CENTRALIZED MODEL FOR LENDING RISK MANAGEMENT SYSTEM

MODELE CENTRALISE POUR SYSTEME DE GESTION DE RISQUE DE CREDIT

English Abstract

This disclosure describes techniques for determining whether to approve or deny a borrower's lending-product request by selectively using a heuristic and statistical model. More specifically, borrower may submit lending-product request to Heuristic-Statistical Risk Management (HS-RM) system, and in doing so HS-RM system may analyze relationship attributes of borrower to determine likelihood of borrower repaying a loan over predetermined time period and avoid being charged off. In some examples, the HS-RM system may execute a plurality of statistical models to determine a charge-off probability score. Each statistical model may be based on a set, or subset of historical lending-product data. A subset of historical lending-product data may be based on a selection bias of shared characteristics within set of historical lending-product data. The selection bias may be based on characteristics of a lending-product request or relationship attributes of borrower.

French Abstract

L'invention concerne des techniques permettant de déterminer s'il faut approuver ou refuser la demande de produit de crédit d'un emprunteur, par la mise en oeuvre sélective d'un modèle heuristique et statistique. Plus particulièrement, l'emprunteur peut soumettre une demande de produit de crédit à un système de gestion de risque heuristique-statistique (HS-RM). Ainsi, le système HS-RM peut analyser les caractéristiques de relation d'un emprunteur afin de déterminer la probabilité de remboursement de l'emprunt par l'emprunteur sur une durée prédéterminée et d'éviter sa radiation. Dans certains exemples, le système HS-RM peut exécuter une pluralité de modèles statistiques pour déterminer un score de probabilité de radiation. Chaque modèle statistique peut être basé sur un ensemble ou sous-ensemble de données historiques de produit de crédit. Un sous-ensemble de données historiques de produit de crédit peut être basé sur un biais de sélection de caractéristiques partagées, dans un ensemble de données historiques de produit de crédit. Le biais de sélection peut être basé sur les caractéristiques d'une demande de produit de crédit ou sur les caractéristiques de relation de l'emprunteur.

Claims

CLAIMS:
1. A statistical risk management system (SRM) for generating a heuristic
analysis of
loan charge-off risk in small, short-term loans for borrowers who have
immediate cash needs,
the SRM comprising:
one or more processors,
a user interface module in communication with the one or more processors, and
memory coupled to the one or more processors, the memory including one or more
modules that are executable by the one or more processors to:
download, from the user interface module, to a client device of a borrower,
for
installation on the client device, an application for generating a lending-
product request;
receive, at the user interface module and from the borrower via the client
device, a
lending-product request that includes at least a loan amount and a borrower
identifier;
retrieve, via the user interface module and from a data-store, client profile
data
associated with the borrower, based at least in part on the borrower
identifier, the client profile
data including one or more relationship attributes that define historical
interactions between the
borrower and at least one financial institution;
generate, via the user interface module, a loan qualifier score based on the
loan amount
and the one or more relationship attributes associated with the borrower;
generate, via a Statistic Model (SM) generation component, at least one
statistical model
to analyze the lending-product request, wherein the at least one statistical
model comprises a
borrower intermediate score based on a plurality of relationship attribute
coefficients;
41

retrieve, via the user interface module in electronic communication with an
independent
data repository maintained by at least one of a financial institution or a
third-party service
provider, historical lending-product data associated with a plurality of
borrowers;
identify, via the SM generation component interacting with the user interface
module,
shared characteristics between the client profile data and the historical
lending-product data;
generate, via the SM generation component, a plurality of statistical models,
individual
statistical models of the plurality of statistical models being based at least
in part on individual
ones of the shared characteristics between the client profile data and the
historical lending-
product data;
determine, via an SM accuracy component interacting with the SM generation
component, an accuracy score for each of the individual statistical models of
the plurality of
statistical models, based at least in part on independent historical lending-
product data which is
different from the historical lending-product data, the accuracy score
reflecting a likelihood
that the borrower defaults or makes an on-time loan payment on the loan
associated with the
lending-product request;
select, as selected individual statistical models, via an SM selection
component
interacting with the SM accuracy component, a portion of the plurality of the
statistical models
that have accuracy scores above a predetermined accuracy score threshold;
perform, simultaneously via an SM analysis component interacting with the SM
selection component, analyses of the lending-product request using the
selected individual
statistical models of the plurality of statistical models and without using
non-selected individual
statistical models of the plurality of statistical models;
42

generate via the SM analysis component an overall charge-off probability score
for the
lending-product request, based at least in part on the analyses of the lending-
product request
using the borrower intermediate score and the selected individual statistical
models and without
using the non-selected individual statistical models;
determine, via the SM analysis component, whether the lending-product request
is
approved, based at least in part on the overall charge-off probability score
being greater than or
equal to a predetermined cutoff threshold, wherein the approval or denial does
not rely on time-
consuming and lengthy credit worthiness checks;
transmit, via a reporting module interacting with the SM analysis component,
an indication of approval or denial of the lending-product request to at least
one of the
client device or a computing device associated with the financial institution;
and
monitor, via the user interface module, the rate of defaults of a plurality of
loans and
adjusting the loan approval cutoff threshold to balance portfolio risk versus
return.
2. The system of claim 1, wherein the one or more modules are further
executable by
the one or more processors to:
retrieve, from one or more third-party services, a set of historical lending-
product data
that includes previously executed lending-products by a plurality of
borrowers;
generate one or more subsets of data from the set of historical lending-
product data,
based at least in part on one or more criteria; and
derive individuals ones of the plurality of relationship attribute
coefficients for the
selected individual statistical models and for the non-selected individual
statistical models,
based at least in part on individual ones of the one or more subsets of data.
43

3. The system of claim 2, wherein the one or more criteria indicates a
selection bias of
historical lending-product data, the selection bias of historical lending-
product data including
one or more of a geographic location, a preceding time-period relative to a
current date, a
lending-product category, or a component of client profile data.
4. The system of claim 2, wherein determining, for each of the statistical
models, the
accuracy score comprises:
retrieving, from a data-store, an additional set of historical lending-product
data that is
associated with previously executed lending-products by a plurality of
borrowers, the additional
set of historical lending-product data being different from the set of
historical lending-product
data used to generate a particular statistical model of the portion of the
plurality of statistical
models;
performing an analysis of the additional set of historical lending-product
data using the
particular statistical model, the analysis to validate an accuracy of a
corresponding set of
relationship attribute coefficients of the particular statistical model;
generating a charge-off probability score for the particular statistical
model, based at
least in part on the analysis;
calculating a delta score that corresponds to a difference between the charge-
off
probability score and actual historical charge-off data within the additional
set of historical
lending-product data; and
generating the accuracy score for the particular statistical model based at
least in part on
the delta score,
44

wherein the one or more modules are further executable by the one or more
processors
to adjust the overall charge-off probability score for the lending-product
request, based at least
in part on the accuracy score for the particular statistical model.
5. The system of claim 4, wherein the one or more modules are further
executable by
the one or more processors to:
determine that the accuracy score of the particular statistical model is less
than a first
predetermined accuracy threshold; and
adjust the overall charge-off probability score associated with the lending-
product
request by excluding the analysis associated with the particular statistical
model.
6. The system of claim 4, wherein the one or more modules are further
executable by
the one or more processors to:
determine that the accuracy score for the particular statistical model is less
than a second
predetermined accuracy threshold;
derive an updated set of relationship attribute coefficients associated with
the particular
statistical model, based at least in part on the additional set of historical
lending-product data;
determine an updated charge-off probability score for the particular
statistical model,
based at least in part on the updated set of relationship attribute
coefficients; and
adjust the overall charge-off probability score associated with the lending-
product
request based at least in part on the updated charge-off probability score.

7. The system of claim 1, wherein the one or more modules are further
executable by
the one or more processors to:
generate individual charge-off probability scores from analyses of the
individual
statistical models of the selected statistical models and without analyses of
the non-selected
individual statistical models, and
wherein the overall charge-off probability score corresponds to a mean value
of the
individual charge-off probability scores.
8. The system of claim 1, wherein the one or more modules are further
executable by
the one or more processors to:
generate individual charge-off probability scores from analyses of the
selected
individual statistical models and without analyses of the non-selected
individual statistical
models, and
wherein the overall charge-off probability score corresponds to a lowest value
of the
individual charge-off probability scores.
9. The system of claim 1, wherein the one or more modules are further
executable by
the one or more processors to:
receive a profit target that is associated with the lending-product request;
and
determine the predetermined cutoff threshold for approval of the lending-
product
request, based at least in part on the profit target.
46

10. One or more non-transitory computer-readable media for generating a
heuristic
analysis of loan charge-off risk in small, short-term loans for borrowers who
have immediate
cash needs, storing computer-executable instructions that, when executed on
one or more
processors, cause the one or more processors to perform acts comprising:
receiving, from a borrower via a client device in communication with the one
or more
processors, a lending-product request that includes at least a loan amount and
a borrower
identifier;
retrieving, from a data-store via a user interface module in electronic
communication
with an independent data repository maintained by at least one of a financial
institution or a
third-party service provider, a set of historical lending-product data that
includes previously
executed lending-products by a plurality of borrowers;
retrieving, from the data-store, client profile data associated with the
borrower, based at
least in part on borrower identifier, the client profile data including one or
more relationship
attributes that define historical interactions between the borrower and at
least one financial
instituti on;
generating, via the user interface module, a loan qualifier score based on the
loan amount
and the one or more relationship attributes associated with the borrower;
generating, via a Statistical Model (SM) generation component, one or more
subsets of
data from the set of historical lending-product data, based at least in part
on one or more criteria
that indicate a selection bias of a lending-product based on a set of
characteristics associated
with the borrower, the selection bias impacting at least the loan amount, the
set of characteristics
including at least a geographic location associated with the borrower or a
component of client
profile data associated with the borrower;
47

identifying, via the SM generation component interacting with the user
interface
module, shared characteristics between the client profile data and the subsets
of data from the
historical lending-product data;
generating, via the SM generation component, individual statistical models to
analyze
the lending-product request, individual statistical models comprise being
based at least in part
on the selection bias and individual ones of the shared characteristics
between the client profile
data and the subsets of data from the historical lending-product data;
determining, via an SM accuracy component interacting with the SM generation
component, for each of the individual statistical models, an accuracy score
that reflects an
accuracy of the individual statistical models, based at least in part on
independent historical
lending-product data which is different from the historical lending-product
data, the accuracy
score reflecting a likelihood that the borrower defaults or makes an on-time
loan payment on
the loan associated with the lending-product request;
selecting, as selected individual statistical models, via an SM selection
component
interacting with the SM accuracy component, a portion of the individual
statistical models that
have accuracy scores above a predetermined accuracy score threshold;
performing, simultaneously via an SM analysis component interacting with the
SM
selection component, analyses of the lending-product request using the
selected individual
statistical models of the plurality of statistical models and without using
non-selected individual
statistical models of the plurality of statistical models;
generating, via the SM analysis component, an overall charge-off probability
score for
the lending-product request, based at least in part on the analyses of the
lending-product request
48

using the borrower intermediate score and the selected individual statistical
models and without
using the non-selected individual statistical models;
determining, via the SM analysis component, that the lending-product request
is
approved, based at least in part on the overall charge-off probability score
being greater than a
predetermined cutoff threshold, wherein the approval or denial does not rely
on time-consuming
and lengthy credit worthiness checks;
transmitting, via a reporting module interacting with the SM analysis
component, an
indication that the lending-product request is approved to at least one of the
client device or a
computing device associated with the financial institution; and
monitoring, via the user interface module, the rate of defaults of a plurality
of loans and
adjusting the loan approval cutoff threshold to balance portfolio risk versus
return.
11. The one or more non-transitory computer-readable media of claim 10,
wherein,
performing, simultaneously, analyses of the lending-product request using the
selected
individual statistical models and without using the non-selected individual
statistical models is
further based at least in part on the client profile data.
12. The one or more non-transitory computer-readable media of claim 10,
further storing
instructions that, when executed cause the one or more processors to perform
acts comprising:
generate a hybrid statistical model based at least in part on individual sets
of relationship
attributes coefficients associated with the portion of the individual
statistical models and
without the remaining portion of the individual statistical models,
49

wherein to generate the overall charge-off probability score is further based
at least in
part on an analysis of the lending-product request via the hybrid statistical
model.
13. The one or more non-transitory computer-readable media of claim 10,
further
storing instructions that, when executed cause the one or more processors to
perform acts
comprising:
generating individual charge-off probability scores from analyses of the
portion of the
individual statistical models and without analyses of the remaining portion of
the individual
statistical models,
wherein the overall charge-off probability score corresponds to one of a mean-
value of
the individual charge-off probability scores, or a lowest-value of the
individual charge-off
probability scores.
14. The one or more non-transitory computer-readable media of claim 10,
further storing
instructions that, when executed cause the one or more processors to perform
acts comprising:
retrieving, from a data-store, an additional set of historical lending-product
data that is
associated with previously executed lending-products by a plurality of
borrowers, the additional
set of historical lending-product data being different from the set of
historical lending-product
data;
determining an accuracy score of a particular statistical model, based at
least in part on
the additional set of historical lending-product data; and
adjusting the overall charge-off probability score for lending-product request
based at
least in part on the accuracy score for the particular statistical model.

