Recommender engine for collections treatment selection

• US 9,792,653 B2
• Filed: 12/13/2011
• Issued: 10/17/2017
• Est. Priority Date: 12/13/2011
• Status: Active Grant

• Current Assignee: Electrifai LLC
• Original Assignee: Opera Solutions U.S.A., Llc
• Inventors: Gancarz, Gregory, Zhang, Jenny, Crowe, Kathleen Marie
• Primary Examiner(s): Goyea, Olusegun
• Application Number:
  US13/324,445
• Publication Number:
  US 20130151383A1
• Time in Patent Office: 2,135 Days
• Field of Search: 705 30, 705 35, 705 36 R, 705 38, 705 40, 705 45, 705 7, 705500, 705 12, 705 731
• CPC Class Codes: G06F 16/00 Information retrieval; Database structures therefor; File system structures therefor
• CPC Class Codes: G06F 17/40 Data acquisition and logging for input to computer G06F3/00
• CPC Class Codes: G06N 20/00 Machine learning
• CPC Class Codes: G06Q 30/00 Commerce
• CPC Class Codes: G06Q 40/00 Finance; Insurance; Tax strategies; Processing of corporate or income taxes
• CPC Class Codes: G06Q 40/02 Banking, e.g. interest calculation or account maintenance credit or loans G06Q40/03
• CPC Class Codes: G06Q 40/08 Insurance
• CPC Class Codes: G06Q 40/12 Accounting

Abstract

A system for automatically recommending treatments for delinquent accounts in collections is provided. The system includes one or more sub-models for analyzing and scoring delinquent accounts and comparing them to historical accounts to identify similar historical accounts. The system can select and recommend one or more treatments for the delinquent account based on treatments for the historical accounts that may have previously provided a favorable outcome for a similar account. The system may test the efficacy of new treatments and incorporate the results into the memory, thereby learning and adapting to new treatments and account behavior.

Claims

1. A system for recommending treatments for one or more delinquent accounts at a particular point in time comprising:

a processor operable to execute programs;

a memory comprising one or more sub-models and reference data for one or more prototype accounts, the prototype accounts being a subset of historical accounts stored in the memory and including results from test accounts created to test the efficacy of new treatments or new sets of rules, thereby providing a mechanism for learning and adapting to new treatments and debtor behavior;

a database associated with said processor and said memory;

a program stored in the memory and executable by the processor, the program being operable for;

expressing the reference data as historical vectors in an n-dimensional feature space;

receiving at least one recommendation request data record for each of one or more delinquent accounts;

converting the at least one recommendation request data record to request vectors using scores from the one or more sub-models;

selecting one or more neighbor vectors from the historical vectors for each of the request vectors, based on an algorithm analysis of data in the n-dimensional feature space;

determining neighbor fitness values for each of the selected one or more neighbor vectors by looking forward in historical time for each of the one or more prototype accounts in the one or more neighbor vectors;

generating, for each of the request vectors, a subset list of allowed treatments by applying one or more sets of policy or regulation rules to a superset list of all treatments, the subset and superset lists being generated regardless of similarities between the one or more delinquent accounts and the one or more neighbor vectors;

computing a treatment fitness score for each of the treatments in the subset list, using the neighbor fitness values from the neighbor vectors that experienced a particular treatment being analyzed and using events outside the one or more neighbor vectors;

selecting at least one of the treatments in the subset list for each of the at least one recommendation request based on the treatment fitness score to recommend treatments for delinquent accounts based on the success of treatments in the subset list; and

presenting the selected at least one treatment to a user,

wherein the one or more prototype accounts in the memory increases speed of operations of the program when the program is executed by the processor.

2. The system of claim 1, wherein the delinquent accounts are selected from the group comprising delinquent accounts, accounts in collection, historical accounts, test accounts, new accounts or combinations thereof.

3. The system of claim 1, wherein overall categories of the allowed treatments may include specific treatment to use, variant of the treatment, sequence of treatments, phone number to use, time of day to call, prioritization of accounts to receive treatment, specific collector to work an account.

4. The system of claim 1, wherein the one or more rules are selected from a group comprising policy and regulation expert rules, policies of the user, any applicable governmental regulations, financial industry regulations or combinations thereof.

