Test yourself on five basic questions on how well you use your data.
The transition from reporting to data warehousing to business intelligence to analytics in the last 20 years has been accompanied by a lot of innovation layered on top of a dramatic, continuing decrease in storage costs and a concurrent increase in the availability of low cost processing power. We all know the story—new tools and new data should increase the power of data to solve thorny insurance business problems like pricing to the risk, smarter risk selection, and prevention of fraud.
But are we really solving those kinds of problems with our analytics?
The short answer is mostly no. Yes, new tools have lowered the cost of cleaning and storing legacy data. And there are new sources of data, much of it unstructured, much of it usefully related to identifying the propensity of a prospect to buy or an insured to file a claim. However, in most cases, insurers are just taking old-style data warehousing and simply calling it analytics. The person consuming this data is still responsible for performing the actual analysis in this more accurate, easier to read format. That’s a plus, but it isn’t analytical. Worse, it may actually be raising labor costs as carriers pay analytics wages and fees for a thin veneer of analysis that rests on top of a huge helping of plain old reporting.
Which brings us to the topic of predictions. Is your analysis actually predictive? Does it recommend or indicate a set of options that will improve future financial outcomes at a lower cost and risk than what is being done today? This is a statistical analysis question, not one about reporting past outcomes.
The most mature, publicly visible examples of predictive analytics in insurance are the services offered by Lexis/Nexis, Fair Isaac, reinsurers and others who create a score or provide other indicators of future experience that have been borne out via back-testing, mortality studies or other means. In the case of reinsurers, they often attach the willingness to have risks ceded to them in real-time based upon their use of predictive analytic techniques and the use of electronic data concerning the risk that isn’t gathered at the point of sale.
It goes without saying that large insurers are making similar investments, but these innovative developers of predictive analytics are cagy about describing their competitive advantage in any more detail than what is absolutely necessary for marketing purposes—or to recruit top quality data scientists who are capable of devising ever more detailed pattern analysis of ever larger data sets.
How then, do you ensure that your work with data is both predictive and analytical? It starts by asking having five basic questions:
In our experience, many insurers are engaged in either consuming or producing predictive analytics in an attempt to produce better returns. These insurers have a strong point of view about build-versus-buy and have addressed the questions introduced in this article. The only question that remains is whether it is working. The answer is, they aren’t saying.
(Russ Bostick is managing partner of the MVP Advisory Group)
Does your firm have access to a statistically significant, reasonably accurate set of data about the risk, the process or the prices that you want to predict? Can that data be refreshed frequently enough to be relevant? Simply having the coded data from the claims experience of your current policyholders may not yield much.
Do you have business leaders who have hypotheses about where the business could improve its performance? Simply looking at correlations won’t derive a conclusion. Predictive analytics initiatives are much more valuable and conclusive when they start with a hypothesis and then iterate new hypotheses. In short, those leaders have to realize that every answer begets a new question for quite some time. The data reveals nothing.
Do you have the brand, culture, and capability to build and retain a staff of data scientists? Or should you focus instead on buying scores and other insight that provides a great deal of the value at a predictable, risk-free cost? The classic build-versus-buy decision is just as present in predictive analytics as it is in other aspects of information technology.
Are you willing to invest in creating a virtuous cycle where future investments are paid for by the gains from prior initiatives? It’s true that storage, processing and tools have never been cheaper, but it’s equally true that the benefits take time to materialize and there are significant upfront people-costs to create a predictive analytics capability in house.
Lastly, are you willing to move from the lab to the field and actually put into practice the products, pricing, and risk selection that the predictive analytics team indicates are profitable? Or is this just another IT science project that checks a box on being innovative?