InsurTech: Predictive Analytics

Durgesh Vyas, Pricing Manager – Casualty, shares his thoughts on why he believes there has been a shift away from predictive analytics. 

Several years ago, the InsurTech market was firmly behind predictive pricing as the solution to reducing loss ratios. Yet, we hear less about this as a key benefit, leaving us to question if the performance in this area has not been producing the expected results. 

A recent report, published by Swiss Re Institute, suggests that industry executives cautioned against expecting large quantitative benefits in the near term, especially concerning improvements in loss ratio. This reality is likely even starker for commercial insurers who typically face more underwriting complexities and a more heterogeneous client base.  

At this stage, it’s worth pointing out that even the most basic rating model is a form of predictive analyticseven when only connected to static rating factors. Rating models, by their nature, are trying to predict or say something about an expected loss cost for a given risk. We must always be mindful of the limitations, in particular, the phenomenon known as ‘Winner’s Curse’ 

Every model will contain some inaccuracy. It stands to reason, that predictive models involved in the risk selection process will have some bias towards risk with lower expected loss ratios. The overall bound portfolio will, therefore, be biased because it is likely the model has under-estimated the expected loss costs for each risk, rather than over-estimated it. 

At Probitas 1492, analytics and data is always front and centre to the business. Our analytics team work directly with the underwriting teams allowing us to understand our data better and react to our pricing models accordingly identifying profitable business, acknowledging competitive pricing, highlighting where adjustments may need to be made and, of course, reporting back on performance and trends. 

There is an opportunity to be reactive when it comes to identifying patterns, such as loss ratios, early in the process. Dissecting customer behaviour to gain insight, which enables us to provide a tailored service, to add value to the insured and, in turn, allows us to more disciplined with our underwriting. Insight from our rating models and data allows us to assess risks with more certainty, they complement the underwriters own assessment of the business that is placed in front of them. 

In simple terms, applications of analytics fall under four areas:  

  1. Enabling growth 
  2. Engaging with customers  
  3. Optimising portfolios  
  4. Improving efficiency

Enabling profitable growth and optimising our portfolios is obviously key to any syndicate. We set out our stall as a business from day one, to capture and record as much risk information (bound, quote or submission) in our rating models. We recognised that the data itself would allow us to optimise our portfolios, engage with customers and enable growth, be that now or in the future.  

We’ve also focused on being streamlined and efficient. Recognising the benefits of being nimble and not constrained by legacy systems, we are able to adapt and pivot, make changes relatively quickly and see the impact immediatelyon our business and, importantly, for brokers and the insured.  

Most incumbent insurers struggle to piece together historic exposures and claims. The simple approach of using tried and tested methods can yield significant insights into our portfolios. But, this isn’t a blackbox approach. Using data and methods which the business is comfortable with, by which I mean, there is buyin from underwriters, encourages a collaborative approach to make changes and evolve our models. 

As a result of being clear of our intentions from day one, we get a bigger bang for our buck. Our data is cleaner and, importantly holds on average more than 5 years of exposure and claims related detail.  

It’s why we know that the value in predictive pricing is so much more than just a reduction in loss ratios.”