@Pixabay.
Artificial intelligence (AI) has long promised to revolutionise the human experience, but more recent developments have opened society’s eyes to it’s true potential. With an ability to scan and process oceans of data in the fraction of the time it would take an army of humans, and its ability to make suggestions on the back of that, the possibilities it throws up for financial services are huge, no more so than in insurance and asset management.
The potential uses for both predictive and generative AI in the world of asset management have been debated for some time now but according to a recent report from Mercer, nine out of 10 asset managers were either currently using (54%) or were planning to use (37%) AI in their investment strategy or asset class research.
"73% of insurers are first looking to reduce overall operational costs,
while 39% are starting to incorporate AI in risk and underwriting."
Elsewhere, a Goldman Sachs survey of 359 CIOs and CFOs in insurance companies, collectively controlling more than $13 trillion in assets, found that 29% of insurers globally were currently using AI with 51% looking at implementing the tech.
Commenting on the findings in the report, Matt Armas, global co-head of insurance at Goldman Sachs Asset Management, said: “What we’re seeing is clients starting to use AI in the highest-return areas first. Approximately 73% of insurers are first looking to reduce overall operational costs, while 39% are starting to incorporate AI in risk and underwriting, particularly in the Americas.
“And there is also an emergence in trying to use it in evaluating investments.”
Insurers and their asset managers have clearly moved beyond considering the use of AI and are now actively implementing it in their businesses or as Joshua Zwick, Partner, Insurance and Asset Management at Oliver Wyman put it: “Generative AI is no longer on the whiteboard. It has come out of the laboratory, into the production lines and that’s creating a real impact.”
Having said that, he is quick to caution against viewing generative AI as the answer to every firm’s productivity and insight challenges.
“It’s not the solution to all problems. Organisations that understand the difference between predictive and generative AI most effectively are the most successful,” he said.
He explained that predictive AI is based on making predictions of what might and might not happen in any given scenario but based upon huge amounts of historical data. Generative AI, on the other hand, works with unstructured data to create new media and new ideas.
“The scope of what it can do is huge, but the question is how you harness that in a way that is useful for asset management. If you use a combination of both predictive and generative AI and understand their relative strengths, that combination could be very effective,” he said.
"We are seeing a lot of solutions that use language models
to collect and construct data."
He outlined three main uses for this technology among asset manager today – using data to calculate a company’s likelihood of default, supporting researchers in their work and summarising hundreds of pages of information into more digestible formats
“There are a lot of great ways this tech can be used in the back office but there are more bold uses. We have seen prototypes that are reading reports and coming up with their own thesis and then feeding that back to the investment team,” said Adam Lieberman, Chief AI Officer at Finastra, a software platform serving the financial services sector.
“But it can also be used for data collection. We are seeing a lot of solutions that use large language models to collect and construct data, whether that be for income statements or balance sheets. The AI will format and cleanse it all for you.”
These examples tend to be the preserve of those who are most comfortable with using AI and that tends to be the quant funds. The more traditional operators, while exploring its use, appear to be just dipping their toes at this stage and in insurance, it looks like they are only now approaching the water’s edge, socks and shows still firmly on.
“There are a host of regulatory requirements and restrictions placed on insurers and if they were to use it to automate decisions, there is always the chance of getting something wrong. In managing claims for example, which would be a horrible situation. These companies are operating in very sensitive domains,” he explained.
He said there was an innate conservatism in the insurance world which is understandable as they deal with potentially life changing decisions daily: “Some of these sectors are more sensitive to the fact that tech must be tried and tested,” he said.
The result of that conservatism?
“I’d be hard pressed to say there are any insurers are at the bleeding edge of this technology. But any meaningful insurer is experimenting with these tools, and I don’t know anyone who is sticking their head in the sand on the issue,” said Zwick.
While that natural conservatism may have allowed the other financial services steal a technological march on insurers in the past, it may be that this time, the caution is well founded.
A report from WTW took an alternative look at the implementation of AI into asset management operations – the risks it brings to well-oiled (and regulated) operations.
Regulatory issues are at the forefront with the report highlighting that the Securities and Exchange Commission in the US has started to flex its muscles around the use of AI in asset management.
“AI has the potential to be a powerful tool in the sector
but its adoption in the industry is limited."
It has been questioning advisers on how they use AI in their algorithmic models and what oversights they have in place while in March this year, it fined two advisers $400,000 each for AI-related breaches including ‘AI washing’, where a firm issues false or misleading statements around the use and value of AI in managing investments.
But the threats go beyond the realms of regulation. Amongst others, the report highlighted the risks around data accuracy, the existence of biased data, compromised cybersecurity and the risk of allowing AI to execute trades.
“AI has the potential to be a powerful tool in the sector but its adoption in the industry is limited,” said William Gibbons, Senior Insurance Investment Consultant, Mercer.
“Before we see further adoption, more work will need to be done to ensure it meets regulatory compliance standards, is robust enough and ensures we can provide rigorous risk management services to our clients.”
This caution is reflected in the approach taken by Sompo, the only insurer out of over 20 contacted that was willing to discuss its use of AI in managing assets.
“We are currently gathering data on how to use AI in asset management and its effects. For example, there is potential to implement AI into predicting economic indicators and assets as well as algorithmic trading. However, as there are issues with accuracy and interpretability, it is in the experimental phase and is difficult to tell if it will be a standard element,” said a spokesperson.
This caution is understandable in such a highly regulated sector where decisions have a real-world impact on real lives. But for those insurers still hovering by the water’s edge, there are ways that AI can be used in a controlled manner, as long as the data source is reputable. And what could be more reputable than your own insight?
“We have seen some people amass every single research note they have written over the last 40 years, collate it into a system and they then have an infinitely scalable, knowledgeable ‘super analyst’ that has access to every single idea they have proposed over last few decades,” said Zwick.“You can ask this super analyst about a company, and it will serve up a view on it. You may agree or disagree with that but just imagine having access to your entire organisation’s history of research,” he added.
And it can be used rather benignly and in-house to test investment ideas or approaches.
“Humans have biases. They think and make connections in certain ways, but these tools think a bit more nonlinearly. So, there is an opportunity to have an explorative conversation with AI, which doesn’t think like you, which might allow you to see a connection that you hadn’t thought of before,” said Zwick.
"It’s not that AI will replace humans but that the humans
who understand AI, will replace those who don’t."
Which sounds like an excellent way to test a thesis but how can asset managers trust the responses they’re getting? How can they know that the AI is delivering a sound appraisal of the situation?
“Trust is one of the limitations of AI – you need that human oversight. To be safe, you have to use AI alongside human judgement,” said Lieberman.
Which supports the belief that AI won’t ultimately push humans out of investment decisions and will instead assume the role of supporting actor to the tried and tested human expert.
“The best outcome comes from a combination of a human and AI. It’s not that AI will replace humans but that the humans who understand AI, will replace those who don’t,” said Zwick.
Or as Gibbons said: “The insurance space has always sought, where appropriate, to adopt technology early to improve decision making and provide a better service to clients and I am sure AI will be no different. However, we don’t see AI replacing the role of humans completely in the asset management space in the short to medium term.”