15. A computer-implemented method for generating a heuristic analysis of loan
charge-
off risk in small, short-term loans for borrowers who have immediate cash
needs, comprising:
under control of one or more processors:
receiving, from a borrower via a client device in communication with the one
or more
processors, a lending-product request that includes at least a lending-product
category identifier
and a loan amount;
retrieving, from a data-store, client profile data associated with the
borrower, the client
profile data including one or more relationship attributes that define
historical interactions
between the borrower and at least one financial institution;
generating, via the user interface module, a loan qualifier score based on the
loan amount
and the one or more relationship attributes associated with the borrower;
retrieving, via a user interface module in electronic communication with one
or more
third-party services, a set of historical lending-product data that includes
at least the lending-
product category, based at least in part on the lending-product category
identifier;
deriving individual sets of relationship attribute coefficients, based at
least in part on the
set of historical lending-product data;
identifying, via the SM generation component interacting with the user
interface
module, shared characteristics between the client profile data and the set of
historical lending-
product data;
generating, via a Statistical Model (SM) generation component, individual
statistical
models to analyze the lending-product request, based at least in part on a
derivation of individual
51

sets of relationship attribute coefficients, wherein at least one of the
individual statistical models
comprises a borrower intermediate score;
determining, via an SM accuracy component interacting with the SM generation
component, for each of the individual statistical models, an accuracy score
that reflects an
accuracy of the individual statistical model, based at least in part on
independent historical
lending-product data which is different from the set of historical lending-
product data, the
accuracy score reflecting a likelihood that the borrower defaults or makes an
on-time loan
payment on the loan associated with the lending-product request;
selecting, as selected individual statistical models, via an SM selection
component
interacting with the SM accuracy component, a portion of the individual
statistical models that
have accuracy scores above a predetermined accuracy score threshold;
generating a hybrid statistical model by aggregating subsets of historical
lending-
product data that are associated with the selected individual statistical
models and without using
non-selected individual statistical models of the individual statistical
models;
determining, via the SM analysis component, that the lending-product request
is
approved, based at least in part on the analyses of the lending-product
request using the hybrid
statistical model , wherein the approval or denial does not rely on time-
consuming and lengthy
credit worthiness checks;
transmitting, via a reporting module interacting with the SM analysis
component, an
indication that the lending-product request is approved to at least one of the
client device or a
computing device associated with the financial institution; and
monitoring, via the user interface module, the rate of defaults of a plurality
of loans and
adjusting the loan approval cutoff threshold to balance portfolio risk versus
return.
52

16. The computer-implemented method of claim 15, further comprising:
generating individual charge-off probability scores for the lending-product
request from
analyses of the portion of the individual statistical models and without
analyses of the remaining
portion of the individual statistical models; and
determining an overall charge-off probability score for the lending-product
request
based at least in part on one of a mean-value of the individual charge-off
probability scores or
a lowest-value of the individual charge-off probability scores,
wherein, determining that the lending-product request is approved is further
based at
least in part on the overall charge-off probability score being greater than a
predetermined cutoff
threshold.
17. The computer-implemented method of claim 15, further comprising:
generating a hybrid statistical model based at least in part on the individual
sets of
relationship attribute coefficients; and
performing an analysis of the hybrid statistical model based at least in part
on the client
profile data; and
determining an overall charge-off probability score for the lending-product
request from
the analysis of the hybrid statistical model,
wherein determining that the lending-product request is approved is further
based at
least in part on an overall charge-off probability score.
53

18. The computer-implemented method of claim 15, further comprising:
identifying a selection bias of historical lending-product data that is
associated with
individual statistical models, the selection bias of historical lending-
product data corresponding
to at least one of a geographic region, preceding time-period relative to a
current date, a type of
lending-product, or a component of client profile data, and
generating subsets of historical lending-product data, based at least in part
on the
selection bias of historical lending-product data,
wherein, the individual sets of relationship attribute coefficients are
further based at least
in part on the subsets of historical lending-product data.
19. The one or more non-transitory computer-readable media of claim 10,
further
storing instructions that, when executed cause the one or more processors to
perform acts
comprising:
determining correlations between the individual statistical models; and
recommending an additional selection bias for the one or more subset of data
from the
sets of historical lending-product data, based at least in part on a disparity
of correlation between
the individual statistical models.
20. The one or more non-transitory computer-readable media of claim 10,
wherein
the selection bias further impacts a lending-product category, and
wherein the set of characteristics associated with the borrower further
includes a portion
of client profile data associated with the borrower.
54

21. A statistical risk management system (SRM) for generating a heuristic
analysis of
loan charge-off risk, comprising:
one or more processors;
a user interface module in communication with the one or more processors, and
memory coupled to the one or more processors, the memory including one or more
modules that are executable by the one or more processors to:
receive, ftom a borrower via a client device in communication with the one or
more
processors, a lending-product request that includes at least a loan amount and
a borrower
identifier;
retrieve, via the user interface module in electronic communication with an
independent data-store, client profile data associated with the borrower,
based at least in part
on the borrower identifier, the client profile data including one or more
relationship attributes
that define historical interactions between the borrower and at least one
financial institution;
generate, via a Statistical Model (SM) generation software component, multiple
statistical models that are each configured to determine a given charge-off
probability score
based on historical lending-product data;
determine, via an SM accuracy software component interacting with the SM
generation software component, for each of the multiple statistical models, an
accuracy score
that reflects an accuracy of the given charge-off probability score, based at
least in part on
independent historical lending-product data;
based on the accuracy scores for the multiple statistical models, select, via
an SM
selection software component interacting with the SM accuracy software
component, a
portion of the multiple statistical models;

determine, via the SM selection software component interacting with the SM
accuracy
software component, that an accuracy score of a particular statistical model
is less than an
accuracy threshold;
in response to determining that the accuracy score of the particular
statistical model is
less than the accuracy threshold, access, via the SM selection software
component interacting
with the SM accuracy software component, updated historical lending-product
data;
generate, via the SM generation software component, an updated particular
statistical
model using the updated historical lending-product data;
determine, via the SM selection software component interacting with the SM
accuracy
software component, that the accuracy score of the updated particular
statistical model is
greater than the accuracy threshold;
based on the accuracy score for the updated particular statistical model being
greater
than the accuracy threshold, select, via the SM selection software component
interacting with
the SM accuracy software component, the updated particular statistical model;
generate, via the SM generation software component, a hybrid statistical model
using
the portion of the multiple statistical models, the updated particular
statistical model, and
without using a remaining portion of the multiple statistical models;
determine, via the SM analysis software component interacting with the SM
generation software component and the user interface module, a charge-off
probability score
for the lending-product request, based at least in part on the client profile
data and the hybrid
statistical model;
determine, via the SM analysis software component, a cutoff threshold based on
monitoring an existing lending-product portfolio using a machine learning
model;
56

determine, via the SM analysis software component, whether the lending-product
request is approved, based at least in part on the charge-off probability
score being greater
than or equal to the cutoff threshold; and
transmit, via a reporting module interacting with the SM analysis software
component
to the client device, an indication of approval or denial of the lending-
product request.
22. The system of claim 21, wherein the one or more modules are further
executable
by the one or more processors to:
determine that the charge-off probability score is greater than or equal to
the cutoff
threshold.
23. The system of claim 21, wherein the one or more relationship attributes
associated
with the borrower include at least one of a length of membership with the at
least one
financial institution, a payment history with the at least one financial
institution, an indication
of direct deposit history with the at least one financial institution, an
indication of a number of
electronic transactions with the at least one financial institution, or a
number of enrolled
financial products at the at least one financial institution.
24. The system of claim 23, wherein the one or more modules are further
executable
by the one or more processors to:
assign individual attribute scores to the one or more relationship attributes
associated
with the borrower;
57

generate a borrower intermediate score, based at least in part on the
individual
attribute scores; and
select the hybrid statistical model to analyze whether the lending-product
request is
approved, based at least in part on the borrower intermediate score being less
than a
predetermined intermediate threshold.
25. The system of claim 21, wherein the lending-product request further
includes a
lending-product identifier, and wherein the one or more modules are further
executable by the
one or more processors to:
retrieve, from one or more third-party services, the historical lending-
product data that
includes data identifying previously executed lending-products associated with
a plurality of
borrowers and that corresponds to a substantially similar type of lending-
product as a type of
lending-product that is associated with the lending-product request; and
generate, for each of the multiple statistical models, a plurality of
relationship attribute
coefficients based at least in part on the historical lending-product data.
26. The system of claim 25, wherein the one or more modules are further
executable
by the one or more processors to:
retrieve, from a data-store, one or more criteria associated with adjusting
the plurality
of relationship attribute coefficients of each of the multiple statistical
models to correct for a
selection bias of the historical lending-product data, the one or more
criteria being based at
least in part on identifying subsets of historical lending-product data that
is associated with
geographic regions of borrowers or portion of client profile data of the
borrower;
58

generate a subset of the historical lending-product data, based at least in
part on the
one or more criteria; and
adjust at least one of the plurality of relationship attribute coefficients
for each of the
multiple statistical models to account for the selection bias of the
historical lending-product
data, based at least in part on the subset of historical lending-product data.
27. The system of claim 21, wherein the one or more modules are further
executable
by the one or more processors to:
receive, from one or more third-party services, the historical lending-product
data that
is associated with a plurality of borrowers,
wherein determining, for each of the multiple statistical models, the accuracy
score
comprises analyzing the historical lending-product data using each of the
multiple statistical
models to validate an accuracy of a plurality of relationship attribute
coefficients associated
with each of the multiple statistical models, and
wherein selecting the portion of the multiple statistical models is based at
least in part
on the accuracy of the plurality of relationship attribute coefficients
associated with each of
the multiple statistical models being greater than a predetermined the
accuracy threshold.
28. The system of claim 21, wherein the one or more modules are further
executable
by the one or more processors to:
receive, from one or more third party services, a first set of the historical
lending-
product data that is associated with a plurality of borrowers; and
59

retrieve, from a data-store, a second set of the historical lending-product
data that is
associated with a plurality of borrowers,
wherein determining, for each of the multiple statistical models, the accuracy
score
comprises:
generating a plurality of relationship attribute coefficients associated with
each of the
multiple statistical models from the first set of historical lending-product
data; and
analyzing the second set of historical lending-product data using each of the
multiple
statistical models to validate the accuracy of the plurality of relationship
attribute coefficients,
and
wherein selecting the portion of the multiple statistical models is based at
least in part
on the accuracy of the plurality of relationship attribute coefficients
associated with each of
the multiple statistical models being greater than a predetermined accuracy
threshold.
29. The system of claim 28, wherein to generate the relationship attribute
coefficients
associated with the statistical model further includes determining a Receiver
Operator
Characteristic (ROC) curve and values of associated Kolmogorov-Smirnov (K-S)
statistic
along the ROC curve.
30. The system of claim 21, wherein the lending-product request further
includes a
loan amount, and wherein the one or more modules are further executable by the
one or more
processors to:
perform one or more analysis iterations using the hybrid statistical model by
modifying the loan amount associated with the lending-product request such
that the charge-

off probability score for the lending-product request is substantially similar
to the cutoff
threshold; and
determine an upper limit of the loan amount, based at least in part on the
charge-off
probability score being substantially similar to the cutoff threshold.
31. One or more non-transitory computer-readable media storing computer
executable
instructions that, when executed on one or more processors, cause the one or
more processors
to perform acts comprising:
receiving, from a client device associated with a borrower that in
communication with
the one or more processors, a lending-product request that includes at least a
borrower
identifier;
retrieving, via the user interface module in electronic communication with an
independent data-store, a client profile associated with the borrower, the
client profile
including one or more relationship attributes;
assigning, via a Statistical Model (SM) analysis software component,
individual
attribute scores to the one or more relationship attributes;
generating a borrower intermediate score, based at least in part on the
individual
attribute scores;
generating, via an SM generation software component, multiple statistical
models that
are each configured to determine a given charge-off probability for a given
lending-product
request;
determining, via an SM accuracy software component interacting with the SM
generation software component, for each of the multiple statistical models, an
accuracy score
61

that reflects an accuracy of the given charge-off probability score, based at
least in part on
independent historical lending-product data;
based on the accuracy scores for the multiple statistical models, selecting,
via an SM
selection software component interacting with the SM accuracy software
component, a
portion of the multiple statistical models;
determining, via the SM selection software component interacting with the SM
accuracy software component, that an accuracy score of a particular
statistical model is less
than an accuracy threshold;
in response to determining that the accuracy score of the particular
statistical model is
less than the accuracy threshold, accessing, via the SM selection software
component
interacting with the SM accuracy software component, updated historical
lending-product
data;
generating, via the SM generation software component, an updated particular
statistical model using the updated historical lending-product data;
determining, via the SM selection software component interacting with the SM
accuracy software component, that the accuracy score of the updated particular
statistical
model is greater than the accuracy threshold;
based on the accuracy score for the updated particular statistical model being
greater
than the accuracy threshold, select, via the SM selection software component
interacting with
the SM accuracy software component, the updated particular statistical model;
generating, via the SM generation software component, a hybrid statistical
model
using the portion of the multiple statistical models, the updated particular
statistical model,
and without using a remaining portion of the multiple statistical models;
62

determining, via the SM analysis software component, whether to analyze the
lending-
product request via a heuristic model or the hybrid statistical model, based
at least in part on
the borrower intermediate score;
generating, via the SM analysis software component interacting with the SM
generation software component and the user interface module, a charge-off
probability score
for the lending-product request, based at least in part on analysis via the
heuristic model or the
hybrid statistical model;
determining, via the SM analysis software component, a cutoff threshold based
on
monitoring an existing lending-product portfolio using a machine learning
model;
determining, via the SM analysis software component, that the lending-product
request
is approved, based at least in part on the charge-off probability score being
greater than the
cutoff threshold; and
transmitting, via a reporting module interacting with the SM analysis software
component, to the client device, an indication of approval or denial of the
lending-product
request.
32. The one or more non-transitory computer-readable media of claim 31,
further
storing instructions that, when executed cause the one or more processors to
perform acts
comprising:
determining that the borrower intermediate score is less than a predetermined
intermediate threshold;
generating, via the heuristic model, a heuristic charge-off probability score,
based at
least in part on the client profile;
63

determining that the lending-product request is approved based at least in
part on the
heuristic charge-off probability score being greater than a predetermined
heuristic cutoff
threshold, the heuristic charge-off probability score being different from the
borrower
intermediate score; and
determining to analyze the lending-product request via the hybrid statistical
model,
based at least in part on the heuristic charge-off probability score being
less than the
predetermined heuristic cutoff threshold.
33. The one or more non-transitory computer-readable media of claim 31,
further
storing instructions that, when executed cause the one or more processors to
perform acts
comprising:
deriving a plurality of relationship attribute coefficients from historical
lending-
product data associated with a plurality of borrowers, the relationship
attributes corresponding
to one or more of a length of membership with at least one financial
institution, a payment
history with the at least one financial institution, an indication of direct
deposit history with
the at least one financial institution, an indication of a number of
electronic transactions with
the at least one financial institution, or a number of enrolled financial
products at the at least
one financial institution; and
generating the statistical model to determine the charge-off probability score
for the
lending-product request, the hybrid statistical model being based at least in
part on the
plurality of relationship attribute coefficients.
64