5. The system of claim 1, further comprising the step of providing a rationale for recommending the selected at least one treatment.

6. The system of claim 1, wherein the sub-models are built using one or more models selected from a group comprising neural networks, regression analysis, integrated rules systems, decision tree models, other statistical models or combinations thereof.

7. The system of claim 1, wherein the reference data comprises account demographics, treatments, model scores, account properties, account behaviors, days past due, the final collections result on the account or combinations thereof.

8. The system of claim 1, wherein the reference data comprises multiple entries for historical account data at different points in time.

9. The system of claim 1, wherein receiving the one or more recommendation request data, further comprises receiving a request for recommendation or notification record from a client device in a predefined API or XML format.

10. The system of claim 1, wherein the algorithm includes a calculation of the similarity between two or more n-feature vectors in an n-dimensional feature space.

11. The system of claim 10, wherein the similarity is a calculation of the distance between two or more n-feature vectors in an n-dimensional feature space.

12. The system of claim 10, wherein the algorithm calculation is selected from a group comprising Mahalanobis distance calculation, Manhattan distance formulas, Euclidean distance formulas, Pearson correlation coefficient, Spearman rank correlation, Kendall'"'"'s τ

correlation, mean squared differences, entropy, adjusted cosine similarity or combinations thereof.

13. The system of claim 1, wherein the treatments are selected from a group which may include a letter sent by mail to the debtor'"'"'s home, a letter sent by mail to the debtor'"'"'s work, an email message sent to the debtor'"'"'s email address, an SMS message sent to the debtors telephone, a phone call made by a collection specialist to the debtors home, a phone call made by a collection specialist to the debtors office, no action, restructuring the delinquent account, forgiving a portion of the delinquent account, utilizing an automatic/automated agent, resending the bill, debt counseling, forbearance program, payment matching program, settlement agreement, running a skip trace search, multiple smaller payments, legal action or combinations thereof.

14. The system of claim 13, wherein the treatments are defined by a time duration, numeric value, frequency or combinations thereof.

15. The system of claim 14, wherein the time duration is selected from a group comprising days, weeks, months, quarters, years or combinations thereof.

16. The system of claim 14, wherein the numeric value is a settlement percentage.

17. The system of claim 14, wherein the frequency is the number of times the treatment is repeated within a fixed time period.

18. The system of claim 1, wherein the fitness score comprises performance measures fed into a fitness function which summarizes the performance measures into a single score.

19. The system of claim 18, wherein the performance measures are selected from a group comprising the average cost incurred on the delinquent accounts through the time the delinquent account exited collections, the recovery amount of the delinquent account, time to recovery on the delinquent account, number of payments, the final outcome on an account, recovery success percentage for the delinquent account, partial payments, completion of payment programs or combinations thereof.

20. The system of claim 18, wherein the fitness function comprises the overall average for the entire historical group and the average for the neighborhood group, weighted by the number of historical accounts in the neighborhood group.

21. A computer-implemented method for recommending treatments for one or more delinquent accounts at a particular point in time the method comprising the computer-implemented steps of:

accessing one or more servers, the one or more servers comprising a database with one or more sub-models, and a memory comprising reference data for one or more prototype accounts, the prototype accounts being a subset of historical accounts stored in the memory and including results from test accounts created to test the efficacy of new treatments or new sets of rules, thereby providing a mechanism for learning and adapting to new treatments and debtor behavior;

expressing the reference data as historical vectors in an n-dimensional feature space;

receiving at least one recommendation request data record for each of one or more delinquent accounts;

converting the at least one recommendation request data record to request vectors using scores from the one or more sub-models;

selecting one or more neighbor vectors from the historical vectors for each of the request vectors, based on an algorithm analysis of data in the n-dimensional feature space;

determining neighbor fitness values for each of the selected one or more neighbor vectors by looking forward in historical time for each of the one or more prototype accounts in the one or more neighbor vectors;

generating, for each of the request vectors, a subset list of allowed treatments by applying one or more sets of policy or regulation rules to a superset list of all treatments, the subset and superset lists being generated regardless of similarities between the one or more delinquent accounts and the one or more neighbor vectors;

computing a treatment fitness score for each of the treatments in the subset list, using the neighbor fitness values from the neighbor vectors that experienced a particular treatment being analyzed and using events outside the one or more neighbor vectors;

selecting at least one of the treatments in the subset list for each of the at least one recommendation request based on the treatment fitness score to recommend treatments for delinquent accounts based on the success of treatments in the subset list; and

presenting the selected at least one treatment to a user,

wherein the one or more prototype accounts in the memory increases speed of operations of the program when the program is executed by the processor.

