≡ Menu

Live from InterACT – Using Risk Applications to Drive Growth


Next up Discover and Fair Isaac talking about Discover’s Enterprise Decision Management initiative. Dave Wodall from Discover co-presented with Xun Shao of Fair Isaac. Discover use Blaze Advisor (rules), Model Builder (analytics) and Decision Optimizer (portfolio optimization). Discover was launched in 1985 and, like Amex, has both the network and the consumer relationship. 50M members, independent and now own the PULSE network, 3Bn transactions and $46B in receivables. Big volumes, lots of accounts and transactions and any kind of decision management must focus on this reality.

Revenue generation and loss reduction are the two keys, of course. Discover had been using decisioning a long time but needed to move to a new level. Wanted to centrally manage decisions to support common, consistent customer treatments and this meant extracting decisions from core applications and becoming true enterprise decision managers. Also needed to drive the time to market down to increase agility and respond to competitive pressure. There is a constant pressure to use data more effectively with new and more sophisticated techniques as well as new data sources and touch points. Right treatment for the right customer at the right time.

Key goals for the decision management program:

  • Advanced Analytics
    Manage complexity, continual improvement, more advanced strategies into more applications, new data attributes and increased targeting precision
  • Business Agility
    Business control, user friendly test environment for non-technical users, less reliance on IT.

Discover has a pretty classic view of decision management – automating, improving and connecting decisions to enhance performance. The connection piece, something you don’t always see in the definition of decision management, is important for a company like Discover with lots of decision points.

Discover started using rule-based criteria for decision management with limited analytics. The next stage involved data mining and statistical analysis to come up with better rules. Now integrating predictive analytics with this to increase precision. BTW this is the classic sequence Neil and I outlined in the book. Next up for them is moving to customer level decisions and net present value optimization. Clearly this process never ends, with new technology and techniques getting added all the time.

Xun came on to talk about how they approached the project and what some of the steps were:

  1. Developed a joint decision management vision and action plan
    “To improve the speed, precision and consistency of customer risk decisions and to rapidly respond to changing customer data and business and/or economic conditions through the use of advanced decision management technology”.
  2. Collaborated to build and enable a world class team
    Initially a joint team but it had to trend towards a Discover-only team. Took a resource plan, lots of knowledge transfer and mentoring.
  3. Design and implement decision management infrastructure
    Created an enterprise repository for rules and models and built a high performance architecture suitable for Discover’s volumes.
  4. Institutionalized best practices
    Developed and documented best practices and facilitated cross-departmental exchanges.
  5. Built solution based on Fair Isaac’s technology and Discover’s expertise
    First few solutions all contained the key elements – predictive analytics, business rules, adaptive control and data management/integration.

Dave wrapped up with some results. First application was in 2006 (Discover started partnering with Fair Isaac over this decision management platform some years ago and was one of the first to implement enterprise decision management or even talk about it) and use has exploded since then. Discover has developed a strong in house competency and now automate millions of customer decisions per day across the customer lifecycle. Tracking against their goals:

  • Leveraged Advanced Analytics
    10 new decision management applications (2 online, 8 batch) and more than 100 model scores introduced in these applications. Now doing customer analytics, not just account level, and can now pick rules/models/strategies as needed for each application.
  • Increased Business Agility
    600 TPS was the online SLA and has been exceeded comfortably and thousands of TPS in batch. Zero downtime and no transaction timeouts in online transactions. Can now tune and redeploy rules very rapidly

Mark listed the huge array of decisions, both batch and online, that Discover is now managing – very impressive. All these store the results of decisions in a data warehouse that itself feeds the analytics and so closes the loop.

Some lessons learned:

  • Data infrastructure really important
    Combined analytic and production data so not separate feeds.
  • Competing priorities become an issue when enterprise-wide
  • Simulation environment very important prior to production
  • Plan for ongoing architecture expansion
  • Properly trained resources are important
  • Senior management commitment needed

Discover is, in many ways, a poster child for enterprise decision management. Very impressive.