There was an interesting article from McKinsey this month – The benefits—and limits—of decision models – by Phil Rosenzweig.
As Phil points out using models (and he is talking about analytic models) helps avoid common biases that undermine leaders’ judgments such as over confidence, a focus on recent data etc. And I would add not just leaders – it also helps avoid the biases of staff throughout the organization. In fact the examples – credit card fraud (not really a recent application but still a great one), loans,pricing, insurance – are not made by “leaders” at all but by front line workers or automated systems. They are operational decisions and the value of using analytic models in these decisions is, if anything, even greater as the people concerned have less time to decide, less experience and often less to gain/lose from the outcome.
Phil draws distinction between things people cannot influence and things they can where “the task isn’t to predict what will happen but to make it happen”. He makes the great point that we are largely predicting things we cannot control using these models. I would add, though, we are predicting these things so we can use the predictions in decisions we do control. For instance, we can’t control the risk that someone will not repay their loan, only predict it. But we can control the rate we offer for the loan to reflect that risk. Plugging these analytic models into a true decision model shows us how they can be used to improve decision-making very explicitly.
Phil also makes the great point that using a model changes the outcomes and so changes the data from which the model is built. This is why it is so important to refresh models regularly, run experiments to see how well models are performing and close the loop with ongoing decision analysis.
Cross-posted from JTonEDM.