Insurers attempting to implement analytics programs face a variety of challenges beyond the technical, observed Ted Vandenberg, director of claims analytics and business intelligence for Farmers Insurance, at the 2013 Insurance Analytics Symposium in Boston last week. Not the least of which is that they simply don’t fit neatly into the top-down corporate structure that is so common at large insurers, he said.
“You may have more knowledge than anyone else in your organization and certainly the executives in the operational world,” Vandenberg said. “You are without peers in a big space. There are few people who understand the mathematics and even fewer who can manage the people who understand the mathematics. There’s a small mix of people turning this into a corporate function.”
Organizing quants, or data scientists, into hierarchical organizations will be a challenge, Vandenberg said, because of the few people who do the job, most are motivated by curiosity and the pursuit of science; fewer are motivated by same things that motivate C-level leaders: cost savings.
To illustrate the point, Vandenberg told the story of coming to that realization years ago, when, as a consultant, he tried to recruit an executive speaker for a conference of quants.
“The sponsor and I approached a chief claims officer and said Here’s a great opportunity. You are going to be speaking to a room of 200 quants at different organizations, and you can tell your story, talk about quantitative science and how important it is.’”
The chief claims officer’s response was less than enthusiastic. “The first time you get a senior executive in that room, they are going to be thinking, Who’s doing this job? Who’s managing it? And maybe we could do it for less,” Vandenberg said, quoting the CCO. “The DNA of an executive is to do things cheaper. Executives are not in the business of making money; we are a cost cut-out. The thought process there: We have to do this effectively and efficiently even though it is hugely valuable.”
In order for analytics to become a part of the corporate structure, it will need to be led by someone with quantitative knowledge, Vandenberg said, who understands how the work gets done and also appreciates the inherent risks.
“Information technology started out as payroll systems; 30 years later it’s an enterprise function,” Vandenberg said. “Pretty quickly, analytics will be in that same realm as IT. It deserves to be. Who would have thought IT was a C-level function back when it was a payroll system? Now IT has formalized processes and knows how to not waste money. That structure is needed in analytics. We need it, we don’t have it.”
Until the issue of integration into the corporate structure is addressed, quants will be powerless, Vandenberg said. Even those models that have been validated, he said, will have difficulty being implemented and put into production because data scientists are not currently part of an enterprise function.
“Whether [business people] use that model or not is up to them,” Vandenberg said. “Do [data scientists] really have VP power? Probably not. [Business people] run the business and whether they want to use the model or not is up to them. How can you be responsible for results it you can’t operationalize that model?”
Analytics face a variety of other hurdles in getting models operationalized. As they routinize decision making, analytics models potentially are job killers, Vandenberg said, and they are inherently risky as long as they are not more widely understood and auditable.
“These analytics efforts are risky and therefore must be organized,” Vandenberg said. “If it doesn’t have controls around it, it’s automatically disorganized.”
There are also ethical considerations, as evidenced by the difficulty involved in the mainstreaming of credit scores for determining rates.
“The data scientist has to be an ethical person. The models can be wrong, and they can be wrong in subtle ways,” Vandenberg said. “They can under sample the data; they can sample it in weird ways. It could be an accident, it could be on purpose. It could be a contractor without your sense of ethics. You need ethical training for your data scientists because who will save us if the model goes wrong? There are very few people who can do that.”
Compounding that issue and the potentially resulting difficulties are the facts that executives don’t know what they don’t know, and that many of the new data scientists will be fresh out of school, not know the insurance industry and may not recognize the limitations of the models they create.
“There is risk here. Anytime there is risk without controls, it’s disorganized,” Vandenberg said. “Who in audit is capable of understanding these models and saying the model is wrong? In an enterprise corporate environment, you would formalize the risk environment. You would have auditors. In banking, they do,” Vandenberg said. “If some of this stuff starts hitting consumers, like credit scores did, [the industry] will be in a world of hurt. And that would be a shame if we don’t control for that and recognize who owns the risk.”
The person who eventually leads the quants is likely to be a business person who learns quant, Vandenberg said, and not a quant, as they are more interested in the science. Nor is that person likely to be an IT person or a finance person.
Related content: 5 Secrets to Successful Analytics.
Register or login for access to this item and much more
All Digital Insurance content is archived after seven days.
Community members receive:
- All recent and archived articles
- Conference offers and updates
- A full menu of enewsletter options
- Web seminars, white papers, ebooks
Already have an account? Log In
Don't have an account? Register for Free Unlimited Access