Predictive Modeling Tools for Analytic Teams

The other session I got to see at FICO World was a panel discussion on analytic modeling tools with representatives from UniCredit (Italy), Ferratum (Finland), Nedbank (South Africa), and Citi.

Challenges and rewards of building an analytic modeling team

The other session I got to see at FICO World was a panel discussion on analytic modeling tools with representatives from UniCredit (Italy), Ferratum (Finland), Nedbank (South Africa), and Citi.

Challenges and rewards of building an analytic modeling team

• Always start from scratch with fresh, inexperienced teams and this is both challenging and rewarding
• Challenges are personal and interpersonal
• Need the right tools to get positive outcomes quickly and to adapt to problems to keep momentum
• Need for math and logic skills can be a challenge when the education system has not delivered the skills you need
• If you have people without the skills you need you have to develop these skills and then fight to retain these people
• Current regulatory environment means that skilled people are spending time on justification and documentation

How much of a regulatory challenge is there?

• Had to make a lot of process changes (how models were developed) to respond to regulations
• Basel II regulations doubled the time to get a model built and deployed(!) and this has forced re-engineering of the processes to reduce this time again.
• Supporting the regulatory and audit processes within the tool, within a graphical modeling tool, made a big difference
• Regulatory environment is broader now with “everything” being considered a model – created large workload for the scoring team

Challenges in the years ahead

• Danger is focusing on regulations rather than on helping the business improve decisions!
• “Don’t bring a knife to a gunfight” – don’t let old models built in a different time rule your business
• Coping with macro-economic cycle (such as number of consumers with credit problems) and integrating this into predictive models
• Tradeoffs between choices and integrating this into a coherent portfolio view
• Streamlining the review and deployment process
• Politics and changing regulations, especially given the US Presidential Election.

There was some good interaction with the audience answering questions but they were hard to capture for the blog post. Some notes though:

• Vitally important to improve the time to build and deploy models
• Regulatory oversight is continuing to increase and not likely to get any easier any time soon
• And international regulation is increasingly driving things also and will reduce the variation between countries
• Managing data is critical
• NEVER lose sight of the business object of a predictive analytic model