Hindsight is said to be 20/20. In numerous business scenarios, this may be taken to suggest that when reviewing the lead up to any negative situation, the signs of the impending problem should have been recognized. As an example, a banking customer may exhibit certain behaviors prior to severing the relationship, such as changing an address, merging balances into a smaller number of accounts, incrementally reducing the number of online bill payments, visiting ATMs less frequently, or increasing electronic transfers to other financial institutions.
Looking backwards, you could point out that the collection of these actions might be indicative of a reduction in engagement, which might presage attrition. Knowing this, the best thing for the bank to do is to reach out to the customer who is poised to shut down the relationship and re-engage them someway by attempting to understand what has triggered the attrition and then to make offers to prevent it.
The problem, though, is that while it is obvious after the fact that the sequence of events occurred, recognizing that the sequence is taking place is a much more difficult challenge, for a number of reasons:
1) There is no exact sequence of events that have to happen; rather, it involves some sequence of some events that, when taken together, are indicative of the negative outcome.
2) It is not clear at which point in the sequence that the transition from “happy customer” to potential attrition occurs.
3) There are numerous events taking place simultaneously involving many individuals.
4) It is difficult to notify the right people in the bank to re-engage the customer even if the transition point can be discretely identified.
This is just one example where a business process could be improved by monitoring for (and taking advantage of) recognition of patterns of event sequences. In my next post, we will look into the problem more abstractly so that we can consider some ideas for solving this puzzle.