Earlier this week I attended a local Business Intelligence SIG to hear Eric Siegel speak. This was essentially a preview of Eric’s keynote presentation at Predictive Analytics World next month on reducing costs with predictive analytics.
Eric is giving a webinar with me as part of the Decision Management Solutions webinar series – Optimizing Business Decisions – How Best to Apply Predictive Analytics – and gave a great introduction to predictive analytics from a cost reduction perspective. He started off with some basics, describing predictive analytics (my definition is here) and discussing how powerful they are when applied to operational decisions. Companies like to learn from their collective experience and predictive analytics is a way to do that using your data. He also pointed out that it is essential for successful application that you understand exactly what it is you are going to predict and what business decision this is going to help with. The first 5 ways to reduce costs (he actually had more) were:
- Response Modeling
Building a model that predicts who is likely to respond to a particular offer allows you to target a smaller number of prospects but get the same response, thus saving you money. If you want a 1,000 responses and average a 1% response rate you need to send 100,000 offers out. If you model let’s you find the folks who are most likely to respond you might be able, for instance, to send you just 20,000 offers at a 5% response rate saving you the cost of 80,000 offers.
- Response Uplift Modeling
A newer technique this builds on the first one. In this case you are using a model to predict who would have responded anyway so you can save the cost of marketing to them. Take our example, perhaps the people in this list are previous customers from last year’s Christmas catalog. How many would order from this year’s even if you did not send them a special offer? If you can find them and avoid sending them the offer you can save more.
- Churn Modeling
A critical tool for any subscription based business – a churn model predicts who is likely to cancel their subscription. Being able to identify those most at risk of canceling and proactively making an offer that will keep them subscribed can save you the cost of finding a new customer to replace them. And finding new customers is typically much more expensive.
- Churn Uplift Modeling
Similar to #2 this involves finding those who would not have left had you not reminded them of your existence! The cell phone customer who did not realize their 2 year contract was up and who is reminded by your retention offer that they are finally free of the penalty for cancellation for example.
- Risk Modeling
Perhaps the most established cost reduction approach – using predictors of fraud or likely bad debt to avoid the costs of these things. Models to predict how risky a transaction is in terms of fraud or a person is in terms of becoming a bad debt are well established ways to reduce costs.
Eric had others and I wrote a post last year with my take on reducing costs with decision management. He also emphasized the importance of control groups and of keeping some data out of your training set so that you can validate the model will work on data that was not used to build it.
If you want to hear more, why not come to hear Eric’s keynote at Predictive Analytics World – I am also speaking and giving a tutorial at the event (on putting predictive analytics to work) and I have a discount code for the event – just email me firstname.lastname@example.org. Eric and I will also both be speaking at the Business Rules Forum/Enterprise Decision Management Summit, a great place to come and learn more.
Finally, don’t forget my webinar next week on the 5 core principles of decision management, a great way to get yourself or one of your colleagues focused on what matters in decisioning, and Eric’s webinar the week after.