I am attending InterACT San Francisco 2007 this week and blogging live (or nearly live) from the sessions. This session is "Continuous Pricing. A Case Study in Captive Optimization" from Dell Financial Services.
DFS is (or soon will be) a captive financing arm for Dell. Target commercial, SMB and consumers with revolving accounts, leases and loands. 42,000 applications per day now, though has some seasonal variation like back-to-school. Dell Financial Services (DFS) had a hard-coded, inflexible system for doing pricing for financing of Dell computers. This was the problem DFS worked with Fair Isaac to solve. The system had long lead-times for changes that had to made by IT (3 months for instance). The system require a large staff to handle its problems and, despite that, only 65% of transactions were reviewed for fraud and risk bands were over-wide and overlapping creating very poor control of pricing. Old system only considered the credit history of a customer, no accounting for the profitability of the product or the current state of Dell. The old organization was siloed and had real problems sharing data or connecting decisions. Problems in their approach also meant that they were being adversely selected against.
The business rules team are part of the business and have control of rules including those derived from SAS models
- Integrated risk and pricing models developed by inhouse decision sciences group
Built in SAS, and all successfully implemented in Blaze Advisor
- Credit and fraud management integrated
Blaze Advisor at the core for business rules (FAQ)used with Falcon Fraud Manager for fraud detection
Both from a technical perspective and supporting all the different silos in a loosely coupled way
- Continuous pricing is the objective
And the new system is getting there
- Common framework and data model across all the business silos
- Incremental rollout
Means that the data sometimes has holes as some silos are not integrated but models are upgraded as and when data becomes available
- Separate instances of Blaze Advisor for fraud and originations
But sharing a common rule repository
- The blended economics of the deal are now taken into account e.g. if an upsell to the customer makes it a more profitable deal then the financing can be made more competitive.
- One result has been dropping of fraud review rate to 18% thanks to targeting.
- Processed 2.2M applications as of April 2007
- Incremental originations: 12%
- Slower liquidation: 66% of originations remaining for UDE vs. 60% for the existing system, after 4.5 months
- Better credit quality: 60+ delinquency rate is 15% lower for UDE than for the existing system, after 4.5 months
Ultimately they would like to make this be interactive so that, as you configure a machine, a financial deal is constructed in parallel to maximize the economic value to Dell. I was interested particularly in their discussion of the organizational change required to adopt EDM and EDM applications. I blogged a little about this before but will do so more.
Here are some selected slides:
You can find the full set of posts from InterACT in this category.