Could AI Change How Insurers See Risk?

This article is by Oli Williamson, Co-Founder and Chief Underwriting Officer – Aspect

Underwriters have always had to make decisions with incomplete information, drawing on their experience to assess everything from property surveys and risk reports to broker commentary. This information generally comes in the form of large volumes of documentation that over time become a complete book, and the basis for detailed risk assessment.

Increasingly however, AI can go further and allow insurers to read between the lines of this documentation, to uncover previously hidden risk indicators that could improve the accuracy of insurance pricing.

The idea of predicting future events might sound like something of fiction – the film Minority Report springs to mind – but AI is bringing the concept a little closer to reality for insurers.

My own team at Aspect recently teamed up with forensic linguistics and AI specialists Fluence World to use AI to test the predictive power of language. Needless to say, the results were eye-opening.

We analysed the broker’s submission pack alone – with no prior knowledge fed into the algorithm – across more than 40,000 files, covering 3,892 commercial property accounts, with a total premium of around £85.6m. We wanted to see whether the language used in insurance submissions contained patterns associated with claims, patterns which even the best underwriters may have been unaware of. The tool was tested without access to our pricing models or underwriting decisions, allowing us to look at what the submissions themselves might reveal.

The results were striking – the tool flagged eight of the ten largest losses, including the largest claim in the book of £4.3m. The half of the portfolio it scored more favourably had a developed loss ratio of 10.5%, compared with 48.1% for the half it scored less favourably. In other words, the two groups performed very differently when it came to claims relative to premiums.

Looking beyond the price

While two businesses might look similar on paper and pay similar premiums, one could present a far different risk profile than the other. The information already sitting in an insurance submission could help point to why.

The premium might reflect an insurer’s assessment of a risk, but it may not tell the whole story about how likely that business is to make a claim. Technology could fill this gap by helping insurers spot risks that deserve a closer look. Rather than just highlighting where they might need to pull back, it could help make better decisions about what risk management changes a client needs to make or highlight where clients are needlessly paying over the odds with another insurer.

It’s important to recognise that AI won’t replace an underwriter’s judgement and it’s important that it doesn’t lead us to a ‘computer says no’ approach to insurance, and risk widening the UK’s already significant insurance gap. The research shows however that technology can give us new means of assessing risk, questioning assumptions and picking up on details they might otherwise overlook.

What could this mean for businesses struggling to find cover?

This could be particularly relevant in specialist insurance, where businesses can be difficult to place because their risks are complex or fall outside standard categories.

Businesses with unusual buildings, specialist equipment or more complex operations need insurers to understand their circumstances, rather than only how they compare with more standard risks. Better information will not make every risk insurable, or guarantee lower premiums, but it could help insurers make more informed decisions about which risks they are able to support.

If AI can help bring those details to light, it could give underwriters a better starting point for conversations with brokers and businesses. Instead of seeing a risk as difficult to assess, they might be able to identify what further information they need before making these decisions.

It is too early to say whether this would lead to more cover being offered or lower premiums, but it raises the question if better assessed information will help insurers make more nuanced decisions about risks that might otherwise be difficult to place.

Could insurance become a real-life Minority Report?

The idea of identifying trouble before it happens might sound like something from Minority Report, but how close could insurance really get to predicting risk? Our research identified patterns in historical data associated with claims, the next step is to establish whether these patterns can help insurers anticipate future problems.

If they can, these tools could help underwriters spot potential warning signs earlier, ask more targeted questions and identify where better risk management might make a difference. Over time, this could help assess risk based on the information available right now to understand how it might develop looking ahead.

While that future is not quite here yet, and the findings need to be tested across other portfolios and over time, the potential is still significant. By uncovering indicators that might otherwise go unnoticed, AI could give underwriters a fuller picture of the risks they take on and help them explore new ways to support businesses that need cover. Minority Report may remain fiction, but a more forward-looking approach to understanding potential risk is a real possibility.