≡ Menu

Live from InterACT – An Enterprise Decision Engine for Originations


Wednesday begins with Antonio Paulo Conde from Citibank Brazil talking about an enterprise decision engine for originations in their retail bank.   Measuring and understanding the risk of new financial products is important and became more so as the sales teams gained more negotiation power. To address this Citibank used Blaze Advisor to build an originations – a credit initiation – engine. Originations is different in Brazil as credit bureaus only have negative data (unlike the US) and so scoring with internal data is even more important than in the US.

Citi is a very large, well established bank (100 counties, 200M customers etc) and in Brazil it has 400,000 customers and 6M cards issued.

The platform covers credit cards, personal installment loans, checking/savings, credit line and overdraft protection. Everything is processed through the same process so that a sales representative can work with a customer to develop whatever products they want. After the sales representatives enter the information it goes through an automatic underwriting process – built in Blaze Advisor – to assign the products, limits etc. The system allows the sales team to enter information in a flexible sequence and the rules fire as needed. These rules might pull bureau data, income data, pre-screened offers etc. The system’s key screen is one for budget management that allows ALL the products to be reviewed and edited. All the rules are applied to see what can be approved, what needs manual review etc. This allows the sales people to simulate how different collections of products might work out for a customer.

The system replaced a typical manual process with multiple systems, policies manual and many contracts filed away. In the old system an application averaged 1.6 passes through the process due to repeats etc. Their target segment is very time sensitive so this was really problematic. With the new system they estimate that 70% will be automated and only 30% will need manual review.

The implementation has about 600 rules in many rulesets and decision tables as well as a scorecard to implement a model. Business users can change the rules without IT and it only took 3-4 months from rules harvesting to user acceptance testing. Having the flexibility to change the rules without a large technology effort was critical to Citi and the system includes rule maintenance applications to support the non-technical users. Although most of the engine was online, interactive, they needed a batch process to support a legacy system and were able to also develop this in Blaze Advisor.

Rulesets developed include:

  • Pre-approved offer rules (to manage how to deal with offers sent out by mail or through outbound calls)
  • Bureau rules (to handle one or multiple applicants and act on the results)
  • Guarantor and collateral rules
  • Income and income proof rules
  • Line assignment rules including scorecard and limit calculations
  • Alternative maximum rules so that the sales person does not have to check every possible value
  • Risk acceptance or credit rules like percentage of disposable income required for loan etc. These are the core credit policies.
  • Documentation requirement rules – forms, disclosures etc.
  • Fraud detection rules that change rapidly and often to combat fraudsters

He identified a number of success factors in the project:

  • Multi-functional team – risk management, sales, operations – not a single “owner”
  • Clearly defined point person in each area
  • Important to balance flexibility and feasibility
  • Strict scope control
  • Direct access to senior management