No one can deny that significant transformation within the insurance market is needed – the financial losses over the past several years speak for themselves. All areas are under scrutiny as the market searches for opportunities to tighten up processes and remove unnecessary expense from the insurance value chain.
Underwriting has often been described as an ‘art’ where good judgement and decision making is exercised. However, there’s a limit to the amount of information and knowledge a human can retain and recall. With data becoming more and more accessible to us as a result of developments in technology, there are significant advantages that can be gained from its use, enabling underwriters to continue to meet the changing demands of clients within an increasingly competitive landscape.
It’s important to also consider the host of cognitive biases present in the human decision-making process and how technology can help underwriters to try and counter this bias. Insurance Day recently published an article¹ highlighting the importance of removing bias from the underwriting process and how artificial intelligence systems can pick up the bias through machine learning.
Having robust machine learning algorithms which base their decisions purely on the available behavioural data rather than on past decisions helps to remove the cognitive bias and provides additional relevant information in a much more detailed way. The analogy I often like to talk about is: traditionally, you may have one or two people who will be responsible for the whole insurance portfolio, identifying trends and then determining the company guidelines, i.e. we should avoid trawler fishing vessels as they are too high a risk. Imagine you then have hundreds of thousands of people who are now responsible for analysing the data, and identifying trends across your entire portfolio, etc. That is what machine learning algorithms allow you to do.
We presented insights to a client recently for cargo vessels where our data analytics revealed that the smaller the cargo vessel, the higher the risk. Our data combined both static and behavioural rating factors including length and breadth of the vessel, detailed journey information and claims and provided the client with a different, more detailed view of risk.
Rather than concluding that all vessels of a certain type carry the same level of risk, the machine learning algorithms analyse the behaviour of individual vessels alongside specific behaviours that lead to a claim and then compare the results to similar vessels. This approach provides a much more granular understanding of the exposure – the benefits of which are twofold: it helps to spot similar vessels that display the same risky behaviours and flag them as high risk and it uncovers vessels with a lower risk. When combined with their experience this insight is extremely valuable in helping underwriters to price and select risk and win new business.
It’s for this reason that we’re likely to see whole fleet policies disappear over time. Policies may start to be written at an individual vessel level; but such an approach may mean an initial increase in administration costs. Businesses need to carefully consider the way forward, but we’re already seeing some clients (who have access to claims information at an individual vessel level) taking this route.
Having access to robust dynamic data can help improve underwriting. A demonstrable example that we talk about frequently is the Fluvius Tamar. In 2017, the cargo vessel was a total loss after sinking in the North Sea. Within our own risk model, we rated the vessel a 4 out of 5 - 1 being the lowest and 5 the highest. Although it will always be difficult to predict total losses, it did highlight that the vessel was undertaking double the amount of port visits and number of unique journeys per year – both of which were important indicators of risk.
Whilst it wouldn’t have given the insurer the ability to completely mitigate against that risk, by having a better understanding of the vessel’s behaviour, more effective action could have been taken when the policy was originally written.
What underwriters are doing currently is taking on a significant debt based on a very limited amount of information. The opportunity to take on the same significant debt based on a far larger amount of information through dynamic data analytics will undoubtedly present a much stronger and surer bet.
Combining their extensive knowledge and experience with the new data insights available will provide underwriters with the superpowers they need to survive market conditions and drive dramatic improvements in loss ratios.