Input

I. Basic information for risks evaluation:

Loans / Debts details:

1. •Debt body
   

2. •Debt percents, penalties etc.
   

3. •Loan date
   

4. •Due date
   

5. •etc.
   

Client personal details:

1. •Name
   

2. •Address and phones
   

3. •Work name
   

4. •Work phones
   

5. •etc.
   

Other details:

1. •Legal actions already undertaken
   

2. •Bank branch, clerk ID
   

3. •Scoring rate obtained at credit authorizing stage
   

4. •etc.
   

2. II.Information for collection scoring:
   

Risk factors from Part I:

1. •Family relations between clients
   

2. •Addresses, phones, work names matching
   

3. •Credit risk
   

4. •Legal risk
   

5. •etc.
   

Compound factors for scoring:

1. •Percentage to debt body ratio
   

2. •Timely payments ratio
   

3. •Psychologic level of debt amount (‘large’ or ‘small’)
   

4. •etc.
   

Debt recovery statistics:

1. •Person contact attempts number
   

2. •Dunning intensity
   

3. •Recovery statistics wrapped into neural net
   (Neural net is trained at scheduled times, thus adapting to slow changes in statistical data gathered by bank).
   

3. III. Information for automatic strategy selection:
   

1. •Recovery prognosis for each account (from Part II)
   

2. •Recovery activities costs
   

3. •Dawn recovery costs per each client
   

4. •Costs of non-acting
   

5. •Employees number (resources)
   

6. •Employees skills
   

7. •Other resources and business limitations
   

File Formats

CSV file format (plain text) is used for input data. Most input data comes in single file, while payments and events data come in separate files.

Other methods of data integration are possible by request.

Fraud Detection

Methods:

1. •Debt parameters preprocessing by fuzzy rules
   

2. •e.g. ‘is debt amount big?’
   

3. •e.g. ‘is overdue time long?’
   

4. •Behavior evaluation
   

5. •timely payments No., Amt.
   

6. •how easy to contact the client
   

7. •recovery actions intensity
   

8. •Fraud and risk factors detection
   

9. •lookup for clients interconnections
   

10. •all available data is used for fuzzy text search
    

11. •permanent knowledge base update for risk factors.
    

Techniques:

1. •Fuzzy set algebra
   

2. •Fuzzy text search
   

3. •Clustering.
   

(more details...)

Collection Scoring

Methods:

1. •Individual cash-flow prognosis for each account
   

2. •Accounts classification
   

3. •Prognosis adjustments by risk level
   

4. •Time-based recovery prognosis
   

5. •Loans portfolio cost evaluation.
   

Techniques:

1. •Neural nets
   

2. •Financial analysis.
   

Automatic Strategy Selection

Methods:

1. •Use neural net to predict debt recovery vs. collection strategy
   

2. •Calculate costs for different collection strategy e.g.:
   

3. •dunning cost (telephone, email, salaries)
   

4. •tariffs based on debt amount (lawyer fees, taxes etc.)
   

5. •Apply business limitations like resources availability, employees skillset, business priorities.
   

6. •Maximize total income prediction vs. total cost.
   

Techniques:

1. •Neural nets
   

2. •Discrete linear programming.
   

Output

1. •Plain text CSV data files for easy integration with IT infrastructure, contains:
   

2. •values of all calculations
   

3. •values of predicted cash flow
   

4. •recommended strategies
   

5. •PDF reports
   

6. •Database storage for narrative data analysis.
   

User interface

1. •Thin client - web browser
   

2. •Web 2.0 Ajax based controls
   

3. •Animation of narrative data (‘Radar’ display)
   

4. •Graphical selection and filtering tools
   

5. •Combined narrative and prognosis trend charts
   

6. •Graphical display of interlinks with point and click traversing to linked accounts
   

7. •Shopping cart - like approach to accounts segmenting and manipulations
   

8. •Graphical display of fuzzy rules risk calculation (cause-effect diagram)
   

9. •Expert knowledge entry that overrides computed risk factors and debt recovery prognosis.
   

Technology

Collect Advantage has been developed with help of true pearls of Open Source software and is delivered to client in open source. Most important components:

1. •web2pyTM Enterprise Web Framework
   

2. •Python programming language
   

3. •jQuery and flot
   

4. •NumPy the fundamental package needed for scientific computing with Python
   

5. •matplotlib a python 2D plotting library
   

6. •ReportLab industry-strength PDF generating solution
   

7. •GLPK GNU Linear Programming Kit
   

8. •Graphviz - Graph Visualization Software.