InsureAI, a software developer led by renowned actuary Ron Richman, has unveiled a general insurance (GI) reserving tool that promises to empower actuaries to make better decisions and contribute more to strategy.
The ReserveAI solution offers reserving actuaries access to 8,000-plus variations of customisable reserving models, including the commonly used approaches such as chain-ladder, advanced machine-learning models and stochastic methods.
But the most novel aspect is how the software helps actuaries pick the most suitable model. The software can quickly run claims data through thousands of models and present the outputs in a similar way to how an “efficient frontier” curve is used to assess the optimal risk and return in asset portfolios.
In this case, the “model curve” presents actuaries with relevant information so they can express their judgement – for example, whether they want a reserving model that is more sensitive to recent claims trends, or is more stable over time.
“Actuaries spend a significant proportion of their time on getting an initial estimate of a triangle, and little time is left to spend on assessing diagnostics and talking to management teams about what the impact of these reserves are,” said Rowald van der Walt, head of reserving and capital at InsureAI.
“Our vision is transforming this into a state where the bulk of actuaries’ time is actually spent on the diagnostics and management insights.”
For example, Richman says, a motor insurer may want to study how their theft trends differ from bodily injury trends, or examine the claims experience they get from different brokers – analysis that ReserveAI can deliver rapidly.
“The whole exercise of reserving allows you to see trends in your data well before these events are picked up by other departments. That proactive analysis with other teams is something that we're aiming towards,” van der Walt said.
The analytics in ReserveAI has its origins in a paper that Richman co-authored, The Actuary and IBNR Techniques: A Machine Learning Approach, which won the Brian Hey Prize in 2020.
The paper developed a numerical scoring method for judging reserving techniques, and has been implemented in rival reserving solutions.
Richman explained InsureAI has further developed this idea and created some proprietary methods, which have not been published, around scoring reserving methods in a way that can identify the techniques that are actuarially reasonable, i.e. that are similar to what an actuary might have done in the situation.
“Instead of having thousands of models that are overwhelming to the actuary, we can say ‘this is what we see as being the best models’,” said Richman. The actuary can then dive deep into a model to conduct the appropriate assurance and adjust assumptions as necessary.
“Our whole ethos is empowering action and not replacing it with AI,” he added.
The InsureAI team has many years’ practical experience in reserving at insurers, and the ReserveAI software includes features that they said would have made their lives easier. These include a modern reserving dashboard to help detect adverse development; the ability to deselect diagonals from claims triangles to enable “as if” analysis; and plotting functionality by period, so an actuary can see how a particular cohort is developing.
ReserveAI has its own cloud-based platform and graphical interface where actuaries can upload data, select assumptions, perform analysis and display outcomes. Workflows are fully documented for governance purposes. But the calculation engine can also be accessed via a software development kit (SDK) that will enable actuarial teams to create custom workflows.
“We're seeing the market increasingly moving towards this approach where they have a data science system in place – something that pings directly to your warehouse and then feeds into your calculation tools. So we've designed this SDK to facilitate that process. It doesn't lock you into an ecosystem,” said van der Walt.
The firm expects this flexible and customisable interface layer will be popular with insurers with a wide variety of data sources underlying their reserving. For example, cell captive operators often have a challenge in integrating data from a wide variety of clients. Similarly, insurers working with managing general agents (MGAs), which have their own standards and ways of working at a licence level, may have a challenge in controlling data.
Richman said there are two South African clients currently using ReserveAI and the company is seeking to expand globally, with an initial focus on the UK, Canada and the US.
The software is licensed by an annual subscription, with no additional charges for stochastic methods, and a “fair use” agreement on cloud usage.