Let’s face it; not knowing whether or not you’re measuring up to expectations is stressful. To avoid that stress, you need to know what you’re measuring, how to measure it accurately, and what those expectations are. With decision management strategies, as with other aspects of life, a clear understanding of those things can go a long way towards ensuring a satisfactory experience. This brings us to the subject of today’s missive: correlating the decisions provided by your decision service with the KPIs (Key Performance Indicators) they are expected to influence.
The challenge with this is that there is often no logically explicit link between the decision data and the data used to measure the KPI. For example, suppose you sell widgets. The goal of selling widgets is to make a profit, so profit is the KPI. Your sales manager believes that you can increase profits by offering various incentives to different classifications of customers: new customers get a “welcome discount,” gold customers get a “thank you discount,” platinum customers get an invitation to a cruise, and dead-beat customers get an email that says “go away.” The logic for classifying the customers and applying the incentives is built into a decision service, which consistently applies the incentive strategy to all customers.
That’s a great start, in that you now have a precise way to know what incentives were offered to which customers. Some basic arithmetic on your sales data should be able to tell you how many widgets each customer then bought, and how much profit you made on them. But that’s not the whole story. You know what the strategy was, and what the profits were, but you don’t know how much the strategy directly contributed to the profits. Those customers might have bought just as many (or more) widgets without the incentives, and the cost of the incentives might have been higher than the revenue from increased sales. How do you correlate the strategy results to the KPI?
This is where the art and science of decision management gets interesting. A business rules-based decision service provides inherent value by applying a decision policy quickly and consistently (hey, did I mention that we have a really great cloud-based product that makes it really easy to build decision services?). But it also provides something of equal or possibly greater value—accurate, up-to-date, coherent data on how the policy was applied. This data can then be applied to more advanced decision management tools and techniques for optimization, data mining and machine learning, champion-challenger testing, etc., to continuously evolve and improve the policy.
Remember that the goal is to positively influence your KPIs. Going back to the widgets example, we have a decision service that produces clean, accurate, normalized data about who bought what after being given which incentive, and we know what the profit was. Now we can apply a whole bag of tricks to that data to see how changing the policy affects the KPI.
- We can use champion-challenger testing, where you tweak the rules to come up with an alternative service (the challenger) and apply it to a subset of the population. This lets you do head-to-head comparisons of the results.
- You can do optimization modeling to find the best balance between competing factors (e.g. what is the best discount to increase sales while maintaining a viable profit margin).
- You can use machine learning to derive predictive models (e.g. to predict the likelihood that a particular customer will accept a particular offer).
If you’re interested in knowing more about the nuts-and-bolts of decision management, FICO World 2016 has a track called Practical and Tactical: Decisioning for Techies. It’s a great opportunity to learn from the experts!
Post originally appeared in FICO Community Blogs