Leveraging Machine Intelligence for Timely & Cost Effective Underwriting

The critical first step in the commercial underwriting and risk management process, gathering and extracting financial data from client submissions, is almost universally a manual, non-responsive, error-prone, non-scalable process.  Timely and consistent application of credit policy across the enterprise is extremely challenging at best and becomes increasingly difficult to operationalize as the organization scales.

Variation in form [electronic files, pdfs, paper statements, etc.], format [unique to every company], language, accounting standards across countries and data delivery methods makes automation a very complex task. XBRL taxonomy, once seen as the panacea of Financial Statement interchange, has gone from being very rigid to virtually no standard structure at all in order to accommodate the unique reporting needs of various constituencies. Adoption rates of XBRL by private companies or by any company outside the United States is anyone’s guess. Meanwhile the challenge of extracting financial data, interpreting it consistently across every conceivable variation, normalizing it to a common standard for credit risk analysis, remains by and large unchanged.

This case study describes the approach taken by a Financial Services major in transforming its operations, dramatically reducing cycle time and establishing consistency and quality in the upstream processes of uploading cleansed and normalized Financial Statements into their internal credit scoring systems and Moody’s RiskCalc.

The Challenge: (Financial Statement Variability):

Variability in Financial Statements covers covered almost the entire spectrum of possibilities. Each statement is unique to the source/company/industry/country it came from. The five major areas of divergence from statement to statement have been organized in to the following categories. 

 Statement form and format variability. Bringing hundreds or thousands Financial  Statements of private [and public] companies, each in their own unique format, which can change without notice, is the first major obstacle to automation 

Variation in content of Financial Statements. There is no standardization in the content of a Financial Statement. Non-standard taxonomies result in inconsistent labels. Analysts often “club” data into the next most appropriate field, without full understanding of the impact.  These issues are amplified for international companies.

 Information embedded in footnotes. Critical information embedded in Financial Statement footnotes need to be identified and applied.  There is not standard or structure for footnotes for Financial Statements.

 Country and Industry specific variation. Wide variation in accounting standards across countries and accounting treatment across industries need to be normalized.

The Financial Services manual processes had to deal with all this variability and complexity on a daily basis for Financial Statements received from 46,000 companies in 35 countries annually. Processing a single statement could involve hours. The current operation was not set up to deliver the timeliness, compliance and quality required for current operations and even less to support any growth. Automating for this degree of variability was not feasible for the firm.

The Solution: (Patterns, Formulas and Rules)

 Leverage a patented extraction engine handles millions of records from a variety of sources, including paper, across many applications on a daily basis. The engine is not positional, i.e. it does not expect data to be in any specific row or column on a Financial Statement. It is able to process a Financial Statement it is encountering for the first time, throwing out for exception processing what it is not able to process automatically. Real time modification of the business logic and rules made dealing with Financial Statement variability relatively straightforward, as demonstrated time and again during the implementation.  No code was generated; the business user drove the process.

Extensive rules ensure accurate mapping of input labels. Following mapping rules drove the mapping process. If a new label that is not mapped appears in a statement it is thrown out for exception processing and the Quality Assurance team would then add a rule to automate the mapping the next time the label appears in a Financial Statement.

Key Breakups not available in Income Statement are automatically identified and extracted from footnotes and populated in normalized output using RAGE technology to interpret semi-structured data.  

Country Specific Rules allow for appropriate normalization of differences in accounting rules.  Industry Specific Rules allow for appropriate normalization of industry specific items.

Differences in the construction logic of Financial Statements are automatically recognized and handled 

It was not just normalized format, it was normalized meaning in context. This requires knowledge workers with a machine intelligence assist.

The Results: 

Compliance: Standards and credit policies institutionalized. One-offs were also handled as per policy. Risk reviews based on reliable, on-time data Automated controls. Click-back to exact extraction location in source documents.

Cost:                Over 75% cost reduction.  Transaction based pricing.

Speed:             80% of Financial Statements are instantly processed. 

Flexibility:        Major changes are rolled out in hours with minimal to no training. Complete insulation from frequently changing source file formats. 

Scalability:        Fully scalable operations. Doubling the volume at any point can be done instantly. 

Quality:          Consistency and institutionalization of the process. Implement bank specific normalization rules, country specific GAAP treatments, easily and rapidly without any programming

Operating Model:  No peak staffing, attrition, training, quality and inconsistency challenges. 

Net; Net:

The application of processes, pattern matching, analytics and business rules, all managed by business folks really delivers results. Truly a smarter process.

This is a highly summarized and anonymous case study based on Rage Frameworks  technology