Big Data and Predictive Analytics
If you check out my IRMI.com bio or LinkedIn bio, you'll see I am "experienced." I got my start in the late 70s. My first big promotion was from a desk underwriter to the home office of the same insurer, elevating to the position of marketing analyst and reporting to the vice president of Marketing.
The timing of the position was perfect for me, as I had spent the last year learning coverages, rating, customer service, and workflow operations. It was time to learn about the big picture.
Among other things, my job entailed competitive analysis. I inherited very large pieces of paper called an "analysis pad."
So the job was to, with pencil and calculator in hand, list a set of fictitious risks, one in each row, and then list the competitors along the top of each column. We did a page for each territory. I literally rated each of the risks for each of the competitors. This required obtaining their manuals from an agent, learning their rate order calculation, and doing the math for each row. Of course, since our major state was California, we didn't attempt to complete a page for every territory, but we did many. We also targeted certain competitors.
The company I worked for wrote nonstandard auto insurance, which is extremely price driven, so this was an important job. The underwriting vice president was in charge of pricing, but the marketing vice president lived and died by convincing agents to sell new business policies. He had to be armed for those discussions as well as have evidence for his boss and peers about the competitive position of our rates.
As you can imagine, the information contained in the small stacks of oversized paper is amazing! If you had good intuition and could follow a thread of thinking through several flips of the page, note-taking, and patience, you could make loads of observations with this data.
The troubling part is it took forever to sift the nuggets of info from this laborious project. However, we had no choice if this was our desire. We knew our profitability was dependent on new business since this line doesn't retain very well. Yet our rates had to be adequate, too, making this a multidimensional effort.
We needed to be competitive on all types of risks. Lowest? Not always, but at least in the "ballpark" so our agents would consider placing the risks with us. We leveraged our salespeople, our good service, and our financial stature to sell the agents on why they should, all things being equal, send the applications our way.
Looking back, I can see the value of what we had accomplished, but I can also see many parts of what this data was telling us because we didn't have a way to do the math and tabulate results. The heavy-lifting methodology we were stuck with left many of the following shortcomings.
- We chose risks that we thought were representative of what we thought we wanted to write. We had no information on what the available pool of applicants was.
- We used the same risk profiles across territories. Not only did this approach limit our knowledge, but it also made it difficult to show agents and our salespeople risk profiles where they were likely to find us competitive.
- We didn't know what insurers were writing in certain territories, other than there was a printed rate in their manual. They may have had no agents in some of the territories and many in other territories.
- We couldn't compare our actual book of business with the sample risks available.
- Most challenging of all, it would take hours to search through the pages of information to do calculations and make inferences based on iterative thinking.
More Than Predictive Analytics
Analytics have been around ever since the abacus. Our tools have gained capability and efficiency as computing power and data have been more accessible. All the buzz in the last 10 years has been about predictive analytics. You cannot pick up an industry trade magazine or read a blog without seeing its mention.
Companies are inundated with calls from salespeople, webinars, and conference exhibits on the topic. Actuaries or other quantitatively trained "data scientists" can write their own ticket if they are good at wielding software to develop models that help them make decisions based on past history. Our business has gotten more complicated with the need to have multiple perspectives when one look cannot do all the heavy lifting.
Predictive's Not as Popular Siblings: Prescriptive and Descriptive
Consider the three members of the analytics family.
- Predictive analytics uses statistical models and forecast techniques to understand the future and answer: "What could happen?"
- Descriptive analytics uses data aggregation and data mining to provide insight into the environment and answer: "What is going on?"
- Prescriptive analytics uses optimization and simulation algorithms to advise on possible outcomes and answer: "What should we do?"1
Analytics, in general, is a more comprehensive mantra for the true data-driven organization. Each one of these tools has its own sweet spot of value to the savvy user. What we were doing in these early days, before regression was chic (or even doable with paper and calculator), would be called descriptive analysis. We felt we were data-driven, and we were, but our capabilities were limited.
Fast-Forward to Modern Descriptive Analytics
Today's tools provide for complete data visualization of a company's competitive status. Competitor companies' pricing is available to be imported into software customized for competitive analysis. Agent files can be imported. Quote/bind history can be imported. Many states have in-force policy information that can be uploaded.
There is no need to rely on hand calculations, guesswork, and anecdotal evidence anymore. With the right data, analytical tools, and a few clicks of the mouse, you should be able to answer the following questions that will drive your growth.
- Will our proposed premiums hit our target bind rates?
- How do we support our competitively-based rate selections in filings?
- Are we leaving money on the table?
- What risk profiles should we advise our agents to quote and not quote for us?
- Are we getting selected against in our most profitable region?
- How much difference is there between company X premiums and ours?
- How do we compare our pricing with other companies when we have very little data to go on?
- How do I show my agents that we are competitive and where we are competitive in their area?
- How can I know where to position my rates when entering a new market?
- How can I show agents where we are not competitive to help them in their marketing?
- How can I direct our sales team as to where it will be most fruitful to appoint agents or to make agency visits?
- How can I determine what characteristics of a risk make us more or less competitive over different geographies?
- What rating factor is making my prices either too low or out of the market where I won't get quoted?
- How can I tell if an agent is selling my quotes where we are competitive?
- How can I simultaneously see our competitive posture versus a group of competitors?
The Pot of Gold
Some of this information was available from our stacks of analysis pads. But it would have taken endless hours of iteration to generate findings to such detail. And these findings based on our selection of sample risks were much less applicable than if they had been based on a representative portfolio of risks, considering the business our company had quoted and either sold or not sold.
Beyond knowing the answers, today's data and tools provide the opportunity to collaborate with actuaries, product managers, underwriters, and marketers. Profitability and growth measures can be balanced. Since everyone has access to the same information, better decisions can be made, and consensus can allow for shared goals, team execution, and winning results.