22. The computer-implemented method of claim 21, wherein the delinquent accounts are selected from the group comprising delinquent accounts, accounts in collection, historical accounts, test accounts, new accounts or combinations thereof.

23. The computer-implemented method of claim 21, wherein overall categories of the allowed treatments may include specific treatment to use, variant of the treatment, sequence of treatments, phone number to use, time of day to call, prioritization of accounts to receive treatment, specific collector to work an account.

24. The computer-implemented method of claim 21, wherein the one or more rules are selected from a group comprising policy and regulation expert rules, policies of the user, any applicable governmental regulations, financial industry regulations or combinations thereof.

25. The computer-implemented method of claim 21, further comprising the step of providing a rationale for recommending the selected at least one treatment.

26. The computer-implemented method of claim 21, wherein the sub-models are built using one or more models selected from a group comprising neural networks, regression analysis, integrated rules systems, decision tree models, other statistical models or combinations thereof.

27. The computer-implemented method of claim 21, wherein the reference data comprises account demographics, treatments, model scores, account properties, account behaviors, days past due, the final collections result on the account or combinations thereof.

28. The computer-implemented method of claim 21, wherein the reference data comprises multiple entries for the historical account data at different points in time.

29. The computer-implemented method of claim 21, wherein receiving the one or more recommendation request data, further comprises receiving a request for recommendation or notification record from a client device in a predefined API or XML format.

30. The computer-implemented method of claim 21, wherein the algorithm includes a calculation of the similarity between two or more n-feature vectors in an n-dimensional feature space.

31. The computer-implemented method of claim 30, wherein the similarity is a calculation of the distance between two or more n-feature vectors in an n-dimensional feature space.

32. The computer-implemented method of claim 30, wherein the algorithm calculation is selected from a group comprises Mahalanobis distance calculation, Manhattan distance formulas, Euclidean distance formulas, Pearson correlation coefficient, Spearman rank correlation, Kendall'"'"'s τ

correlation, mean squared differences, entropy, adjusted cosine similarity or combinations thereof.

33. The computer-implemented method of claim 21, wherein the treatments are selected from a group which may include a letter sent by mail to the debtor'"'"'s home, a letter sent by mail to the debtor'"'"'s work, an email message sent to the debtor'"'"'s email address, an SMS message sent to the debtors telephone, a phone call made by a collection specialist to the debtors home, a phone call made by a collection specialist to the debtors office, no action, restructuring the delinquent account, forgiving a portion of the delinquent account, utilizing an automatic/automated agent, resending the bill, debt counseling, forbearance program, payment matching program, settlement agreement, running a skip trace search, legal action, multiple smaller payments, or combinations thereof.

34. The computer-implemented method of claim 33, wherein the treatments are defined by a time duration, numeric value, frequency or combinations thereof.

35. The computer-implemented method of claim 34, wherein the time duration is selected from a group comprising days, weeks, months, quarters, years or combinations thereof.

36. The computer-implemented method of claim 34, wherein the numeric value is a settlement percentage.

37. The computer-implemented method of claim 34, wherein the frequency is the number of times the treatment is repeated within a fixed time period.

38. The computer-implemented method of claim 21, wherein the fitness score comprises performance measures fed into a fitness function which summarizes the performance measures into a single score.

39. The computer-implemented method of claim 38, wherein the performance measures are selected from a group comprising the average cost incurred on the delinquent accounts through the time the delinquent account exited collections, the recovery amount of the delinquent account, time to recovery on the delinquent account, number of payments, the final outcome on an account, recovery success percentage for the delinquent account, partial payments, completion of payment programs or combinations thereof.

40. The computer-implemented method of claim 38, wherein the fitness function comprises the overall average for the entire historical group and the average for the neighborhood group, weighted by the number of historical accounts in the neighborhood group.