34. The one or more non-transitory computer-readable media of claim 31,
further
storing instructions that, when executed cause the one or more processors to
perform acts
comprising:
determining that the borrower intemediate score is greater than a
predetermined
intermediate threshold; and
determining to analyze the lending-product request via the hybrid statistical
model,
based at least in part on the borrower intermediate score being less than the
predetermined
intermediate threshold.
35. The one or more non-transitory computer-readable media of claim 34,
further
storing instructions that, when executed cause the one or more processors to
perform acts
comprising:
applying a distribution function to the borrower intermediate score,
wherein generating, via the hybrid statistical model, the charge-off
probability score is
further based at least in part on an evaluation of the distribution function.
36. A computer-implemented method, comprising:
under control of one or more processors:
receiving, from a client device in communication with the one or more
processors, a
lending-product request that includes at least a borrower identifier
associated with a borrower;
retrieving, via a user interface module in communication with the one or more
processors and from an independent data-store, client profile data associated
with the
borrower, based at least in part on the borrower identifier, the client
profile data including one

or more relationship attributes that define historical interactions between
the borrower and at
least one financial institution;
generating, via a Statistical Model (SM) generation software component,
multiple
statistical models that are each configured to determine a given charge-off
probability score
based on historical lending-product data;
determining, via an SM accuracy software component interacting with the SM
generation software component, for each of the multiple statistical models, an
accuracy score
that reflects an accuracy of the given charge-off probability score, based at
least in part on
independent historical lending-product data;
based on the accuracy scores for the multiple statistical models, selecting,
via an SM
selection software component interacting with the SM accuracy software
component, a
portion of the multiple statistical models;
determining, via the SM selection software component interacting with the SM
accuracy software component, that an accuracy score of a particular
statistical model is less
than an accuracy threshold;
in response to determining that the accuracy score of the particular
statistical model is
less than the accuracy threshold, accessing, via the SM selection software
component
interacting with the SM accuracy software component, updated historical
lending-product
data;
generating, via the SM generation software component, an updated particular
statistical model using the updated historical lending-product data;
66

determining, via the SM selection software component interacting with the SM
accuracy software component, that the accuracy score of the updated particular
statistical
model is greater than the accuracy threshold;
based on the accuracy score for the updated particular statistical model being
greater
than the accuracy threshold, selecting, via the SM selection software
component interacting
with the SM accuracy software component, the updated particular statistical
model;
generating, via the SM generation software component, a hybrid statistical
model
using the portion of the multiple statistical models, the updated particular
statistical model,
and without using a remaining portion of the multiple statistical models;
determining, via the SM analysis software component interacting with the SM
generation software component and the user interface module, a charge-off
probability score
for the lending-product request, based at least in part on the client profile
data and the hybrid
statistical model;
determining, via the SM analysis software component, a cutoff threshold based
on
monitoring an existing lending-product portfolio using a machine learning
model;
determining, via the SM analysis software component, that the lending-product
request
is denied, based at least in part on the charge-off probability score being
less than the cutoff
threshold; and
transmitting, via a reporting module interacting with the SM analysis software
component to the client device, an indication that the lending-product request
is denied.
37. The computer-implemented method of claim 36, wherein lending-product
request
further includes a first loan amount, and further comprising:
67

performing one or more iterations using the hybrid statistical model by
incrementally
modifying a value of the first loan amount such that the charge-off
probability score is
substantially similar to the cutoff threshold;
determining a second loan amount, based at least in part on the charge-off
probability
score being substantially similar to the cutoff threshold; and
transmitting, to the client device, an additional indication that a revised
lending-
product is approved, the revised lending-product including the second loan
amount.
38. The computer-implemented method of claim 36, further comprising:
assigning individual attribute scores to the one or more relationship
attributes
associated with the client profile data; and
generating a borrower intermediate score, based at least in part on the
individual
attribute scores,
wherein determining the charge-off probability score via the hybrid
statistical model is
further based at least in part on the borrower intermediate score being less
than a
predetermined intermediate threshold.
39. The computer-implemented method of claim 36, further comprising:
generating a borrower intermediate score for the borrower, based at least in
part on the
client profile data;
determining that the borrower intermediate score is greater than a
predetermined
intermediate threshold; and
68

generating, via a heuristic model, a heuristic charge-off probability score,
based at
least in part on the client profile data, the heuristic charge-off probability
score being different
from the borrower intermediate score; and
determining that the heuristic charge-off probability score is less than a
predetermined
heuristic cutoff threshold,
wherein determining that the charge-off probability score via the hybrid
statistical
model is further based at least in part on the heuristic charge-off
probability score being less
than the predetermined heuristic cutoff threshold.
40. The computer-implemented method of claim 36, further comprising:
retrieving, from one or more third-party services, historical lending-product
data from
a plurality of borrowers; and
generating a plurality of relationship attribute coefficients, based at least
in part on the
historical lending-product data,
wherein, generating the multiple statistical models is based at least in part
on the
plurality of relationship attributed coefficients.
69

Description

86785221
CENTRALIZED MODEL FOR LENDING RISK MANAGEMENT
SYSTEM
BACKGROUND
[0001] Consumer may occasionally need to borrow small amounts of money for a
short
amount of time to maintain financial sustainability. While most consumers have
access to the
financial services and products offered by financial institutions such as
banks and credit unions,
traditional lending practices of such financial institutions are not well
suited to provide such
small dollar value, short-term loans to consumers. These traditional lending
practices are
generally designed to provide long-term loans of relatively large amounts of
funds for major
goals based on collateral of valuable assets owned by the consumers.
Additionally, these
traditional lending practices may rely on time-consuming and lengthy credit
worthiness checks,
in many cases even when the consumers are existing customers of the financial
institutions,
which are impractical for meeting the immediate cash needs of consumers. As a
result, some
consumers who desire small short-term loans may be forced to turn to third-
party lenders that
do not view the consumers as long-term customers, and who also do have any
incentive to
educate the consumers in the responsible use of credit.
SUMMARY OF THE INVENTION
[0001a] According to one aspect of the present invention, there is
provided a statistical
risk management system (SRM) for generating a heuristic analysis of loan
charge-off risk in
small, short-term loans for borrowers who have immediate cash needs, the SRM
comprising:
1
Date recue / Date received 2021-12-09

86785221
one or more processors, a user interface module in communication with the one
or more
processors, and memory coupled to the one or more processors, the memory
including one or
more modules that are executable by the one or more processors to: download,
from the user
interface module, to a client device of a borrower, for installation on the
client device, an
application for generating a lending-product request; receive, at the user
interface module and
from the borrower via the client device, a lending-product request that
includes at least a loan
amount and a borrower identifier; retrieve, via the user interface module and
from a data-store,
client profile data associated with the borrower, based at least in part on
the borrower identifier,
the client profile data including one or more relationship attributes that
define historical
interactions between the borrower and at least one financial institution;
generate, via the user
interface module, a loan qualifier score based on the loan amount and the one
or more
relationship attributes associated with the borrower; generate, via a
Statistic Model (SM)
generation component, at least one statistical model to analyze the lending-
product request,
wherein the at least one statistical model comprises a borrower intermediate
score based on a
plurality of relationship attribute coefficients; retrieve, via the user
interface module in
electronic communication with an independent data repository maintained by at
least one of a
financial institution or a third-party service provider, historical lending-
product data associated
with a plurality of borrowers; identify, via the SM generation component
interacting with the
user interface module, shared characteristics between the client profile data
and the historical
lending-product data; generate, via the SM generation component, a plurality
of statistical
models, individual statistical models of the plurality of statistical models
being based at least in
part on individual ones of the shared characteristics between the client
profile data and the
historical lending-product data; determine, via an SM accuracy component
interacting with the
la
Date recue / Date received 2021-12-09

86785221
SM generation component, an accuracy score for each of the individual
statistical models of the
plurality of statistical models, based at least in part on independent
historical lending-product
data which is different from the historical lending-product data, the accuracy
score reflecting a
likelihood that the borrower defaults or makes an on-time loan payment on the
loan associated
with the lending-product request; select, as selected individual statistical
models, via an SM
selection component interacting with the SM accuracy component, a portion of
the plurality of
the statistical models that have accuracy scores above a predetermined
accuracy score threshold;
perform, simultaneously via an SM analysis component interacting with the SM
selection
component, analyses of the lending-product request using the selected
individual statistical
models of the plurality of statistical models and without using non-selected
individual statistical
models of the plurality of statistical models; generate via the SM analysis
component an overall
charge-off probability score for the lending-product request, based at least
in part on the
analyses of the lending-product request using the borrower intermediate score
and the selected
individual statistical models and without using the non-selected individual
statistical models;
determine, via the SM analysis component, whether the lending-product request
is approved,
based at least in part on the overall charge-off probability score being
greater than or equal to a
predetermined cutoff threshold, wherein the approval or denial does not rely
on time-consuming
and lengthy credit worthiness checks; transmit, via a reporting module
interacting with the SM
analysis component, an indication of approval or denial of the lending-product
request to at
least one of the client device or a computing device associated with the
financial institution;
and monitor, via the user interface module, the rate of defaults of a
plurality of loans and
adjusting the loan approval cutoff threshold to balance portfolio risk versus
return.
lb
Date recue / Date received 2021-12-09

86785221
[00011)1 According to another aspect of the present invention, there is
provided one or
more non-transitory computer-readable media for generating a heuristic
analysis of loan charge-
off risk in small, short-term loans for borrowers who have immediate cash
needs, storing
computer-executable instructions that, when executed on one or more
processors, cause the one
or more processors to perform acts comprising: receiving, from a borrower via
a client device
in communication with the one or more processors, a lending-product request
that includes at
least a loan amount and a borrower identifier; retrieving, from a data-store
via a user interface
module in electronic communication with an independent data repository
maintained by at least
one of a financial institution or a third-party service provider, a set of
historical lending-product
data that includes previously executed lending-products by a plurality of
borrowers; retrieving,
from the data-store, client profile data associated with the borrower, based
at least in part on
borrower identifier, the client profile data including one or more
relationship attributes that
define historical interactions between the borrower and at least one financial
institution;
generating, via the user interface module, a loan qualifier score based on the
loan amount and
the one or more relationship attributes associated with the borrower;
generating, via a Statistical
Model (SM) generation component, one or more subsets of data from the set of
historical
lending-product data, based at least in part on one or more criteria that
indicate a selection bias
of a lending-product based on a set of characteristics associated with the
borrower, the selection
bias impacting at least the loan amount, the set of characteristics including
at least a geographic
location associated with the borrower or a component of client profile data
associated with the
borrower; identifying, via the SM generation component interacting with the
user interface
module, shared characteristics between the client profile data and the subsets
of data from the
historical lending-product data; generating, via the SM generation component,
individual
lc
Date recue / Date received 2021-12-09

86785221
statistical models to analyze the lending-product request, individual
statistical models comprise
being based at least in part on the selection bias and individual ones of the
shared characteristics
between the client profile data and the subsets of data from the historical
lending-product data;
determining, via an SM accuracy component interacting with the SM generation
component,
for each of the individual statistical models, an accuracy score that reflects
an accuracy of the
individual statistical models, based at least in part on independent
historical lending-product
data which is different from the historical lending-product data, the accuracy
score reflecting a
likelihood that the borrower defaults or makes an on-time loan payment on the
loan associated
with the lending-product request; selecting, as selected individual
statistical models, via an SM
selection component interacting with the SM accuracy component, a portion of
the individual
statistical models that have accuracy scores above a predetermined accuracy
score threshold;
performing, simultaneously via an SM analysis component interacting with the
SM selection
component, analyses of the lending-product request using the selected
individual statistical
models of the plurality of statistical models and without using non-selected
individual statistical
.. models of the plurality of statistical models; generating, via the SM
analysis component, an
overall charge-off probability score for the lending-product request, based at
least in part on the
analyses of the lending-product request using the borrower intermediate score
and the selected
individual statistical models and without using the non-selected individual
statistical models;
determining, via the SM analysis component, that the lending-product request
is approved,
based at least in part on the overall charge-off probability score being
greater than a
predetermined cutoff threshold, wherein the approval or denial does not rely
on time-consuming
and lengthy credit worthiness checks; transmitting, via a reporting module
interacting with the
SM analysis component, an indication that the lending-product request is
approved to at least
id
Date recue / Date received 2021-12-09

86785221
one of the client device or a computing device associated with the financial
institution; and
monitoring, via the user interface module, the rate of defaults of a plurality
of loans and
adjusting the loan approval cutoff threshold to balance portfolio risk versus
return.
[0001c] According to another aspect of the present invention, there is
provided a computer-
implemented method for generating a heuristic analysis of loan charge-off risk
in small, short-
term loans for borrowers who have immediate cash needs, comprising: under
control of one or
more processors: receiving, from a borrower via a client device in
communication with the one
or more processors, a lending-product request that includes at least a lending-
product category
identifier and a loan amount; retrieving, from a data-store, client profile
data associated with
the borrower, the client profile data including one or more relationship
attributes that define
historical interactions between the borrower and at least one financial
institution; generating,
via the user interface module, a loan qualifier score based on the loan amount
and the one or
more relationship attributes associated with the borrower; retrieving, via a
user interface module
in electronic communication with one or more third-party services, a set of
historical lending-
product data that includes at least the lending-product category, based at
least in part on the
lending-product category identifier; deriving individual sets of relationship
attribute
coefficients, based at least in part on the set of historical lending-product
data; identifying, via
the SM generation component interacting with the user interface module, shared
characteristics
between the client profile data and the set of historical lending-product
data; generating, via a
Statistical Model (SM) generation component, individual statistical models to
analyze the
lending-product request, based at least in part on a derivation of individual
sets of relationship
attribute coefficients, wherein at least one of the individual statistical
models comprises a
borrower intermediate score; determining, via an SM accuracy component
interacting with the
le
Date recue / Date received 2021-12-09

86785221
SM generation component, for each of the individual statistical models, an
accuracy score that
reflects an accuracy of the individual statistical model, based at least in
part on independent
historical lending-product data which is different from the set of hi
storicallending-product data,
the accuracy score reflecting a likelihood that the borrower defaults or makes
an on-time loan
payment on the loan associated with the lending-product request; selecting, as
selected
individual statistical models, via an SM selection component interacting with
the SM accuracy
component, a portion of the individual statistical models that have accuracy
scores above a
predetermined accuracy score threshold; generating a hybrid statistical model
by aggregating
subsets of historical lending-product data that are associated with the
selected individual
statistical models and without using non-selected individual statistical
models of the individual
statistical models; determining, via the SM analysis component, that the
lending-product
request is approved, based at least in part on the analyses of the lending-
product request using
the hybrid statistical model , wherein the approval or denial does not rely on
time-consuming
and lengthy credit worthiness checks; transmitting, via a reporting module
interacting with the
SM analysis component, an indication that the lending-product request is
approved to at least
one of the client device or a computing device associated with the financial
institution; and
monitoring, via the user interface module, the rate of defaults of a plurality
of loans and
adjusting the loan approval cutoff threshold to balance portfolio risk versus
return.
[0001d] According to another aspect of the present invention, there is
provided a statistical
risk management system (SRM) for generating a heuristic analysis of loan
charge-off risk,
comprising: one or more processors; a user interface module in communication
with the one or
more processors, and memory coupled to the one or more processors, the memory
including
one or more modules that are executable by the one or more processors to:
receive, from a
if
Date Recue/Date Received 2022-11-21

