Solving common insurance issues such as customer churn and managing risk through the application of AI, can provide more accurate and predictive outcomes.

Ai and insuranceInsurance has long been a sector which has appreciated the value of its data. No wonder, since in comparison to many industries, it is data rich. Insurers’ databases are full of demographic, lifestyle, credit-worthiness, health, property, wealth and automotive-related information. As a result, insurers were among the first to adopt techniques such as database marketing, CRM and customer segmentation as a way of reducing churn, improving customer acquisition and driving cross sell opportunities. And given the challenging business environment such as the push from government and pro-consumer organisations to encourage switching, the rise of price comparison sites and growing numbers of new entrants to the market, this became essential to achieve competitive advantage.

But that’s where it has stayed. This early data innovation stagnated and today many of the models powered by insurance data like Home Rebuild Cost are still being used despite the methodology being years old. This means they are not as predictive as they could and should be.

The volume of insurance data has in some ways acted as an anticatalyst, lulling the industry into a false sense of security. As a result, unstructured and third party data such as customer service interactions, website clicks, mortgage application data etc. have been largely ignored due to the fact it is hard to collect, store and analyse – besides the sector didn’t need it because of the wealth of its own data – right? No, not really. This new data can be enormously powerful when combined with its structured counterpart, and with the advent of AI, it is now possible to more easily utilise this data and enhance existing methodologies to make them more effective.

Building customised models with AI

An example of how data science can be used to enhance, or indeed disrupt established ways of doing things is by building new, customised models to replace established methodologies. Going back to the estimation of Home Rebuild Cost, the original foundation for such models is BCIS rates but these are typically only suitable for ‘standard’ properties. As soon as variables like listed buildings, high value postcodes, unique or large homes come into play the predictions are widely inaccurate. As a result, broker valuations for these properties tend to follow market value or historic guessed value. Yet by applying deep learning to a wealth of property data, such as build date, number of bedrooms, location, etc. it’s possible to build a bespoke model that accounts for these non-standard properties and delivers more accurate rebuild costs, thus reducing the risk for the insurer.

More predictive and accurate Home Move Triggers

Up until now Home Move Triggers have been simplistic models built around when a property is listed for sale, but this approach does not provide the chance to identify and engage with existing and prospective customers at the ideal moment in the moving process. It merely gives an indication that the resident ‘might’ be on the move. By applying AI and data science and supplementing the listing trigger with historical data, location-based data and other sources of information relevant to the property, we can build models that far outperform human rule-based logic, in terms of precision and reactivity. Moreover, by providing Confidence Scores to the modelled move date, it allows businesses to tailor the use of data to their needs and quantify the impact of the prediction.

This means if a house takes a long time to sell you are not communicating with the customers too early, or if it sells quickly, you are not too late. The wealth of variables also means that insurers have a very granular view of each potential home mover and can increase the relevance of their communications as well as sending at the right moment in the customer journey.  

The importance of applying Confidence Scores to modelled data

The issue with AI is that its development has an inverse relationship with confidence. As models become increasingly complex, understanding how they arrive at their conclusions becomes harder. As a result there are some organisations that prefer to stay with the tried and tested old ways, which might deliver less accuracy and prediction power, but are easier to comprehend and explain. That’s why Confidence Scores are becoming increasingly important – particularly in the insurance industry, which is all about minimising risk! To aid decision making and improve trust in these new methodologies Confidence Scores can be attached to all modelled data so that the reliability of the derived information can be quickly and easily established. Ultimately a modelled attribute without a Confidence Score is just as a bad as a guess.

AI ultimately leads to better decision making

Through the application of AI, it is therefore possible for the insurance industry to once again become data trailblazers and find more predictive, faster and cost-effective ways to solve business problems. If data is the lifeblood of the industry, data science should become its beating heart.