The pragmatic applications of AI in the Insurance Industry
By Nick Henthorn
This article was published by Modern Insurance Magazine on Dec 12 2018.
Increasingly artificial intelligence and deep learning are hitting the headlines. Recently it was reported that IBM has created a secret algorithm that develops scents.
The perfume industry famed for its highly paid human ‘noses’ is about to be disrupted as the technology giant developed a system that studies more than 1.7 million existing fragrance formulas (from perfumes to scents used for toothpaste, candles, shower gels, face creams etc.) and then compares the ingredients to other data sets, like geography and customer age. This algorithm will be used to develop new perfumes that will target very specific market segments. For instance if asked for a fragrance that would appeal to millennials living in Brazil the algorithm would compare formulas against scents that were popular in that area and age range. One of the perfumes it created had a fruity, floral smell with scents of Osmanthus tea, lychee and patchouli. When tested the fragrance received positive feedback from focus groups, and came out on top even when tested against other perfumes popular with Brazilian millennials.
Another exciting application of AI from Georgia Institute of Technology is the creation of a machine learning system that has taught itself video game design by ‘watching’ footage of people playing classic games such as Super Mario and Kirby. By working out the components of what makes a good game the system can create complex programmes to significantly enhance games created by human programmers.
So what has all this got to do with the insurance industry?
Quite a lot actually! AI is becoming increasingly mainstream and is no longer the domain of boffins in computer labs. Data scientists are now common place in organisations or as a part of an outsourced marketing team. And unsurprisingly given the depth of data generated by the insurance sector data scientists have turned their attention to creating AI-enhanced models that are reshaping traditional claims, distribution, underwriting and pricing models. It’s no surprise therefore that The Society of Insurance Research has identified data science and AI as the two most important disciplines in terms of securing competitive advantage. As a result there are plenty of articles to be found pontificating on the future of the insurance industry and AI – outlining what is likely to be possible, but it is much harder to find more pragmatic literature explaining what is possible using machine learning right now.
The key lies in understanding more about the customer. By doing this insurers can use AI to better target them at the moment they are ready to buy with the right premium, aid the application process and enhance the relationship over the course of the policy.
To gain a comprehensive view of potential risk, insurers need to be able to rely on accurate data that can be applied to every decision. Through AI key variables can be generated to support insurers in identifying potential risks and therefore optimise pricing models. For home insurance for example these might be:
- Water escape: Developing models to accurately understand the number of bathrooms and any building works undertaken on the property that are all key factors of risk
- Size: Gaining an accurate view of the number of bedrooms and square footage of the property
- Home Rebuild Cost: creating a model that predicts the rebuild cost for houses across Great Britain including confidence scores to enable autonomy in creating thresholds around the pricing policy
- Roof Construction Type: Using deep learning with satellite imagery to categorise roof types (flat, pitched or thatched).
Research shows that 55 million insurance policy purchases are abandoned each year during the application process. This is because it takes time and effort for consumers to fill in lengthy insurance application forms. By pre-populating online applications using an API makes it quicker and easier for prospects to complete forms and means the information is more accurate as the data is based on fully verified data rather than customer hearsay. It is proven that customers can sometimes ‘stretch’ the truth about their home or driving behaviour if they believe it will lower their premium. Common white lies might include claiming a tall tree is further away from the home than it actually is or estimating down annual mileage.
Through sophisticated data management techniques it is possible to fill in the application forms on behalf of the customer. Car insurance applications can do this by using data from DVLA whilst home insurers can use household data to pre-populate questions such as number of bedrooms and bathrooms, property type, age of property, and rebuild costs.
All the customer has to do is check that the information in the form is correct, taking seconds, rather than minutes to complete, significantly reducing industry churn.
In addition, these datasets can also be used to run marketing campaigns with a fully underwritten home insurance quote, as opposed to an estimate. This in turn, will drive a stronger call to action, allowing the consumer to simply enter a unique code on an insurers’ website and go straight to payment.
Finally, using the latest data science technique it is possible to create more accurate and predictive customer segmentations, or layer multiple segmentations together at lower cost and greater speed to align business objectives. By building insights about customers it is possible to use AI understand why customers behave as they do and determine what motivates them to act.
Data can be monitored and interpreted over time to use change as a predictive variable – for instance children coming of driving age – in order to interpret trends and predict future customer needs. This means that AI can predict when individual customers are in the market for a specific product, know what type of communication will result in a purchase and understand how to engage then for future policy purchases. Again all of this must be underpinned by confidence scores so that insurers can be confident in the models they are using to set prices, identify risk and communicate with their customers.
AI is, without a doubt, a future forward technology. Seemingly on a monthly basis new developments are reported. However, what is important to remember is not what might be possible next year, in five years, or in 10 years but how AI can be applied to the business right now or risk being left behind!