(Posted by guest blogger, James Taylor)
I recently started reading an interesting new blog - the Decision Strategist - and saw a post titled New Experiences Improve Decision Making. It was an interesting (if short) post that made the critical point that "one of the best ways of improving our decision making was to vary our experiences". This is clearly true and the author goes on to give some good reasons for this (of which more in a minute). What to do, though, if you have automated the decision? Say you have built a decision service to automate and manage an operational decision. How can you ensure that this decision service continues to improve its decision making? How does a service get "new experiences"?
A decision service can get new experiences in two ways - automatically and manually. Ensuring that a decision service gets new experiences automatically means either developing adaptive analytics that respond to new data as it is collected or using adaptive control. Adaptive analytics are sometimes appropriate - when you get very quick feedback on how good a decision is and when variations are not driven by your own actions but by the market as a whole. Adaptive control, the establishment of multiple decision strategies that can be tracked and compared over time, can always be used, however. By running some transactions through alternatives and by tracking the result you can establish a decision service that, more or less, learns from a greater variety of "experiences".
Manually driving the results of broader experience into a decision service is a business intelligence/performance management issue in large part. If the business users who control the rules in your decision service have good reporting and performance management tools, then they can use these to investigate their business. Widening the range of information being considered allows them to make changes to their rules that reflect a wider experience. Of course, they have to want to do this. Even good tools can still be used to merely reinforce existing prejudices.
The post identifies four specific ways in which this helps and three seem to be relevant here:
- Correcting the confirmation bias. Not only do wider experiences help with this, the use of predictive analytic models to replace human judgment can often do this very effectively (such as in the use of credit scoring to replace loan officers value judgments).
- Expanding automatic associations. As the post notes "a subconscious preference ... can lead to suboptimal decision making" and gathering, and analyzing data more widely can help correct this. Using descriptive analytics to understand the true associations can be particularly effective.
- Enlarging our idea of normal. We might think that our normal approach will work better but adaptive control can show us that an alternative will work better.
Just because you are automating a decision, don't forget to give your decision services new experiences when you can.