86785221
borrower via a client device in communication with the one or more processors,
a lending-
product request that includes at least a loan amount and a borrower
identifier; retrieve, via the
user interface module in electronic communication with an independent data-
store, client profile
data associated with the borrower, based at least in part on the borrower
identifier, the client
profile data including one or more relationship attributes that define
historical interactions
between the borrower and at least one financial institution; generate, via a
Statistical Model
(SM) generation software component, multiple statistical models that are each
configured to
detattnine a given charge-off probability score based on historical lending-
product data;
determine, via an SM accuracy software component interacting with the SM
generation
software component, for each of the multiple statistical models, an accuracy
score that reflects
an accuracy of the given charge-off probability score, based at least in part
on independent
historical lending-product data; based on the accuracy scores for the multiple
statistical models,
select, via an SM selection software component interacting with the SM
accuracy software
component, a portion of the multiple statistical models; determine, via the SM
selection
software component interacting with the SM accuracy software component, that
an accuracy
score of a particular statistical model is less than an accuracy threshold; in
response to
determining that the accuracy score of the particular statistical model is
less than the accuracy
threshold, access, via the SM selection software component interacting with
the SM accuracy
software component, updated historical lending-product data; generate, via the
SM generation
software component, an updated particular statistical model using the updated
historical
lending-product data; determine, via the SM selection software component
interacting with the
SM accuracy software component, that the accuracy score of the updated
particular statistical
model is greater than the accuracy threshold; based on the accuracy score for
the updated
lg
Date Recue/Date Received 2022-11-21

86785221
particular statistical model being greater than the accuracy threshold,
select, via the SM
selection software component interacting with the SM accuracy software
component, the
updated particular statistical model; generate, via the SM generation software
component, a
hybrid statistical model using the portion of the multiple statistical models,
the updated
particular statistical model, and without using a remaining portion of the
multiple statistical
models; determine, via the SM analysis software component interacting with the
SM generation
software component and the user interface module, a charge-off probability
score for the
lending-product request, based at least in part on the client profile data and
the hybrid statistical
model; determine, via the SM analysis software component, a cutoff threshold
based on
.. monitoring an existing lending-product portfolio using a machine learning
model; determine,
via the SM analysis software component, whether the lending-product request is
approved,
based at least in part on the charge-off probability score being greater than
or equal to the cutoff
threshold; and transmit, via a reporting module interacting with the SM
analysis software
component to the client device, an indication of approval or denial of the
lending-product
request.
[0001e] According to another aspect of the present invention, there is
provided one or more
non-transitory computer-readable media storing computer executable
instructions that, when
executed on one or more processors, cause the one or more processors to
perform acts
comprising: receiving, from a client device associated with a borrower that in
communication
.. with the one or more processors, a lending-product request that includes at
least a borrower
identifier; retrieving, via the user interface module in electronic
communication with an
independent data-store, a client profile associated with the borrower, the
client profile including
lh
Date Recue/Date Received 2022-11-21

86785221
one or more relationship attributes; assigning, via a Statistical Model (SM)
analysis software
component, individual attribute scores to the one or more relationship
attributes; generating a
borrower intermediate score, based at least in part on the individual
attribute scores; generating,
via an SM generation software component, multiple statistical models that are
each configured
to determine a given charge-off probability for a given lending-product
request; determining,
via an SM accuracy software component interacting with the SM generation
software
component, for each of the multiple statistical models, an accuracy score that
reflects an
accuracy of the given charge-off probability score, based at least in part on
independent
historical lending-product data; based on the accuracy scores for the multiple
statistical models,
selecting, via an SM selection software component interacting with the SM
accuracy software
component, a portion of the multiple statistical models; determining, via the
SM selection
software component interacting with the SM accuracy software component, that
an accuracy
score of a particular statistical model is less than an accuracy threshold; in
response to
determining that the accuracy score of the particular statistical model is
less than the accuracy
threshold, accessing, via the SM selection software component interacting with
the SM
accuracy software component, updated historical lending-product data;
generating, via the SM
generation software component, an updated particular statistical model using
the updated
historical lending-product data; determining, via the SM selection software
component
interacting with the SM accuracy software component, that the accuracy score
of the updated
particular statistical model is greater than the accuracy threshold; based on
the accuracy score
for the updated particular statistical model being greater than the accuracy
threshold, select, via
the SM selection software component interacting with the SM accuracy software
component,
the updated particular statistical model; generating, via the SM generation
software component,
ii
Date Recue/Date Received 2022-11-21

86785221
a hybrid statistical model using the portion of the multiple statistical
models, the updated
particular statistical model, and without using a remaining portion of the
multiple statistical
models; determining, via the SM analysis software component, whether to
analyze the lending-
product request via a heuristic model or the hybrid statistical model, based
at least in part on
the borrower intermediate score; generating, via the SM analysis software
component
interacting with the SM generation software component and the user interface
module, a charge-
off probability score for the lending-product request, based at least in part
on analysis via the
heuristic model or the hybrid statistical model; determining, via the SM
analysis software
component, a cutoff threshold based on monitoring an existing lending-product
portfolio using
a machine learning model; determining, via the SM analysis software component,
that the
lending-product request is approved, based at least in part on the charge-off
probability score
being greater than the cutoff threshold; and transmitting, via a reporting
module interacting with
the SM analysis software component, to the client device, an indication of
approval or denial of
the lending-product request.
[0001f] According to another aspect of the present invention, there is
provided a computer-
implemented method, comprising: under control of one or more processors:
receiving, from a
client device in communication with the one or more processors, a lending-
product request that
includes at least a borrower identifier associated with a borrower;
retrieving, via a user interface
module in communication with the one or more processors and from an
independent data-store,
client profile data associated with the borrower, based at least in part on
the borrower identifier,
the client profile data including one or more relationship attributes that
define historical
interactions between the borrower and at least one financial institution;
generating, via a
lj
Date Recue/Date Received 2022-11-21

86785221
Statistical Model (SM) generation software component, multiple statistical
models that are each
configured to determine a given charge-off probability score based on
historical lending-
product data; determining, via an SM accuracy software component interacting
with the SM
generation software component, for each of the multiple statistical models, an
accuracy score
that reflects an accuracy of the given charge-off probability score, based at
least in part on
independent historical lending-product data; based on the accuracy scores for
the multiple
statistical models, selecting, via an SM selection software component
interacting with the SM
accuracy software component, a portion of the multiple statistical models;
determining, via the
SM selection software component interacting with the SM accuracy software
component, that
.. an accuracy score of a particular statistical model is less than an
accuracy threshold; in response
to determining that the accuracy score of the particular statistical model is
less than the accuracy
threshold, accessing, via the SM selection software component interacting with
the SM
accuracy software component, updated historical lending-product data;
generating, via the SM
generation software component, an updated particular statistical model using
the updated
historical lending-product data; determining, via the SM selection software
component
interacting with the SM accuracy software component, that the accuracy score
of the updated
particular statistical model is greater than the accuracy threshold; based on
the accuracy score
for the updated particular statistical model being greater than the accuracy
threshold, selecting,
via the SM selection software component interacting with the SM accuracy
software
component, the updated particular statistical model; generating, via the SM
generation software
component, a hybrid statistical model using the portion of the multiple
statistical models, the
updated particular statistical model, and without using a remaining portion of
the multiple
statistical models; determining, via the SM analysis software component
interacting with the
lk
Date Recue/Date Received 2022-11-21

86785221
SM generation software component and the user interface module, a charge-off
probability
score for the lending-product request, based at least in part on the client
profile data and the
hybrid statistical model; determining, via the SM analysis software component,
a cutoff
threshold based on monitoring an existing lending-product portfolio using a
machine learning
model; determining, via the SM analysis software component, that the lending-
product request
is denied, based at least in part on the charge-off probability score being
less than the cutoff
threshold; and transmitting, via a reporting module interacting with the SM
analysis software
component to the client device, an indication that the lending-product request
is denied.
BRIEF DESCRIPTION OF THE DRAWINGS
[0002] The detailed description is set forth with reference to the
accompanying figures. In the
figures, the left-most digit(s) of a reference number identifies the figure in
which the reference
number first appears. The use of the same reference numbers in different
figures indicates
similar or identical items or features.
11
Date Recue/Date Received 2022-11-21

CA 03087631 2020-07-03
WO 2019/135860
PCT/US2018/064698
[0003] FIG. 1 illustrates a schematic view of a computing environment that
facilitates an
analysis of a lending-product request via a Statistical Risk Management (SRM)
system.
[0004] FIG. 2 illustrates a block diagram of a SRM system that is configured
to select a
heuristic model analysis or a statistical model analysis to determine whether
to approve a
lending-product request.
[0005] FIG. 3 illustrates a block diagram of a SRM system that is configured
to
concurrently execute a plurality of statistical models to determine whether to
approve a
lending-product request.
[0006] FIG. 4 illustrates a block diagram of a SRM system that is configured
to generate
a hybrid statistical model based on a subset of statistical models
100071 FIG. 5 illustrates a block diagram showing various components of a SRM
system
that is configured to analyze a lending-product request.
[0008] FIG. 6 illustrates a SRM system process to select one a heuristic model
analysis or
a statistical model analysis to determine whether to approve a lending-product
request.
[0009] FIG. 7 illustrates a SRM system process to generate a statistical model
based on a
set of historical lending-product data.
[0010] FIG. 8 illustrates a SRM system process to execute a plurality of
statistical models
to determine an overall charge-off probability score for approval of a lending-
product
request.
[0011] FIG. 9 illustrates a SRM system process to generate a hybrid
statistical model
based on an aggregated set of historical lending-product data.
2

CA 03087631 2020-07-03
WO 2019/135860
PCT/US2018/064698
DETAILED DESCRIPTION
100121 This disclosure describes techniques for selectively using a heuristic
model or a
statistical model to analyze relationship attributes of a borrower to
determine whether to
approve or deny a lending-product request. More specifically, a Statistical
Risk
Management (SRM) system is described that determines a probability of a
borrower
repaying a loan over a predetermined time, and avoiding being charged off. A
charge-off is
a declaration by a creditor that an amount of debt is unlikely to be
collected. In one example,
a charge-off may occur when a borrower becomes delinquent on a repayment of
the loan
amount for predetermined time-period, such as 30-days. However, any
predetermined time
period is possible
[0013] The SRM system may approve or deny various types of lending-product
requests
based on various criteria such as a borrower's loan qualification, the loan
amount solicited
by the borrower, and the intended purpose of the loan amount. In one example,
approval of
a lending-product request may be based on a combination of a borrower's loan
qualification
and the loan amount solicited by the borrower. While a borrower may solicit
any loan
amount, the SRM system may be configured such that different loan amounts
invite different
gradations of heuristic or statistical analyses. For example, a financial
institution may
consider a loan amount of $500 to be low-risk, and thus attract a heuristic
analysis of the
borrower's loan qualification. Alternatively, a loan amount of $5,000 may be
considered a
high-risk in terms of being charged-off, and thus attract a different
gradation of a heuristic
analysis or a statistical analysis of the borrower's loan qualification.
[0014] In another example, approval of a lending-product request may be based
on a
combination of a borrower's loan qualification and the intended purpose of the
loan. A
lending-product request for the purchase of an automobile may invite a
different gradation
of heuristic analysis or statistical analysis for a loan intended for a
recreational vehicle.
3

CA 03087631 2020-07-03
WO 2019/135860
PCT/US2018/064698
[0015] Additionally, the SRM system determine the approval of a lending-
product request
by comparing a charge-off probability score associated with the lending-
product request and
an approval cutoff threshold. The approval cutoff threshold can be set up an
operator of the
SRM system as a method of managing a loan poi folio risk.
[0016] In one example, the SRM system may receive a lending-product request
from a
borrower. A borrower may initiate a lending-product request from an online
portal that is
operated by a financial institution using a web browser installed on a client
device. The
client device may access the online portal via a local area network (LAN), a
larger network
such as a wide area network (WAN), or a collection of networks, such as the
Internet. The
online portal may provide a loan request interface page that enables the
borrower to initiate
a loan request. The loan request interface page may be configured to permit
the borrower to
initiate the lending-product request after the borrower has submitted
authentication
credentials that authenticates an identity of the borrower as an existing
customer of the
financial institution. Alternatively, the borrower may initiate a lending-
product request via
a SRM application that is native to the client device, and communicatively
coupled to the
SRM system.
[0017] In various examples, the borrower may submit a lending-product request
that
includes a borrower identifier, a lending-product identifier, a loan amount,
or any
combination thereof. The borrower identifier may be used to retrieve client
profile data
associated with the borrower from a repository that is maintained by, or on
behalf of, the
SRM system. The client profile data may further include relationship
attributes that describe
observable characteristics of the borrower's relationship with a financial
institution.
Alternatively, or additionally, the SRM system may retrieve relationship
attributes from a
relationship attribute repository that is maintained by, or on behalf of, the
SRM system.
4

CA 03087631 2020-07-03
WO 2019/135860
PC171.182018/064698
[0018] In a non-limiting example, relationship attributes may include a length
of
relationship of the borrower with the financial institution, a payment history
that includes
the number of times the borrower paid open and closed loan payments on time, a
direct
deposit history that includes the number of direct deposits for which the
borrower is a
primary account holder, electronic transaction history that includes the
number of electronic
transactions for which the borrower is a primary account holder, and an
aggregated deposit
balance during a predetermined transaction period.
[0019] The lending-product identifier may identify categories of lending-
products based on
an intended purpose for the lending-product. For example, financial
institutions may
selectively offer different terms of repayment for a lending-product (i.e.
repayment period,
interest rate, deposit amount, penalties for default payments, etc.) based on
the intended
purpose of the lending-product. For example, a lending-product that is
intended for use in
purchasing an automobile may have more favorable terms (i.e. interest rate,
deposit amount)
relative to a lending-product that is intended to purchase a recreational
vehicle. Moreover,
categories of lending-products may include, but are not limited to,
automobiles, recreational
vehicles, marine vehicles, financing repayment of outstanding health expenses,
household
expenses, or any other type of personal or business expense.
[0020] In response to receiving a lending-product request, the SRM system may
determine
whether to approve or deny the lending-product request based on a heuristic
model analysis
or a statistical model analysis. In one example, a heuristic model analysis
includes assigning
a select number of relationship attributes with a numerical score based on a
relative value
of the borrower's relationship attribute. For example, a borrower's 11-year
length of
relationship may equate to a numerical score of 5 on a scale of 1 to 5; the
numerical score
of 5 being the upper-limit of benefit that may be assigned to a length of
relationship,
5

