AI requires human touch to fulfil risk management promise

AI offers huge opportunity for risk management but is just another tool with limitations, so risk professional must question its output and overlay their own analytical skills before relying on its conclusions, said experts at the Brokerslink conference.

Panellists at the event in Abu Dhabi warned of inherent bias within AI and a lack of credibility in some of the data sets it uses. This can throw up incorrect risk management assumptions and create group think based on answers that have little or no accountability, said a leading risk manager.

However, on the plus side, AI’s speed and breadth of data processing offers predictive modelling, risk mitigation and customised risk solutions, the experts said. They also believe insurers will be able to harness AI’s predictive powers to deliver better products more quickly, particularly around emerging risks.

Ultimately, the panellists agreed that there is no option for risk professionals than to carefully embrace AI. Those that don’t will find themselves falling behind the game.

“AI is just a tool, and like many other tools, it can be very beneficial if used wisely, but it will always need to be manned,” said Maurizio Castelli, head of Brokerslink’s risk management practice and CEO of consultant Augustas Risk Services in Italy. “So we in the risk profession can certainly take advantage of this tool that offers more opportunity than threats as long as it is managed properly.”

Shane Keating, chief data officer at The Ardonagh Analytics Lab, said it is vital that AI’s outputs are transparent and traceable to build trust. But above all, humans must question its output, he said. “Human analytical skills will always be needed with AI,” added Keating.

Ole Ohlmeyer, global head of customer and distribution management and ART at Swiss Re Corporate Solutions, agreed. “Whatever a models says, if you don’t apply the human touch you will face problems and potentially draw the wrong conclusions,” he said.

Fellow speaker Manuel Padilla, vice-president of risk management and insurance at MacAndrews & Forbes Incorporated, as well as a Rims board member, said there are clear risks and challenges when it comes to AI, alongside the opportunities.

“I would agree that the opportunities are there. But if we lull ourselves into a false sense of comfort believing AI is going to tell us exactly what needs to be done, and someone is going to bypass our knowledge base to get the answer, I think that is dangerous. We need to understand what it does and its limitations, and we need to question it because AI’s bias is the key issue,” he said.

The risk manager said the potential lack of credibility in data sets used by AI can be a problem and let group think set in. Padilla added that risk managers must be aware that AI doesn’t really care if it is right or wrong ­– because it is often just programmed to give an answer – and has little or no accountability. There is a real risk that people become over reliant on the tool, he stressed.

“Algorithms won’t gave bad information maliciously but can be programmed to provide a response regardless of what the response is. So if you are making a business decision, you can have a big problem,” he said.

Like others, Padilla stressed that AI is there to support rather than replace risk managers and should be used as part of a “digital partnership”.

He and other panellists put AI’s efficiency, speed of processing and ability to deliver deeper insight with enhanced accuracy on the opportunity side of the scale. “AI presents opportunities for predictive modelling, risk mitigation and customised risk solutions,” said Padilla.

Keating said AI’s ability to help insurers more deeply understand risk creates “big opportunities” around better product development and solutions that are a “better fit for clients and more aligned with what they need”.

“With AI, the real opportunity here is around predictive application and especially in the area of recently emerged or emerging risks. We are an industry that has relied very heavily on years and years of claims history. But how do we develop solutions for risks that have no claims history? What are you going to build or base your model on? For some of these risks emerging, there isn’t time to wait for two or three years of experience, so we need to find ways to better predict. Having the ability to apply predictive technology to things you don’t necessarily know about now will apply to short-tail as well as long-tail business,” he said.

Ohlmeyer noted that AI has “significant” ability to write, check and compare insurance wordings.

“A benefit is enhancing brokers knowledge around the right product as large language models allow us to compare policy schedules, covers and all sort of other things really quickly, so you are not spending days referring back to the policy. You can get all of those answers now in a fairly self-contained, accurate environment within your own data. That will allow us to respond to things faster,” agreed Keating.

Ohlmeyer added that AI can do a lot of the “grunt work” that takes up a time at insurers, so underwriters can focus on understanding risk and developing key services. AI will deliver efficiency gains when it comes to compliance, while producing documentation will become a lot easier and be taken out of day-to-day activity, said Keating.

He believes that with prudence, risk managers, brokers and insurers can use AI to be more ambitious, and urged all of these risk professionals to embrace the technology.

“I just encourage you to accept that it is here and changing things whether we like it or not, so lean into it but in a way that is prudent. Take small steps initially, get comfortable with the results and how you are going to apply it and the way you are doing it, and go from there,” he said.