Why Historical Insurance Data Models Don’t Work in our Current Environment

Classic data modeling and history-based actuarial models do not comprehensively work anymore. In order to get useful insights that can be implemented to support customers and the business, insurers must rapidly incorporate new data sources in their analysis. Current events that are different than anything we have seen in recent history force us to get used to a new world that cannot totally be evaluated and analyzed based on historical experience and knowledge. Unfortunately, this makes the actuarial models that insurers have run their business on for decades less useful. A new world requires new data, and new ways to gain insights.

How historical data lead to incorrect predictions

History and models concluded that car insurers would pay out fewer claims during the pandemic since most people are not using their cars as much as before, or at all. Based on those historical insights, car insurers in the USA, and to a lesser degree in other parts of the world refunded their customers 15-20% of their premiums paid over a period of 2-3 months for now (Forbes).  However, the actual experience started showing that, although claims volume might be down, claims severity has gone up. Insurers are paying out more per claim than expected based on relying on those historical models. Initial assumption-based analyses and anecdotal insights show that this might be caused by excessive speeding (much easier on empty roads), less strict law enforcement (cities want to avoid their police officers coming in close contact with citizens), and/or increased alcohol and drugs use during this pandemic. None of these triggers were modeled when deciding to refund premiums. 

Another example includes wearing masks. There is anecdotal information that mask-wearing increases attempted robbery in stores and banks as reported by CBS Los Angeles in May. I am sure that in underwriting the property policies for these businesses, nobody accounted for the fact that everybody who enters the facility would be covering their face. Apple, Google, etc. have had to change their algorithms for facial recognition to accommodate the need for identifying a person with a mask. So even new world companies and new data sources need to rapidly evolve.

Big areas of uncertainty can be found in all sorts of liability -from product liability on PPE, employment practices liability on who can go to their workplace, when and under what conditions, to questions on who and what to blame if I, or one of my loved ones,  get infected after we start opening. There are big coverage questions around Business Interruption insurance, and to a lesser degree travel insurance; also, the underwriting processes in Life insurance are rapidly changing.

Insurers must incorporate new Data Sources

These examples clearly demonstrate that classic data modeling and history-based actuarial exercises do not comprehensively work anymore. In order to get useful insights that can be implemented to support customers and the business, insurers must rapidly incorporate new data sources in their analysis and support all the formats these sources come in: text, voice, pictures, videos, streaming, IoT, social media, structured and unstructured.  This is not an easy task. It requires a business strategy that identifies and supports the need for all of this information, a data strategy that defines, prioritizes, governs, and executes on the data requirements, a cloud strategy to figure out how to store and compute the data, and a use case roadmap. This must all be supported by an enterprise data platform that enables business users to access and analyze data when, where, and how they need. 

Challenging as these past few months have been for all of us (and please continue to stay safe and healthy), it is inspiring to see so many of our insurers step up and truly try to gather and analyze new relevant data to support their customers in these complicated times.  I am thankful for the opportunity to support some of them with use cases, market insights, and roadmaps. Since this is one of these times when we can only win together. With data.

Join this webinar series to hear updated Insurance use cases for how you can use big data to make a big impact on your business. You can also learn more about Cloudera Data Platform and how it can support your Enterprise Data Cloud needs.  

Monique Hesseling
Monique Hesseling

2 Comments

by Marty Ellingsworth on

I think there is a lot of biased data loaded into the points made above on severity especially on short time to close total losses with fatalities – normally those tend to be longer tailed, so Covid period estimates are small and exaggerated

by monique hesseling on

Hi Marty, thank you for your comment! I agree with you that accidents with fatalities are long tail claims. In auto, some claims data show a recent uptick in physical damage and total losses, as well as in theft . For example the Philadelphia Inquirer  posted an article today by AP’s Stefanie Dazio quoting that that Year over Year so far this year, NY saw 63% more vehicle larcenies, and Los Angeles saw an increase of 17%.  I used these insights as examples for the need to include new data in predictive analytics.  Now, I understand that this example -as well as the one about mask wearing- might be anecdotal, but my point was that we need to include data points we never thought to include before  in our analytics to make better sense of our new world.

Leave a comment

Your email address will not be published. Links are not permitted in comments.