CA 03087631 2020-07-03
WO 2019/135860
PCT/US2018/064698
attribute. It is noteworthy that a numerical scale of 110 5 is used for
exemplaiy purposes
only. Any numerical scale is possible.
[0021] In doing so, the heuristic model analysis involves generating a loan
qualifier score
for the borrower based on a mathematical combination of the numerical scores
for the select
number of relationship attributes, along with other criteria such as a loan
amount, or an
intended purpose of the lending-product. In this example, the SRM system may
approve the
lending-product request based at least in part on the loan qualifier score
being greater than
a predetermined heuristic threshold. The predetermined heuristic threshold may
be set by
an administrator of the SRM system and intended to approve lending-product
requests for
low-risk borrowers, low-risk lending-products, or a combination of both. For
example, the
predetermined heuristic threshold may be influenced by an intended purpose of
the lending-
product request, the loan amount, or a combination of both.
[0022] In some examples, the SRM system may detelinine that the loan qualify
score for
the borrower is less than the predetermined heuristic threshold. In doing so,
the SRM system
may generate one or more statistical models to determine whether to approve
the lending-
product request. In one example, the SRM system may retrieve historical
lending-product
data for a plurality of borrowers. The historical lending-product data may be
maintained by
the SRM system within a historical lending-product data repository.
Alternatively, or
additionally, the historical lending-product data may be maintained by a
financial institution
or a third-party service provider on behalf of the SRM system or the financial
institution.
[0023] The historical lending-product data may include client profile data
associated with a
plurality of borrowers that submitted lending-product requests over a
predetermined time
period. The historical lending-product data may further include historical
records of
corresponding lending-products, historical records lending-product request
denials,
6

CA 03087631 2020-07-03
WO 2019/135860
PCT/US2018/064698
historical records of actual charge-offs (i.e. defaults) associated with
approved lending-
product requests, or any combination thereof.
100241 The SRM system may selectively generate a statistical model based on
the historical
lending-product data. More specifically, the SRM system may select a random
sample set
of lending-product data from within the historical lending-product data. In
some examples,
the SRM system may apply a correction for selection bias within the historical
lending-
product data, based on an incomplete randomness in the sample set of
borrowers.
100251 Additionally, the SRM system may selectively generate a plurality of
statistical
models by first parsing through the historical lending-product data to
generate a plurality of
subsets of the historical lending-product data. In doing so, the SRM may
generate an
individual statistical model for each subset of historical lending-product
data. In one
example, an administrator of the SRM system may identify a selection bias for
subsets of
historical lending-product data based on shared characteristics of a lending-
product request
or relationship attributes of a borrower. For example, shared characteristics
of a lending-
product request may include, but is not limited to, a loan amount, a lending-
product category
(i.e. intended purpose of lending-product request), or a combination of both.
Similarly,
shared characteristics of relationship attributes of a borrower may include,
but is not limited
to, a geographic location, portions of client profile data (i.e. employment
status, income
bracket, etc.). Alternatively, or additionally, the SRM system may detect and
recommend a
selection bias for subsets of historical lending-product data based on a
disparity of
correlations between a plurality of existing, statistical models,
10026.1 Each statistical model may provide a set of relationship attribute
coefficients that are
useful for determining whether borrowers are able to repay their loans.
Relationship
attribute coefficients may include an aggregate deposit (Dep) coefficient, one
or more length
of relation (LoR) coefficients, one or more payment history (PayH)
coefficients, a direct
7

CA 03087631 2020-07-03
WO 2019/135860
PCT/US2018/064698
deposit (DirDep) coefficient, an electronic transactions (ElecTr) coefficient,
a banking
product (Prod) coefficient, a bill pay coefficient (BillPay), an affiliate
coefficient (All), or
any combination thereof.
[0027] In response to generating one or more statistical models and the
ensuing relationship
attribute coefficients, the SRM system may generate individual borrower
intermediate
scores based on analyses of a borrower's relationship attributes relative to
relationship
attribute coefficients of each statistical model. The borrower intermediate
score may reflect
a likelihood of the borrower repaying the loan amount associated with the
lending-product
request.
[0028] Additionally, the SRM system may calculate a charge-off probability
score of the
borrower not defaulting (i.e. charging-off) on payment associated with a
lending-product
for a predetermined time, based on each intermediate borrower score. In one
example, the
SRM system may apply a distribution function, such as a Standard Normal
Cumulative
Distribution Function (CDF), to an intermediate borrower score to calculate
the charge-off
probability score.
[0029] Moreover, the SRM system may selectively calculate an overall charge-
off
probability score for a lending-product request based on the plurality of
charge-off
probability scores. In one example, the SRM system may determine an overall
charge-off
probability score by determining a mean-value or median-value of the plurality
of charge-
off probability scores. Alternatively, the SRM system may select a lowest-
value charge-off
probability score as the overall charge-off probability score.
[0030] In any case, the SRM system may compare the overall charge-off
probability score
with an approval cutoff threshold to determine whether to approve or deny the
lending-
product request. In one example, SRM system may determine that the overall
charge-off
probability score is greater than or equal to the approval cutoff threshold.
In this example,
8

CA 03087631 2020-07-03
WO 2019/135860
PCT/US2018/064698
the SRM system may approve the lending-product request and transmit a message
to a client
device of the borrower, indicating that the lending-product request has been
approved.
Alternatively, the SRM system may determine that the overall charge-off
probability is less
than the approval cutoff threshold, and further transmit a message to the
client device
indicating the denial of the lending-product request.
[0031] The approval cutoff threshold may be used as a method of balancing a
lending-
product portfolio risk versus return. Raising a value of the approval cutoff
threshold may
lead to a lower rate of default (i.e. charge-offs) of approved loans, but may
also reduce the
volume of approved lending-product requests. Alternatively, lowering the value
of the
approval cutoff threshold may increase a number of defaults (i.e. charge-offs)
of approved
loans, but increase the volume of approved lending-product requests. In
various examples,
the SRM system may monitor the volume and rate of defaults of an existing
lending-product
portfolio to determine whether a change to an approval cutoff threshold may
support a
revenue-based and/or profit-based target. The SRM system may use one or more
trained
machine learning models to monitor an existing lending-product portfolio and
adjust the
approval cutoff threshold. Alternatively, an administrator of the SRM system
may adjust
the approval cutoff threshold, via a user interface module.
[0032] In the illustrated example, the SRM system may validate the accuracy of
a statistical
model using an independent set of historical lending-product data that is
different from
historical lending-product data used to generate the statistical model. For
example, the SRM
system may retrieve, from one or more third-party services, the independent
set of historical
lending-product data that is associated with a plurality of borrowers The
independent set of
historical lending-product data may include a same type of data that is used
to generate the
statistical model, such as client profile data associated with a plurality of
borrowers that
submitted lending-product requests over a predetermined time period,
historical records of
9

CA 03087631 2020-07-03
WO 2019/135860
PCT/US2018/064698
corresponding lending-products, historical records lending-product request
denials,
historical records of actual charge-offs (i.e. defaults) associated with
approved lending-
product requests, or any combination thereof.
[0033] Thus, the SRM system may parse through the independent set of
historical lending-
product data and select a set of relationship attributes associated with an
independent
historical lending-product request previously submitted by a borrower. In this
example, the
SRM system may use the statistical model being validated to calculate a
borrower
intermediate score for the independent historical lending-product request,
based on the
borrower's relationship attributes. The SRM system may further calculate a
charge-off
probability score of the borrower not defaulting (i.e. charging-off) on
payment associated
with the independent historical lending-product request, and in doing so,
compare the
charge-off probability score with an actual record of whether the borrower did
in fact default
on payment of the historical lending-product.
[0034] In this way, the SRM system may generate an accuracy score for the
statistical model
using an independent set ¨ or subset - of independent historical lending-
product data, and
further determining whether the statistical model correctly predicts a
borrower's default or
on-time payment. The accuracy score may be an alpha-numeric expression (i.e. 0
to 10, or
A to F), a descriptive expression (i.e. low, medium, or high), based on color
(i.e. red, yellow,
or green), or any other suitable scale that reflects a degree of correlation
charge-off
probability score generated by the statistical model and the actual record of
a borrower's
default or on-time loan payment.
10035] Moreover, the SRM system may selectively determine whether the accuracy
score
of the statistical model is equal to or greater than a predetermined accuracy
threshold. The
predetermined accuracy threshold may be a mean-value accuracy score (i.e. 5,
C, medium,
or yellow), or any other accuracy score that is set by an administrator of the
SRM system.

CA 03087631 2020-07-03
WO 2019/135860
PCT/US2018/064698
The predetermined accuracy threshold may reflect an accurate correlation of
some, but not
all, of independent historical lending-product requests associated with the
set of independent
lending-product data.
[0036] In response to determining that the accuracy score is greater than the
predetermined
accuracy threshold, the SRM system may continue to use the statistical model,
as validated,
to determine whether to approve or deny future lending-product request. In
some examples,
the SRM system may determine that the accuracy score of the statistical model
is less than
the predetermined accuracy threshold. In these instances, the SRM system may
re-generate
the statistical model based on a revised set of historical lending-product
data that includes
the independent set of historical lending-product data, or another updated set
of historical
lending-product data.
[0037] In the illustrated example, the SRM system may further generate a
hybrid statistical
model by combining subsets of historical lending-product data associated with
one or more
statistical models. For example, the SRM system may concurrently execute a
plurality of
statistical models. Each individual statistical model may be based on a
different subset of
historical lending-product data that is biased towards different, borrower
characteristics.
[0038] The SRM system may further generate an accuracy score for each
statistical model
to determine whether each statistical model correctly predicts a borrower's
default or on-
time payment. In this way, the SRM system may select a portion of the total
number of
statistical models, based at least in part on their relative accuracy scores.
In one example,
the SRM system may select an n-number [any integer number], but not all, of
the total
number of statistical models with the highest relative accuracy score. In
another example,
the SRM system may select an n-number of statistical models with accuracy
scores that are
greater than or equal to a predetermined accuracy threshold.
11

CA 03087631 2020-07-03
WO 2019/135860
PCT/US2018/064698
[0039] In doing so, the SRM system may create an aggregate subset of
historical lending-
product data by selectively combining the subsets of historical lending-
product data of the
selected statistical models. In this way, the SRM system may further generate
a hybrid
statistical model using the aggregate subset of historical lending-product
data.
[0040] Further, the term "techniques," as used herein, may refer to system(s),
method(s),
computer-readable instruction(s), module(s), algorithms, hardware logic,
and/or
operation(s) as permitted by the context described above and through the
document.
[0041] FIG. 1 illustrates a schematic view of a computing environment 100 that
facilitates
an analysis of a lending-product request via a Statistical Risk Management
(SRM) system
102. The SRM system 102 may receive a lending-product request 104 from a
client device
106 associated with a borrower 108. In some examples, the borrower 108 may
submit the
lending-product request 104 via a SRM application 110 native on the client
device 106.
Alternatively, the borrower 108 may submit the lending-product request 104 via
an online
portal that is operated by a financial institution, using a web browser
installed on the client
device 106. The lending-product request 104 may include a borrower identifier,
a lending-
product identifier, a loan amount, or any combination thereof.
[0042] In one example, the SRM system 102 may retrieve client profile data
associated with
the borrower 108 from a client profile data-store maintained by the SRM system
102. In
doing so, the SRM system 102 may generate a loan qualifier score for the
borrower 108
based at least in part on the lending-product request 104, and further
determine whether to
approval the lending-product request 104 based via a heuristic model analysis.
[0043] In some examples, the SRM system 102 may determine that a statistical
model
analysis is required to determine whether the lending-product request may be
approved,
based on the loan qualifier score being less than a predetermined heuristic
threshold. In these
instances, the SRM system 102 may retrieve historical lending-product data 112
from a
12

CA 03087631 2020-07-03
WO 2019/135860
PCT/US2018/064698
historical lending-product repository 114, and in doing so, generate one or
more statistical
models to analyze the lending-product request 104. The historical lending-
product
repository 114 may be a repository that is maintained by the SRM system 102, a
computing
device 116 of a financial institution 118, a third-party service provider on
behalf of the
financial institution 118, or any combination thereof.
[0044] In some examples, the SRM system 102 may analyze the lending-product
request 104 via one statistical model, a combination of statistical models, or
a hybrid
statistical model that aggregates subsets of historical lending-product data
from multiple
statistical models. In this way, the SRM system 102 may determine an approval
of the
lending-product request, based on a charge-off probability score relative to
an approval
cutoff threshold. The SRM system 102 may then generate and transmit a loan
decision 120
to the client device 106 informing the borrower 108 of an approval or denial
of the lending-
product request 104.
[0045] In the illustrated example, the client device 106 may correspond to any
sort of
electronic device operating on the telecommunications network, such as a
cellular phone, a
smart phone, a tablet computer, an electronic reader, a media player, a gaming
device, a
personal computer (PC, a laptop computer), etc. The client device 106 may have
a subscriber
identity module (SIM), such as an eSIM, to identify the respective electronic
device to a
telecommunications service provider network (also referred to herein as
.. "telecommunications network").
[0046] Additionally, the SRM system 102 may operate on one or more distributed
computing resource(s). The one or more distributed computing resource(s) may
include one
or more computing device(s) 122(1)-122(N) that operate in a cluster or other
configuration
to share resources, balance load, increase performance, provide fail-over
support or
.. redundancy, or for other purposes. The one or more computing device(s)
122(1)-122(N)
13

CA 03087631 2020-07-03
WO 2019/135860
PCT/US2018/064698
may include one or more interfaces to enable communications with other
networked
devices, such as the client device 106, the financial institution 118, via one
or more
network(s) 124.
[0047] The one or more network(s) 124 may include public networks such as the
Internet,
private networks such as an institutional and/or personal intranet, or some
combination of
private and public networks. The one or more network(s) 124 can also include
any type of
wired and/or wireless network, including but not limited to local area network
(LANs), wide
area networks (WANs), satellite networks, cable networks, Wi-Fi networks, Wi-
Max
networks, mobile communications networks (e.g. 36, 4G, and so forth), or any
combination
thereof.
100481 FIG. 2 illustrates a block diagram of a SRM system 202 that is
configured to select
a heuristic model analysis or a statistical model analysis to determine
whether to approve a
lending-product request 204. More specifically, a borrower 206 may generate
and transmit
a lending-product request 204 to the SRM system 202 via a client device 208,
The lending-
product request 204 may include at least one of a borrower identifier, a
lending-product
identifier, or a loan amount. In doing so, the SRM system 202 may retrieve
relationship
attributes associated with the borrower 206, based at least in part on the
lending-product
request 204. The SRM system 202 may retrieve the relationship attributes from
a client
profile data repository or a relationship attribute repository native to the
SRM system 202.
100491 In some examples, the SRM system 202 may perform a heuristic model
analysis 210
to determine whether to approve the lending-product request 204. In some
examples, the
SRM system 202 may selectively bypass the heuristic model analysis 210 based
at least in
part on one or more criteria, and automatically perform a statistical model
analysis 212 of
the lending-product request 204. The one or more criteria may relate to the
intended purpose
.. for the lending-product, the loan amount, or a relationship attribute of
the borrower. In one
14

