Time for a quick session or two at FICO World. Chip Clarke and Andrew Beckman presented on the use of customer-level TRIAD to continuously improve their collection results for their retail banking products – to do “Adaptive Control”. Adaptive control, for those not familiar with the term, means continually challenging the way you make decisions to see if there is a better way to make those decisions. Chip’s group weaves analytics into a whole series of things with a focus on using analytics to continuously improve how they operate. KeyBank is based in Cleveland and is one of the US’ largest bank with $90Bn in assets. Over 1,0000 branches and 1,500 ATMs as well as their website.
KeyBank runs customer-level TRIAD and was an early adopter of the customer-centric (rather than account-centric) product. This gives them a full view of their customers and households so they can make decisions based on the totality of their customer relationship rather than having each product team focus on their own accounts. KeyBank is using TRIAD to make account-level decisions used customer-level (multiple-account) enterprise data. KeyBank is not yet making holistic customer-centric decisions so much as using customer-level, multiple account data to make better account-level decisions. KeyBank’s use of TRIAD covers consumer lines and loans, business lines and loans, education loans, equipment finance and DDA overdrafts.
To make decisions about delinquent accounts, KeyBank use account level, customer level and household data multiple internal account servicing systems. They use data from their data warehouse, a collections data warehouse (with promises to pay promises kept, calls made etc) as well as external data such as that from credit bureaus. They use this data in TRIAD to prioritize delinquent inventory based on the severity of the default and the probability of payment. They also decide about changes to customer’s lines of credit (reducing the amount of a line for instance) using the same process.They have extended this with FICO’s Business Rules Management System FICO Blaze Advisor to develop custom predictive characteristics as well as empirical (non-data-driven) risk scorecards.It has been important for them that their business team can make the changes they need without IT engagement. This is important given the high tempo of challenger strategies within their adaptive control environment.
A consistent data source is really important. Strategy development and reporting use the same data source so that if the analytics say that the decision-making approach will affect a certain number of customers then the report should come back with the same number. To do this they created the equivalent of an analytic data mart that drives the TRIAD decision-making, predictive analytic model development as well as reporting. All of the approaches used are very transparent not only in terms of using easy to read models and rules but also in terms of strategy proposals, documentation of every step, scorecard validations being posted to internal website.
From an implementation perspective, KeyBank had to begin with building consensus across the different groups who collectively “own” the decision-making. Once implemented they have improved loss mitigation and exposure management. They have reduced collections efforts by focusing on customers and they met their business case return within 12 months. The core team is truly cross-functional with consultants, business analysts, programmers, risk/compliance folks and technology people.
Lessons learned included building a cross-functional team, owning the technology in the business and making sure that the decision management project is tied to company’s overall objectives. Next up are NPV and propensity to pay scoring, more fine grained segmentation of accounts, analytic placement of collections to specific collections agencies and plan to develop optimized collection and recovery strategies.