Below is a response to my blog, “Predictive Modeling Is Here to Stay” in a discussion published on the Linked In group, Workers Compensation Roundtable last week.)
Responder 1: Predictive modeling is particularly useful on smaller policies. An account with millions of dollars in payroll and high premiums generally gets more attention so the quality of information is higher. Given the size, the history is much more predictive of future results. On the smaller accounts, you don’t typically have as high quality information and a lack of data on which to make sound assumptions. You have to underwrite the book more than the individual account. Predictive modeling allows a carrier to do that fairly, as you point out, by find those metrics that are real drivers or results.
Responder 2: Predictive Modeling clearly is being effectively engaged to improve the WC underwriting process. There is also a growing initiative to apply predictive models in the post claim world, using predicted loss costs from FROI data to create triage protocols, severity specific claim handling strategies, managed care referral triggers, and other resource allocation decisions. Finding the Pareto 20% quickly and deploying high case cost mitigation strategies can be achieved with predictive modeling tools. This is an equally important initiative for employers to engage with carriers and vendors managing their WC programs.
Me: Agreed! I will be covering how SI employers can use predictive modeling in a future post.
Responder 3: Annmarie, thanks for your posting. I followed through on your link, then to the other links…the article in Contingencies was hard to read online and I would very much like to see it in an easier to read format. It is my impression that predictive modeling has in effect been in workers comp for decades. What IT has done in the past 5 – 10 years, with more power, more accurate data, and far easier integration of disparate data sources, is to given real meaning to the word “model.” As the phrase goes, “all models are wrong; some are useful.” That is, one has to have the mind set to persistent in trying out different ways of looking into the future, always be willing (and able) to experiment. And this brings up for me the great barrier to predictive modeling in workers comp: the resistance of some claims and underwriting executives to challenge their own ways of predicting, which are crafted over many years and based on very flawed perceptions of reality. They thrive on anecdotes that they will not give up on.
Me: Part of predictive modeling is using new non-insurance-related data sets. This is definitely new. Barriers are being overcome. They must for insurers to stay competitive. One workers’ compensation insurer wrote me after my post asking how small-to-medium-sized carriers are handling predictive modeling when data is limited. More insurance-related data is becoming available. More workers’ compensation carriers are adopting predictive modeling and I will be posting something on this by the end of the week. Sorry about the Contingencies article. I like to refer people to the actual publication’s posting. If you go under my “work samples” tab on my blog, you will also find it there. To save time, I am giving you the link to it below. It is a paper copy I had scanned in. Here’s the link: https://annmariecommunicatesinsurance.files.wordpress.com/2012/01/workerscomppredmod-contingencies2.pdf You are correct that computer technology is also making predictive modeling more possible. As I am covering in an article yet-to-be published, more carriers are installing new systems that support predictive modeling. Thanks for reading and for your comments! Stay warm!
Responder 4: (Responder 3) is right The “dinosaur” mentality based on years ( actually decades…) of “gut feeling” still prevails out there. It is in claims, sales, loss control, risk management, etc. Part of the problem is that people inherently resist “change”. When “the model” does not agree with the “historical perception” then “the model” must be wrong, because the person making that “right / wrong” decision has made those type of decisions based on a “gut feeling” (i.e. – experience) for the past 40 years….. When the data CONSISTENTLY swings in favor of “modeling”, then the “gut feeling” as the “primary” critera has to be “repositioned” as a secondary consideration, rather than the primary consideration. In time, as models improve, maybe 95% of WC claims can be successfully “modeled”. The remaining 5% are going to have to be reviewed by a person that determines that the model does not apply to this SPECIFIC claim and maybe a “gut feeling” decision is more appropriate, in this single case. At the end of the day, every claim is unique. WC modeling has come a long way in a short time, and like everything, will continue to improve as more valid data is compiled and analyized. You just have to convince “the boss” that the data is more accurate than 40 years of “gut feelings”……
Responder 3: Annmarie, thanks for posting the article in a readable format. It is the best discussion I have ever read on predictive modeling in workers comp. To be sure, you are light on the resistance to modeling such as (one responder) cogently points out. The first serious model I saw as introduced into a workers comp insurer about 2000 and was abandoned eventually because adjusters felt threatened. But your article is a gift to the community. My abiding interest in modeling took me to nutty extremes last September when I visited the grave site of the poet, Rilke, whose poems of 1900-1925 were, I feel, an aesthetic interpretation of modeling, in his case, how you change your whole life, time and again.
Me: Thanks for the compliment on my article. Just want to address (Responder 3) and (Responder 4). For underwriting purposes, predictive modeling often quantifies what underwriters knew with their gut feeling and that alone is important. But it goes far beyond that. Workers’ compensation predictive modeling got started around 12 years ago by the largest wc insurers and applications keep growing. I do not agree that the “dinosaur” mentality is holding back insurers, or not at least like it once did. It can’t. There is just too much profit to be made, not to mention all the market segmenting opportunities insurers can use to fine-tune their competitive strategies.
The bandwagon is filling up and nobody wants to be left behind. Predictive modeling, as it is now understood, started with Progressive Insurance around 30 years ago. It forever changed how auto, and later other personal lines, would be underwritten by the entire industry. It will not take that long for the same to happen with workers’ compensation predictive modeling. Commercial lines actuaries and statisticians have learned a lot from the personal lines experience and there are more tools available. Automation and the latest computer systems support predictive modeling much better than those of the past. Some companies are developing computer systems with predictive modeling in mind. The article I posted does cover barriers to predictive modeling, but they are quickly being overcome thanks to good old fashioned profit motive. Workers’ compensation insurance will experience the Progressive Insurance phenomenon. If employers did not have a good enough financial argument for following all the great advice to reduce losses, they do now.
Responder 5: Excellent article. Re: your comment “predictive modeling often quantifies what underwriters knew with their gut feeling and that alone is important” – don’t underestimate the importance of this. Underwriters are trained in the Joe Friday, “just the facts” school and are hesitant to make a decision without something in the file. On the claims side, this is also true. I was involved in putting together a very simple model which showed UR denial rates or treating physicians – a simple provider scorecard. When the examiners saw this, they were delighted to have real evidence that Dr. X was a problem.
Me: Thanks (Responder 5). I have been keeping my focus on predictive modeling as an underwriting/rating application because this is the most common and immediate way employers will notice a difference in premiums based on their experience. Claims predictive modeling is also picking up some speed and we can learn a lot from group health care in this regard. I have more articles coming out on predictive modeling and when they are published, I will share! Appreciate all the comments everyone!
(Note: Permission was granted by one of the group’s managers to publish this discussion with the names of commenters removed.)