CA 03087631 2020-07-03
WO 2019/135860
PCT/US2018/064698
example, one criteria may stipulate that a lending-product request 204
intended for certain
depreciating assets, such as recreational vehicles, are analyzed using a
statistical model
analysis 212. In another example, one criteria may also indicate that loan
amounts above a
predetermined value are to be analyzed using a statistical model analysis 212.
In yet another
example, one criteria may state that lending-product requests associated with
borrowers that
have a length of relationship with a financial institution (i.e. relationship
attribute) that is
less than a predetermined number of years, are to be analyzed using a
statistical model
analysis 212.
[0050] In the illustrated example, the SRM system 202 may selectively perform
the
heuristic model analysis and generate a loan qualifier score for the borrower
206. The loan
qualifier score may correspond to a mathematical combination of numerical
scores for a
select number of relationship attributes, along with one or more other
criteria such as a loan
amount and an intended purpose of the lending-product. The SRM system 202 may
assign
a numerical score to each relationship attribute and/or criteria relative to a
predetermined
scale. In one example, the SRM system 202 may assign a numerical score of 2 on
a scale of
1 to 5 for a loan amount that is greater than a predetermined amount. The
numerical score
of 2 may indicate that the loan amount is considered a high-risk value.
Similarly, the SRM
system 202 may assign a numerical score of 4 on scale of 1 to 5 to a lending-
product that is
intended for purchase of a commuter vehicle, indicating that the intended
purpose of the
lending-product is considered low-risk.
[0051] In response to generating a loan qualifier score, the SRM system 202
may compare
the loan qualifier score with a predetermined heuristic threshold. In one
example, the SRM
system 202 may generate a loan decision 214 for the lending-product request
204 based at
least in part on the loan qualifier score being greater than or equal to the
predetermined
heuristic threshold. The loan decision 214 may correspond to a loan approval
or a loan

CA 03087631 2020-07-03
WO 2019/135860
PCT/US2018/064698
denial. In another example, the SRM system 202 may determine that the loan
qualifier score
is less than the predetermined heuristic threshold. In doing so, the SRM
system may analyze
the lending-product request 204 via a statistical model analysis 212, and
further generate a
corresponding, loan decision 216 to approve or deny the lending-product
request 204.
[0052] FIG. 3 illustrates a block diagram of a SRM system 302 that is
configured to
concurrently execute a plurality of statistical models to determine whether to
approve a
lending-product request. In some examples, the SRM system 302 may parse
through a set
of historical lending-product data to generate one or more subset(s) of
historical lending-
product data 304(1)-304(N). Each subset of historical lending-product data
304(1)-304(N)
may share similar characteristics that include geographic location, portions
of client profile
data (i.e. employment status, income bracket, etc.), lending-product category
(i.e. intended
purpose of a lending-product), loan amount, or any combination thereof.
[0053] Moreover, the SRM system 302 may generate a plurality of statistical
model(s) 306(1)-306(N) based on the one or more subset(s) of historical
lending-product
data 304(1)-304(N). In other words, each subset of historical lending-product
data 304(1)-
304(N) may be used to generate corresponding, statistical model(s) 306(1)-
306(N).
[0054] In response to generating the plurality of statistical model(s) 306(1)-
306(N), the
SRM system 302 may generate corresponding, charge-off probability score(s)
308(1)-
308(N) for individual statistical models of the plurality of statistical
model(s) 306(1)-
306(N). The charge-off probability score(s) 308(1)-308(N) may then be used to
generate an
overall charge-off probability score 310 upon which to base a loan decision
312 (i.e.
approval or denial) for a lending-product request. The overall charge-off
probability score
310 may correspond to a mean-value or median value of the charge-off
probability score(s)
308(1)-308(N). Alternatively, the SRM system 302 may determine the overall
charge-off
16

CA 03087631 2020-07-03
WO 2019/135860
PCT/US2018/064698
probability score 310 by selecting a lowest-value of the charge-off
probability score(s)
308(1)-308(N).
[0055] In the illustrated example, the SRM system 302 may generate a loan
decision 312
by comparing the overall charge-off probability score 310 with an approval
cutoff threshold.
The SRM system 302 may approve a lending-product request based on the overall
charge-
off probability score 310 being greater than or equal to an approval cutoff
threshold.
Alternatively, the SRM system 302 may deny a lending-product request based on
the overall
charge-off probability score 310 being less than the approval cutoff
threshold.
[0056] FIG. 4 illustrates a block diagram of a SRM system 402 that is
configured to generate
a hybrid statistical model 404 based on a plurality of statistical models
406(1)-406(N). In
various examples, the SRM system may generate a hybrid statistical model 404
by
aggregating a selection of multiple subsets of historical lending-product
data.
[0057] In the illustrated example, the SRM system 402 may selectively generate
the
plurality of statistical models 406(1)-406(N), based on individual subsets of
historical
lending-product data 408(1)-408(N). Each subset of historical lending-product
data 408(1)-
408(N) may be based on a selection bias of shared borrower or lending-product
characteristics within a set of historical lending-product data. For example,
a selection bias
of shared characteristics may include, but are not limited to, a geographic
location, portions
of client profile data (i.e. employment status, income bracket, etc.), lending-
product
category (i.e. an intended purpose of a lending-product), loan amount, or any
combination
thereof. In a non-limiting example, a subset of historical lending-product
data may be biased
towards on a particular lending-product category (i.e. automobile,
recreational vehicle,
marine vehicle, etc.) within the set of historical lending-product data. In
another non-
limiting example, a subset of historical lending-product data may be biased
towards a
particular geographic location within the set of historical lending-product
data.
17

CA 03087631 2020-07-03
WO 2019/135860
PCT/US2018/064698
[0058] In response to generating the plurality of statistical model(s) 406(1)-
406(N), each
based on an individual subset of historical lending-product data 408(1)-
408(N), the SRM
system 402 may execute each statistical model and determine corresponding
accuracy
score(s) 410(1)-410(N). An accuracy score is intended to quantify the degree
to which a
statistical model correctly predicts a borrower's default or on-time loan
payment (i.e.
charge-off probability score). The SRM system 402 may generate each accuracy
score using
an independent of historical lending-product data that is different from each
subset of
historical lending-product data 408(1)-408(N) used to generate the statistical
models.
[0059] In this way, the SRM system 402 may parse through the independent set
of historical
lending-product data and identify at least one independent historical lending-
product request
previously submitted by a borrower. The SRM system 402 may then execute the
plurality
of statistical model(s) 406(1)-406(N) to calculate borrower intermediate
score(s), and
ultimately charge-off probability score(s) that corresponds to the independent
historical
lending-product request. The charge-off probability score(s) may be compared
with an
actual record ¨ from the independent set of historical lending-product data ¨
of whether the
borrower did in fact default on payment of the historical lending-product.
[0060] Moreover, the SRM system 402 may generate accuracy score(s) 410(1)-
410(N) for
each of the statistical model(s) 406(1)-406(N) based on whether the
statistical model(s)
406(1)-406(N) correctly predict an actual charge-off or on-time payment of the
historical
lending-product. In some examples, the SRM system 402 may monitor the accuracy
score(s)
410(1)-410(N) (i.e. by re-generating the accuracy scores) of the statistical
model(s) 406(1)-
406(N) to ensure that each accuracy score is greater than or equal to a
predetermined
accuracy threshold. The predetermined accuracy threshold may be set by an
administrator
of the SRM system 402. If the accuracy score of a statistical model is below
the
predetelinined accuracy threshold, the SRM system 402 may selectively
discontinue use of
18

CA 03087631 2020-07-03
WO 2019/135860
PCT/US2018/064698
the statistical model. Alternatively, the SRM system 402 may replace the
statistical model
with another statistical model that is based on an alternate set of historical
lending-product
data. In one example, the alternate set of historical lending-product data may
include the
independent set of historical lending-product data used to generate the
accuracy score.
[0061] The SRM system may monitor the accuracy score(s) 410(1)-410(N) of the
statistical
model(s) 406(1)-406(N) on a continuous basis, per a predetermined schedule, or
in response
to a triggering event. The predetermined schedule may correspond to any time
interval, such
as one day, one week, or one month. A triggering event may be initiated by an
administrator
of the SRM system 402, or an independent accounting of a predetermined number
of
inaccurate charge-off predictions made by the statistical model(s) 406(1)-
406(N).
10062] In the illustrated example, the SRM system 402 may generate the hybrid
statistical
model 404 by selecting a number of statistical models, based on their relative
accuracy
scores. In one example, the SRM system 402 may select a predetermined number
of
statistical models with the highest relative accuracy scores among the
statistical models
406(1)-406(N). Alternatively, the SRM system 402 may select statistical models
with an
accuracy score above a predetermined hybrid accuracy threshold. The
predetermined hybrid
accuracy threshold may be set by an administrator of the SRM system 402, and
be
substantially similar to the predetermined accuracy threshold that is used to
determine
whether to selectively discontinue use of a statistical model. Alternatively,
the
____________________________________________________________ predetern
lined hybrid accuracy threshold maybe different from the predetermined
accuracy
threshold.
[0063] In response to selecting a number of statistical models based on their
relative
accuracy scores, the SRM system 402 may create an aggregate subset of
historical lending-
product data by combining the subsets of historical lending-product data
associated each of
19

CA 03087631 2020-07-03
WO 2019/135860
PCT/US2018/064698
the selected statistical models. In this way, the SRM system 402 may generate
a hybrid
statistical model 404 based on the aggregate subset of historical lending-
product data.
100641 FIG. 5 illustrates a block diagram showing various components of a SRM
system
that is configured to analyze a lending-product request. In the illustrated
example the SRM
system 502 may include routines, program instructions, objects, and/or data
structures that
perform particular tasks or implement abstract data types. Further, the SRM
system 502 may
include input/output interface(s) 504. The input/output interface(s) 504 may
include any
type of output interface known in the art, such as a display (e.g. a liquid
crystal display),
speakers, a vibrating mechanism, or a tactile feedback mechanism. Input/output
interface(s)
504 also include ports for one or more peripheral devices, such as headphones,
peripheral
speakers, or a peripheral display. Further, the input/output interface(s) 504
may further
include a camera, a microphone, a keyboard/keypad, or a touch-sensitive
display. A
keyboard/keypad may be a push button numerical dialing pad (such as on a
typical
telecommunication device), a multi-key keyboard (such as a conventional QWERTY
.. keyboard), or one or more other types of keys or buttons, and may also
include a joystick-
like controller and/or designated navigation buttons, or the like.
100651 Additionally, the SRM system 502 may include network interface(s) 506.
The
network interface(s) 506 may include any sort of transceiver known in the art.
For example,
the network interface(s) 506 may include a radio transceiver that performs the
function of
transmitting and receiving radio frequency communications via an antenna. In
addition, the
network interface(s) 506 may also include a wireless communication transceiver
and a near
field antenna for communicating over unlicensed wireless Internet Protocol
(IP) networks,
such as local wireless data networks and personal area networks (e.g.
Bluetooth or near field
communication (NFC) networks). Further, the network interface(s) 506 may
include wired
communication components, such as an Ethernet port or a Universal Serial Bus
(USB).

CA 03087631 2020-07-03
WO 2019/135860
PCT/US2018/064698
[0066] Further, the SRM system 502 may include one or more processor(s) 508
that are
operably connected to memory 510. In at least one example, the one or more
processor(s)
508 may be a central processing unit(s) (CPU), graphics processing unit(s)
(GPU), a both a
CPU and GPU, or any other sort of processing unit(s). Each of the one or more
processor(s)
508 may have numerous arithmetic logic units (ALUs) that perform arithmetic
and logical
operations as well as one or more control units (CUs) that extract
instructions and stored
content from processor cache memory, and then executes these instructions by
calling on
the ALUs, as necessary during program execution. The one or more processor(s)
508 may
also be responsible for executing all computer applications stored in the
memory, which can
be associated with common types of volatile (RAM) and/or nonvolatile (ROM)
memory.
100671 In some examples, memory 510 may include system memory, which may be
volatile
(such as RAM), non-volatile (such as ROM, flash memory, etc.) or some
combination of
the two. The memory may also include additional data storage devices
(removable ad/or
non-removable) such as, for example, magnetic disks, optical disks, or tape.
[0068] The memory 510 may further include non-transitory computer-readable
media, such
as volatile and nonvolatile, removable and non-removable media implemented in
any
method or technology for storage of information, such as computer readable
instructions,
data structures, program modules, or other data. System memory, removable
storage and
non-removable storage are all examples of non-transitory computer-readable
media.
Examples of non-transitory computer-readable media include, but are not
limited to, RAM,
ROM, EEPROM, flash memory or other memory technology, CD-ROM, digital
versatile
disks (DVD) or other optical storage, magnetic cassettes, magnetic tape,
magnetic disk
storage or other magnetic storage devices, or any other non-transitory medium
which can
be used to store the desired information.
21

CA 03087631 2020-07-03
WO 2019/135860
PCT/US2018/064698
[0069] In the illustrated example, the memory 510 may include an operating
system 512, a
user-interface module 514, an input module 516, a Statistical (HS) process
selection module
518, a heuristic-model (HM) analysis module 520, a statistical model (SM)
analysis
module 522, a reporting module 524, a recommendation module 526, and one or
more data
repositories 528. The operating system 512 may be any operating system capable
of
managing computer hardware and software resources.
[0070] The input module 516 may be configured to receive a borrower's lending-
product
request. In some examples, the input module 516 may receive a lending-product
request
from an online portal operated by a financial institution. A borrower may
access the online
portal via a web browser installed on the client device. Alternatively, or
additionally, the
input module 516 may receive a lending-product request directly from a SRM
application
native on a borrower's client device. Further, the input module 516 may
receive sets of
historical lending-product data from a third-party service provider, or
financial institution,
for the purpose of generating statistical models, or validating an accuracy of
existing
statistical models.
[0071] The HS process selection module 518 may be configured to determine
whether to
conduct an initial analysis of a lending-product request via a heuristic model
analysis or a
statistical model analysis. In one example, the HS process selection module
518 may include
one or more criteria that selectively bypass a heuristic model analysis, and
instead,
automatically perform a statistical model analysis. The one or more criteria
may be
configured by an administrator of the SRM system, via the user-interface
module 514, and
may relate to the intended purpose for the lending-product, the loan amount,
or a relationship
attribute of the borrower, or any combination thereof
[0072] The HM analysis module 520 may be configured to perform a heuristic
model
analysis of a lending-product request. The HM analysis module 520 may further
include a
22

