For the last presentation today, I listened to John Lucker of Deloitte discuss what they’ve developed in the area of predictive pricing models for property and casualty insurance. Pricing insurance is a bit trickier than pricing widgets: it’s more than just cost of goods sold plus a profit factor, there’s also the risk factor, and calculating these risks and how they affect pricing is what actuaries do for a living. However, using predictive models can make this pricing more accurate and more consistent, and therefore provides insurance companies with a way to be more competitive and more profitable at the same time.
I know pretty much nothing about predictive modeling, although I think that the algorithms are related to the pattern recognition and clustering stuff that I used to do back in grad school. There’s a ton of recent books on analytics, ranging from pop culture ones like Freakonomics to the somewhat more scholarly Competing on Analytics. I’m expecting Analytics for Dummies to come out any time now.
Predictive modeling is used heavily in credit scoring — based on your current assets, spending habits and debt load, how likely are you to pay on time — and in the personal insurance business, but it hasn’t really hit the commercial insurance market yet. However, the insurance industry recognizes that this is the future, and all the big players are at least dabbling in it. Although a lot of them have previously considered this in order to just do more consistent pricing, what they’re trying to do now is have the predictive models integrate together with business rules in order to drive results. This is helping to reduce the number of lost customers (by providing more competitive pricing), reducing expenses (by providing straight-through processing), increasing growth (by targeting new business areas), and profitability (by providing more accurate pricing).
He talked about how the nature of targeting insurance products is moving towards micro-segmentation, such as finding the 18-year-old male drivers who aren’t bad drivers or the roofing companies with low accident rates, then selling to them at a better price than most insurance companies would offer to a broader segment, such as all 18-year-old male drivers or all roofers. He didn’t use the words long tail, but that’s what he’s talking about: this is the long tail of insurance underwriting. There’s so much data about everything that we do these days, both personal and business, that it’s possible to do that sort of micro-segmentation by gathering up all that data, applying some predictive modeling to extract many more parameters of the data than would have been done in a manual evaluation, and develop the loss predictive model that allows a company to figure out whether you’re a good risk or not, and what price to charge you in order to mitigate that risk. Violation of privacy? Maybe. Good insurance business? Definitely.
The result of all this is a segmented view of the market that allows a company to decide which parts they want to focus on, and how to price any of those parts. Now it gets really interesting, because now these models can be fed into the business rules in order to determine the price for any given policy: a non-negotiable price, much like Saturn does with its cars. This disintermediates both the agents and the underwriters in the sales process, since all of the decisions about what risks to accept and how to price the policies is automated based on the predictive models and the business rules. Rules can even be made self-optimizing based on emerging trends in the data, which I discussed in my presentation this morning, although this practice is not yet mainstream.
Lucker’s message is that business rules are what leverages the power of the predictive models into something that makes a difference for a business, namely, improving business processes: reducing manual processes and associated costs, enhancing service and delivery channels, targeting sales on profitable niches (that long tail), and improving point-of-sale decision-making at an agency.
He ended up describing a top-down approach for designing business rules, starting with organizational strategy, decomposing to the functional areas (business operations, sales, customer service, distribution), then developing the business rules required to help meet the objectives of each of the areas.
