≡ Menu

Some thoughts on Adaptive Decision Management


Paul Haley wrote an interesting piece last week on Adaptive Decision Management. This is a really good piece and I highly recommend it. I do have a couple of things to add, however.

While Paul is correct that Fair Isaac did not talk about “decision management” until after the HNC merger, Fair Isaac had a long history prior to that in doing decision management – combining predictive analytic models and scores with business rules into what they called “strategies”. Post the HNC merger we simply formalized a name to describe what both companies had been doing. Fair Isaac/HNC Products like Falcon, TRIAD and Capstone are not what you might call traditional applications but neither are they tools. They are decision services (wiki) with software for case management, data integration and other related tasks wrapped around them. They plug into backbone applications (core banking systems, Loan Origination Systems) and provide those applications with the critical decisions they need.

But, enough history. Like Paul I think that dashboards, while useful, are not enough (see this post, for instance) and I agree that that neither rules nor predictive analytics alone are enough for true decision management. That’s why decision management also has something we call adaptive control and optimization (wiki). Indeed the book has a whole chapter on it and Paul goes on to discuss it a little – it is often called Champion/Challenger testing or A/B testing. Today, this is largely a manual feedback loop.

Automatically learning analytics are clearly the next step and were one of the future technologies we discussed briefly in the book. There are some challenges – decisions where the consequences take a long time to play out can be challenging and you don’t want models to adapt to changes you caused, for instance. Say you ran a SuperBowl ad and this generated a vast increase in page hits. Well you would hardly want your analytics to start adapting to it as though this were just a random increase. That said, for many problems you can not only automate the adaptive part you can even automate the champion/challenger testing and work to produce algorithms that generate challengers that will teach you something useful is well underway.
Like Paul I think this is an area where there is still much to do but there is plenty of raw technology out there with which to do it.

One last comment, though. If you have not externalized and managed your decisions you will not be able to have them adapt effectively. As one of the founders of Fair Isaac famously said “if you want to improve a decision you have to grab it by the throat and not let go”.