CA 03087631 2020-07-03
WO 2019/135860
PCT/US2018/064698
HM rules component 530. In one example, the HM analysis module 520 may
generate a
loan qualifier score for a borrower associated with a lending-product request.
More
specifically, the HM analysis module 520 may parse through the lending-product
request to
retrieve one or more characteristics of a lending-product request (i.e. loan
amount, lending-
product category) and one or more data repositories to retrieve a select
number of
relationship attributes of the borrower associated with the lending-product
request. In doing
so, the HM analysis module 520 may generate a loan qualifier score for the
lending-product
request based on one or more rules from the HM rules component 530.
100731 The HM rules component 530 may include one or more rules that assign
numerical
scores to a select number of relationship attributes associated with a
borrower, a select
number of characteristics of a lending-product request, or a combination of
both. For
example, the HM analysis module 520 may assign one numerical score to a length
of
relationship of the borrower with a financial institution, a second numerical
score for a loan
amount associated with a lending-product request, and a third numerical score
for an
intended purpose of the lending-product request. The numerical score may be
assigned on
a numerical scale of 1 to 5, 1 to 10, or any other numerical scale. In the
illustrated example,
an increase in a numerical score favorably tends toward a loan approval
decision.
100741 Further, the HM rules component 530 may include one or more rules for
generating
a loan qualifier score by mathematically combining the numerical scores of the
select
number of relationship attributes, select number of characteristics of a
lending-product
request, or a combination of both.
10075 The HM analysis module 520 may generate a loan qualifier score based on
the one
or more rules form the HM rules component 530. In doing so, the 1-IM analysis
module 520
may compare the loan qualifier score with a predetermined heuristic threshold
to determine
whether to approval the lending-product request. In some examples, the
predetermined
23

CA 03087631 2020-07-03
WO 2019/135860
PCT/US2018/064698
heuristic threshold may be stored in the HM rules component 530 and may be set
by an
administrator of the SRM system 502 via the user-interface module 514.
100761 In one example, the HM analysis module 520 may approve the lending-
product
request based on the loan qualifier score being greater than or equal to the
predetermined
heuristic threshold. In this example, the HM analysis module 520 may transmit
a message
to the reporting module 524 indicating that the lending-product request is
approved. In
another example, the HM analysis module 520 may transmit an indication the HS
process
selection module 518 to analyze the lending-product request via a statistical
model analysis,
based at least in part on the loan qualifier score being less than the
predetermined heuristic
threshold.
100771 In the illustrated example, the SM analysis module 522 may be
configured to
perform a statistical model analysis of a lending-product request. The SM
analysis module
522 further includes an SM selection component 532, an SM rules component 534,
an SM
generation component 536, an SM analysis component 538, and a SM accuracy
component
540.
100781 The SM selection component 532 may be configured to determine whether
to
analyze the lending-product request using one statistical model or a plurality
of statistical
models. Further, the SM selection component 532 may also determine which
statistical
models to use for the statistical model analysis. In one example, the SM
selection component
532 may parse through the lending-product request to retrieve one or more
characteristics
of the lending-product request (i.e. loan amount, lending-product category,
andior so forth).
In doing so, the SM selection component 532 may identify one or more
statistical models
that were generated with a selection bias towards the loan amount or the
lending-product
category identified within the lending-product request. In a non-limiting
example, the SM
selection component 532 may parse through a lending-product request to
identify a lending-
24

CA 03087631 2020-07-03
WO 2019/135860
PCT/US2018/064698
product request for $1000 that is intended for a purchase of a marine vehicle
(i.e. lending-
product category). Thus, the SM selection component 532 may identify two
relevant
statistical models from a repository; a first statistical model that was
generated with a
selection bias of loan amounts that approximate $1000, and a second
statistical model that
was generated with a selection bias of a lending-product category for marine
vehicles.
[0079] Similarly, the SM selection component 532 may retrieve, from one or
more data
repositories, a select number of relationship attributes of a borrower
associated with the
lending-product request. In this way, the SM selection component 532 may
identify one or
more statistical models that were generated with a selection bias towards the
select number
of relationship attributes of the borrower.
100801 Additionally, the SM selection component 532 may filter selection of a
statistical
model based on an accuracy score indication from the SM accuracy component
540. For
example, the SM accuracy component 540 component may transmit a message to the
SM
selection component 532 to continue use or discontinue use of a statistical
model for
analyses of lending-product requests.
[0081] SM rules component 534 may include one or more rules associated with
analysis of
a lending-product request via one or more statistical models. In one example,
the SM rules
component 534 may include a register of statistical models that were generated
with a
selection bias towards characteristics of a lending-product request or
relationship attributes.
Thus, the SM rules component 534 may include one or more rules that associate
a lending-
product request with one or more statistical models, based on a comparison of
characteristics
of the lending-product request or relationship attributes of the borrower, and
the selection
bias of the one or more statistical models.
100821 Further, the SM rules component 534 may further include a register of
statistical
models that were generated without a selection bias of underlying date. In
this example, the

CA 03087631 2020-07-03
WO 2019/135860
PCT/US2018/064698
one or more rules may associate a lending-product request with one or more
statistical
models, based on an accuracy score of the statistical model, or some other
criteria specified
by an administrator of the SRM system 502. In some examples, the one or more
rules may
further indicate a predetermined accuracy threshold of statistical models. For
example, the
one or more rules may indicate that a statistical model may be used to analyze
a lending-
product request in response to an accuracy score for the statistical model
being greater than
or equal to the predetermined accuracy threshold. Additionally, the one or
more rules may
specify a calculation method for an overall charge-off probability score from
a plurality of
charge-off probability scores.
[0083] Moreover, the SM rules component 534 may include one or more thresholds
used
by the SM analysis module 522, such as but not limited to, a predetermined
accuracy
threshold, a predetermined hybrid accuracy threshold, and an approval cutoff
threshold.
Each of these one or more thresholds may be set by an administrator of the SRM
system
502 via the user-interface module 514.
[0084] The SM generation component 536 may generate a statistical model, based
on a set
of historical lending-product data. In some examples, the SM generation
component 536
may retrieve the set of historical lending-product data from a repository that
is maintained
by the SRM system 502, a financial institution, or a third-party service
provider on behalf
of the SRM system or the financial institution.
[0085] In doing so, the SM generation component 536 may selectively generate a
statistical
model based on the set of historical lending-product data. More specifically,
the SRM
system may select a random sample set of lending-product data from within the
historical
lending-product data. In some examples, the SRM system may apply a correction
for
selection bias within the historical lending-product data, based on an
incomplete
randomness in the sample set of borrowers.
26

CA 03087631 2020-07-03
WO 2019/135860
PCT/US2018/064698
[0086] In some examples, the SM generation component 536 may generate a
plurality of
statistical models by first parsing through the set of historical lending-
product data to
generate a plurality of subsets of the historical lending-product data. In
some examples, each
subset of historical lending-product data may be based on a selection bias of
characteristics
of a lending-product request or relationship attributes of a borrower. The
selection bias of
characteristics for a lending-product request may include a loan amount, a
lending-product
category. Similarly, the selection bias of characteristics for a borrower may
include a
geographic location, portions of client profile data (i.e. employment status,
income bracket,
etc.).
[0087] Further, the SM generation component 536 may selectively generate a
hybrid
statistical model by aggregating one or more subsets of historical lending-
product data. The
SM generation component 536 may rely on one or more rules from the SM rules
component
534 to determine which subsets of historical lending-product data are to be
aggregated. In
one example, the SM generation component 536 may aggregate a number of subsets
of
historical lending-product data, based on relative accuracy scores of
corresponding
statistical models.
[0088] SM analysis component 538 may analyze a lending-product request using
one or
more statistical models. The SM analysis component 538 may generate a borrower
intermediate score using each statistical model by analyzing a borrower's
relationship
attributes relative to relationship attribute coefficients associated with
each statistical model.
The borrower's intermediate score may reflect a likelihood of the borrower
repaying the
loan amount of the lending-product request.
[0089] The SM analysis component 538 may further calculate a charge-off
probability score
of the borrower not defaulting (i.e. charging-off) on payment associated with
a lending-
product for a predetermined time, based on each intermediate borrower score.
In one
27

CA 03087631 2020-07-03
WO 2019/135860
PCT/US2018/064698
example, the SM analysis component 538 may apply a distribution function, such
as a
Standard Normal Cumulative Distribution Function (CDF), to an intermediate
borrower
score to calculate the charge-off probability score.
[0090] In some examples, the SM analysis component 538 may selectively
calculate an
overall charge-off probability score in response to executing a plurality of
statistical models.
A determination of the overall charge-off probability score may be based on
one or more
rules within the SM rules component 534. The overall charge-off probability
score may be
based on a mean-value or a median-value of charge-off probability scores from
each
statistical model. Alternatively, the overall charge-off probability score may
be a lowest-
value charge-off probability score from among the plurality of statistical
models.
100911 Moreover, the SM analysis component 538 may compare a charge-off
probability
score (i.e. for an analysis using a single statistical model) or an overall
charge-off probability
score (i.e. for an analysis using multiple statistical models) with an
approval cutoff
threshold. In one example, the SM analysis component 538 may determine that
the charge-
.. off probability score or overall charge-off probability score is greater
than or equal to the
approval cutoff threshold. In this example, the SM analysis component 538 may
transmit an
indication to the reporting module 524 that the corresponding lending-product
request is
approved. Alternatively, the SM analysis component 538 may determine that the
charge-off
probability score or overall charge-off probability score is less than the
approval cutoff
threshold, in this latter example, the SM analysis component 538 may transmit
an indication
to the reporting module 524 that the corresponding lending-product request is
denied.
10092] In some examples, the SM analysis component 538 may determine whether
an
approval cutoff threshold that is used to determine whether to approve or deny
a lending-
product request supports a revenue-based and/or profit based target. In some
examples, an
administrator of the SRM system 502 may input a revenue-based or profit-based
target via
28

CA 03087631 2020-07-03
WO 2019/135860
PCT/US2018/064698
the user-interface module 514, In this example, the SM analysis component 538
may
monitor the volume and rate of defaults of an existing lending-product
portfolio to determine
whether a change to an approval cutoff threshold may support the revenue-based
and/or
profit-based target. The SM analysis component 538 may use one or more trained
machine
learning models to monitor an existing lending-product portfolio and
automatically adjust
the approval cutoff threshold. Alternatively, the SM analysis component 538
may transmit
a message to the recommendation module 526 indicating a recommended adjustment
to the
approval cutoff threshold. In doing so, an administrator of the SRM system 502
may
selectively adjust the approval cutoff threshold, via the user-interface
module 514.
100931 The SM accuracy component 540 may be configured to determine an
accuracy score
for a statistical model. An accuracy score is intended to quantify the degree
to which a
statistical model correctly predicts a borrower's default or on-time loan
payment (i.e.
charge-off probability score). In one example, the SM accuracy component 540
may
determine an accuracy score for a statistical model using an independent set ¨
or subset - of
independent historical lending-product data. For example, the SM accuracy
component 540
may parse through an independent set of historical lending-product data and
identify at least
one independent historical lending-product request previously submitted by a
borrower. The
SM accuracy component 540 may retrieve a borrower's relationship attributes
that are
associated with the independent historical lending-product request from the
independent set
of historical lending-product data (i.e. client profile data), or a repository
that is maintained
by, or on behalf of, the SRM system 502.
10094] Further, the SM accuracy component 540 may execute the statistical
model using
the borrower's relationship attributes associated with the independent
historical lending-
product request. In doing so, the SM accuracy component 540 may generate a
borrower
29

CA 03087631 2020-07-03
WO 2019/135860
PCT/US2018/064698
intermediate score via the statistical model; and further, by applying a
distribution function
(i.e. a Standard Normal CDF), calculate a corresponding charge-off probability
score.
100951 The SM accuracy component 540 may further generate an accuracy score
for the
statistical model by comparing the charge-off probability score with an actual
record ¨ from
the independent set of historical lending-product data ¨ of whether the
borrower did in fact
default on payment of the independent historical lending product.
100961 In some examples, the SM accuracy component 540 may further compare the
accuracy score with a predetermined accuracy threshold. In one example, the SM
accuracy
component 540 may determine that the accuracy score is greater than or equal
to a
predetermined accuracy threshold. In this example, the SM accuracy component
540 may
transmit a message to the SM selection component 532 indicating that the
statistical model
may be used to analyze a lending-product request. Alternatively, the SM
accuracy
component 540 may determine that the accuracy core is less than the
predetermined
accuracy threshold. In this latter example, the SM accuracy component 540 may
transmit a
message to the SM selection component 532 to discontinue use of the
statistical model.
Further, the SM accuracy component 540 may transmit an additional message to
the
recommendation module 526 indicating a discontinued use of the statistical
model.
100971 The SM accuracy component 540 may monitor accuracy scores of each
statistical
model on a continuous basis, per a predetermined schedule, or in response to a
triggering
event. The predetermined schedule may correspond to any time interval, such as
one day,
one week, or one month. A triggering event may be initiated by an
administrator of the SRM
system 502, or an independent accounting of a predetermined number of charge-
off
predictions made by a statistical model.
100981 The user-interface module 514 may enable an administrator to interact
with the SRM
system 502 via data input devices and data output devices. The data input
devices may

CA 03087631 2020-07-03
WO 2019/135860
PCT/US2018/064698
include, but are not limited to, combinations of one or more keypads,
keyboards, mouse
devices, touch screens that accept gestures, microphones, voice or speech
recognition
devices, and any other suitable devices or other electronic/ software
selection methods. The
data output devices may include visual displays, speakers, virtual reality
(VR) gear, haptic
feedback devices, and/or so forth. In some examples, the administrator may use
the user-
interface module 514 to cause the SM analysis module 522 to adjust the
approval cutoff
threshold. The administrator may monitor portfolio metrics, such as charge
offs, to achieve
a desired balance between portfolio risk and return.
[0099] In other examples, the administrator may use the user-interface module
514 to
configure a loan qualify score formula or a borrower score formula for use the
HM analysis
module 520 or the SM analysis module 522, respectively.
[00100] The reporting module 524 may be configured to transmit a message to a
client
device of a borrower indicating whether a lending-product has been approved or
denied. In
some examples, the reporting module 524 may receive a message from the HM
analysis
module 520 or the SM analysis component 538 of the SM analysis module 522
indicating
whether a lending-product is denied or approved. Further, the reporting module
524 may be
configured to transmit a message to a financial institution indicating that a
lending-product
request associated with a borrower has been approved or denied.
[00101] The recommendation module 526 may be configured to transmit one or
more
recommendation(s) to an administrator of the SRM system 502. For example, the
recommendation module 526 may receive a message from the SM accuracy component
540
indicating a discontinued use of a statistical model, based at least in part
on an accuracy
score being less than a predetermined threshold. In this example, the
recommendation
module 526 may transmit a recommendation to an administrator of the SRM system
502
31

CA 03087631 2020-07-03
WO 2019/135860
PCT/US2018/064698
recommending an update to historical lending-product data associated with the
statistical
model.
1001021 Further, the recommendation module 526 may receive a message from the
SM
analysis component 538 indicating a recommended adjustment to the approval
cutoff
threshold. In doing so, the recommendation module 526 may transmit a
recommendation to
an administrator of the SRM system 502 that includes the recommended
adjustment to the
approval cutoff threshold.
1001031 Alternatively, or additionally, the SRM system may detect and
recommend a
selection bias for subsets of historical lending-product data based on a
disparity of
correlations between a plurality of existing, statistical models.
1001041 The data repositories 528 may include, but is not limited to, a
relationship attribute
repository, a statistical model repository, and a historical lending-product
data repository.
The data repositories 528 may be maintained by the SRM system 502.
Alternatively, the
data repositories 528 may be maintained by a financial institution or a third-
part service
provider on behalf of the financial institution or the SRM system 502. The
relationship
attribute repository may include client profile data associated with a
plurality of borrowers.
The client profile data may include relationship attributes, such as a length
of relationship
of the borrower with the financial institution, a payment history that
includes the number of
times the borrower paid open and closed loan payments on time, a direct
deposit history that
includes the number of direct deposits for which the borrower is a primary
account holder,
electronic transaction history that includes the number of electronic
transactions for which
the borrower is a primary account holder, an aggregated deposit balance during
a
predetermined transaction period.
1001051 The statistical model repository may include one or more statistical
models used to
analyses lending-product requests via the SRM system 502. In various examples,
the
32

CA 03087631 2020-07-03
WO 2019/135860
PCT/US2018/064698
statistical model repository may include statistical models that are currently
in use, and
statistical models that were historically in-use but currently discontinued.
[00106] The historical lending-product data repository may include sets of
historical
lending-product data that was used to generate one or more statistical models.
In some
examples, the historical lending-product data may also include subsets of
historical lending-
product data that are derived from a set of historical lending-product data.
The historical
lending-product data repository may include sets and subsets of historical
lending-product
data that is current in use, and historically in-use but currently
discontinued.
1001071 FIGS. 6, 7, 8, and 9 present processes 600, 700, 800, and 900 that
relate to
operations of the SRM system. Each of processes 600, 700, 800, and 900
illustrate a
collection of blocks in a logical flow chart, which represents a sequence of
operations that
can be implemented in hardware, software, or a combination thereof. In the
context of
software, the blocks represent computer-executable instructions that, when
executed by one
or more processors, perfolin the recited operations. Generally, computer-
executable
.. instructions may include routines, programs, objects, components, data
structures, and the
like that perform particular functions or implement particular abstract data
types. The order
in which the operations are described is not intended to be construed as a
limitation, and any
number of the escribed blocks can be combined in any order and/or in parallel
to implement
the process. For discussion purposes, the processes 600, 700, 800, and 900 are
described
with reference to the computing environment 100 of FIG. 1.
[00108] FIG. 6 illustrates a SRM system process to select one a heuristic
model analysis or
a statistical model analysis to determine whether to approve a lending-product
request. The
heuristic model analysis model may be configured to approve a lending-product
request for
a low-risk borrower, a low-risk lending-product, or a combination of both.
33

CA 03087631 2020-07-03
WO 2019/135860
PCT/US2018/064698
[00109] Al 602, the SRM system may receive, via a client device, a lending-
product request
that includes at least a borrower identifier. In some examples, the lending-
product request
may further include a lending-product identifier that is associated with an
intended purpose
of the lending-product request (i.e. lending-product category). For example,
the lending-
product request may be associated with the purchase of an automobile,
recreational vehicle,
marine vehicle, or to finance payment of outstanding health, household, or any
other type
of consumer product or service.
[00110] At 604, the SRM system may retrieve client profile data associated
with the
borrower, based at least in part on the borrower identifier. The client
profile data may
include one or more relationship attributes that describe observable
characteristics of the
borrower's relationship with a financial institution. Alternatively, the SRM
system may
retrieve the one or more relationship attributes from a relationship attribute
repository native
to the SRM system, based at least in part on the borrower identifier.
[00111] At 606, the SRM system may generate a loan qualifier score for the
borrower, based
at least in part on a select number of relationship attributes associated with
the borrower.
For example, the SRM system may assign numerical scores to each of the select
number of
relationship attributes. In doing so, the loan qualifier score may correspond
to a
mathematical combination of each of the numerical scores. In some examples,
the loan
qualifier score may be further based on criteria such as a loan amount and
intended purpose
of the lending-product.
[00112] At 608, the SRM system may determine that the loan qualifier score for
the
borrower is less than the predetermined heuristic threshold. In this instance,
the SRM system
may select at least one statistical model to analyze the lending-product
request. More
specifically, the SRM system may generate at least one statistical model that
further
generates a plurality of relationship attribute coefficients. In doing so, the
SRM system may
34

CA 03087631 2020-07-03
WO 2019/135860
PCT/US2018/064698
calculate a borrower intermediate score for the lending-product request by
applying the
borrower's relationship attiibutes to the plurality of relationship attribute
coefficients
generated by the statistical model.
[00113] Alternatively, the SRM system may determine that the loan qualifier
score is greater
than or equal to a predetermined heuristic threshold. In this way, the SRM
system may
approve the lending-product request. In some examples, an administrator of the
SRM system
may configure a predetermined heuristic threshold to approve lending-product
requests for
low risk borrowers, low-risk products, or a combination of both.
1001141 At 610, the SRM system generate a charge-off probability score for the
lending-
product request based at least in part on the borrower intermediate score. The
SRM system
may calculate the charge-off probability score by applying a Normal CDF to the
intermediate borrower score.
[00115] At 612, the SRM system may determine whether to approve or deny the
lending-
product request, based at least in part on a comparison of the charge-off
probability score
relative to an approval cutoff threshold. In one example, the SRM system may
determine
that the charge-off probability score for the lending-product request is
greater than or equal
to the approval cutoff threshold. In doing so, the SRM system may process an
approval of
the lending-product request and further transmit an indication to the client
device of the
borrower indicating that the lending-product request has been approved.
[00116] Alternatively, the SRM system may determine that the charge-off
probability score
for the lending-product request is less than the approval cutoff threshold. In
this instance,
the SRM system may process a denial of the lending-product request and further
transmit
an indication to the client device of the borrower indicating that the lending-
product request
has been denied. In some examples, the approval cutoff threshold may be used
as a method
of balancing a lending-product portfolio risk versus return.

CA 03087631 2020-07-03
WO 2019/135860
PCT/US2018/064698
[00117] FIG. 7 illustrates a SRM system process to generate a statistical
model based on a
set of historical lending-product data. In some examples, the SRM system may
generate a
subset of historical lending-product data from the set of historical lending-
product data,
based on a selection bias of shared characteristics within the set of
historical lending-product
data. In a non-limiting example, a lending-product request that is associated
with a borrower
that resides within a particular geographic region may be analyzed using a
statistical model
that is generated by a subset of historical lending-product data that is
biased towards that
same, particular geographic region.
1001181 At 702, the SRM system may retrieve, from a data repository, a set of
historical
lending-product data. In some examples, the data repository may be maintained
by a
financial institution, the SRM system, or a third-party service provider on
behalf of one of
the financial institution or the SRM system. Further, the set of historical
lending-product
data may include client profile data associated with the borrower,
relationship attributes
associated with a plurality of borrowers, historical records of corresponding
lending-
products, denials of lending-product requests, actual charge-offs associated
with approved
lending-products, or any combination thereof
[00119] At 704, the SRM system may identify a selection bias for borrowers
that share
certain characteristics within the set of historical lending-product data. In
some examples,
the selection bias may be directed towards shared characteristics, such as a
geographic
location, portions of client profile data (i.e. employment status, income
bracket, etc.),
lending-product category (i.e. an intended purpose of a lending-product), loan
amount, or
any combination thereof
[00120] At 706, the SRM system may parse through the set of historical lending-
product
data and generate one or more subsets of historical lending-product data based
at least in
part on the selection bias.
36

CA 03087631 2020-07-03
WO 2019/135860
PCT/US2018/064698
[00121] At 708, the SRM system may generate individual statistical models,
based at least
in part on the individual subsets of the historical lending-product data. In
other words, the
SRM system may generate individual statistical models that focus on shared
borrower
characteristics, such as a geographic location, portions of client profile
data (i.e.
employment status, income bracket, etc.), lending-product category, loan
amount, or any
combination thereof.
[00122] FIG. 8 illustrates a SRM system process to execute a plurality of
statistical models
to determine an overall charge-off probability score for approval of a lending-
product
request. In various examples, the SRM system may generate a plurality of
statistical models
that are based on a selection bias of subsets of historical lending-product
data. Each subset
of historical lending-product data may share one or more similar
characteristics such as a
geographic location, portions of client profile data (i.e. employment status,
income bracket,
etc.), lending-product category (i.e. intended purpose of a lending-product),
loan amount,
or any combination thereof.
[00123] At 802, the SRM system may receive, via a client device, a lending-
product request
that includes at least a borrower identifier. The lending-product request may
further include
a lending-product identifier, a loan amount, of a combination of both.
[00124] At 804, the SRM system may generate or retrieve a client profile
associated with
the borrower, based at least in part on the borrower identifier. The client
profile may include
one or more relationship attributes that describe observable characteristics
of the borrower's
relationship with a financial institution.
[00125] At 806, the SRM system may concurrently execute a plurality of
statistical models
using the one or more relationship attributes associated with the borrower. In
doing so, the
SRM system may generate borrower intermediate scores from the results of each
of the
individual statistical models; and further, by applying a distribution
function (i.e. a Standard
37

CA 03087631 2020-07-03
WO 2019/135860
PCT/US2018/064698
Normal CDF), calculate corresponding charge-off probability scores for each of
the
borrower intermediate scores.
1001261 At 808, the SRM system may determine an overall charge-off probability
score for
the lending-product request based at least in part on the charge-off
probability scores
associated with the plurality of statistical models. In one example, the
overall charge-off
probability score may be based on a mean-value or a median-value of the charge-
off
probability scores. Alternatively, the overall charge-off probability score
may be based on
the lowest-value of the charge-off probability scores.
1001271 At 810, the SRM system may generate a loan decision for the lending-
product
request, based at least in part on a comparison of the overall probability
score relative to an
approval cutoff threshold. In this example, the loan decision may correspond
to an approval
of the lending-product request based at least in part on the overall charge-
off probability
score being greater than or equal to the approval cutoff threshold.
Alternatively, the loan
decision may correspond to a denial of the lending-product request, based
least in part on
the overall probability score being less than the approval cutoff threshold.
1001281 FIG. 9 illustrates a SRM system process to generate a hybrid
statistical model based
on an aggregated set of historical lending-product data. In one example, the
SRM system
may selectively aggregate subsets of historical lending-product data that are
associated with
statistical models that maintain a threshold accuracy score. An accuracy score
of statistical
model is intended to quantify the degree to which a statistical model
correctly predicts a
borrower' default or on-time loan payment (i.e. charge-off probability score).
1001291 At 902, the SRM system may retrieve an independent set of historical
lending-
product data that is different from historical lending-product data used to
generate the one
or more statistical models of the SRM system. The SRM system may retrieve the
independent set of historical lending-product data from a financial
institution, or one or more
38

CA 03087631 2020-07-03
WO 2019/135860
PCT/US2018/064698
third-party service providers that maintain such data on behalf of a financial
institution or
on behalf of a similar organization.
[00130] At 904, the SRM system may identify, from the independent set of
historical
lending-product data, at least one independent historical lending-product
request previously
submitted by a borrower. The SRM system may use the independent historical
lending-
product request to validate an accuracy of a statistical model, before
aggregating subsets of
historical lending-product data to generate a hybrid statistical model. The
SRM system may
identify different historical lending-products requests for each statistical
model. In one non-
limiting example, a SRM system may determine that a particular statistical
model is based
on a subset of historical lending-product data that biased towards a
particular characteristic,
such as a particular geographic location. In this instance, the SRM system may
select an
independent historical lending-product request from the independent set of
historical
lending-product data that corresponds to the selection bias of the particular
geographic
location.
[00131] Further, the SRM system may further retrieve a borrower's relationship
attributes
that are associated with the independent historical lending-product request.
In one example,
the borrower's relationship attributes may be retrieved from the independent
set of historical
lending-product data (i.e. client profile data), or a repository that is
maintained by, or on
behalf of, the SRM system.
[00132] At 906, the SRM system may concurrently execute a plurality of
statistical models
using the borrower's relationship attributes associated with the independent
historical
lending-product request from the independent set of historical lending-product
data. In
doing so, the SRM system may generate borrower intermediate scores from the
results of
each of the individual statistical models; and, further, by applying a
distribution function
39

CA 03087631 2020-07-03
WO 2019/135860
PCT/US2018/064698
(i.e. a Standard Normal CDF), calculate corresponding charge-off probability
scores for
each of the borrower intermediate scores.
1001331 At 908, the SRM system may generate an accuracy score for each of the
statistical
models, based at least in part on corresponding charge-off probability scores.
For example,
the SRM system may compare each charge-off probability score with an actual
record ¨
from the independent set of historical lending-product data¨ of whether the
borrower did in
fact default on payment of the historical lending-product.
1001341 At 910, the SRM system may the select a portion of the plurality of
statistical
models based on their relative accuracy scores. In one example, the SRM system
may select
a predetermined number of statistical models with the highest relative
accuracy scores
among the plurality of statistical models. Alternatively, the SRM system may
select
statistical models with an accuracy score above a predetermined hybrid
accuracy threshold.
The predetermined hybrid accuracy threshold may be set by an administrator of
the SRM
system,
1001351 At 912, the SRM system may generate a hybrid statistical model by
aggregating the
subset of historical lending-product data associated with the statistical
models that were
selected based on their relative accuracy scores.
CONCLUSION
1001361 Although the subject matter has been described in language specific to
features
and methodological acts, it is to be understood that the subject matter
defined in the
appended claims is not necessarily limited to the specific features or acts
described herein.
Rather, the specific features and acts are disclosed as exemplary forms of
implementing the
claims.

Representative Drawing