People of a certain age remember word processers. These computing devices were not computers per se. Unlike the modern desktop, or even the phone in your pocket, a word processor could do one thing, and one thing only: help people to produce documents.

These highly technical devices, when compared to a simple IBM Selectric, were as a jumbo jet to the Wright Brothers’ plane at Kitty Hawk. Operating them was a job for a skilled technician. In those early days, therefore, users had to be specialists. These people, like the machines, were also often also known word processors. Sometimes they were called ‘word processing engineers’, which in retrospect is an extraordinarily lofty moniker for a typist.  

At a PKF Littlejohn dinner this week, invited carriers, MGAs, and your diarist gathered to discuss the gamut of topics to do with MGAs. One such was the recruitment needs of the MGA of the future. Inevitably, perhaps, the diners agreed that people who are skilled at AI would be essential.

Your humble correspondent broke the unanimity.

“No,” I said. “Future MGAs will need exactly the same kind of people MGAs need right now: experienced risk experts with excellent judgement that they apply well, to deliver positive underwriting results the majority of the time.”

In the same way that we don’t all need to employ specialist word processors or telephone operators, and we don’t now and never will need to hire internet engineers, underwriting shops don’t need trained AI specialists. They’re unnecessary even in the IT department, unless the MGA in question is executing an ambitious build, which I don’t recommend (see Diaries passim). All the regular IT people will have sufficient AI skills to do what’s needed.

Further, even those MGAs that claim technology as competitive advantage probably don’t need an IT department (unless they’re rather large), let alone an AI man. MGAs rely on electricity, but they don’t employ a sparky. A great many have even outsourced their accounting.

Already everyone uses AI. You’ve noticed the rapid evolution of internet search results, for example. They’re no longer a page of relevant links. Now  Google or Bing throw up a quasi-coherent answer to the search term’s implied question. It comes in the form of a summary of the potentially correct information that the internet has on it. That’s AI, in the form of a large language Model or LLM, hard at work.

It’s everywhere. By now almost everyone has used an AI chatbot either at home or in the office, and probably both. Fuzzy logic and machine learning are in use in everything from the fridge in the kitchen to the Netflix interface. And we can all use Word, Microsoft’s universal word processor, without special training, too.

No, the MGA of the future will need experienced underwriters. It will also need juniors who will become the next tranche of experienced underwriters. That’s not to say MGAs do not require AI. They most certainly do, just like they need Word, the internet, and plug sockets. Already to underwrite without AI is to play from the back of the field. That doesn’t mean you should rely on AI for underwriting. Certainly not. It is a tool to help underwriters be better at their jobs by giving them more space to exercise their judgement, and more information on which to base them.  

I have seen even the most traditional underwriters find a use for AI tools, and swiftly. Well-deployed LLMs bring them greater clarity about the data they already have, and therefore allow them to use their judgement and expertise to make even better underwriting decisions. Meanwhile they appreciate the utility of AI to automate mundane tasks like submission triage and document summation. That reduces expense ratios (provided they don’t spend too much on AI), which helps them produce sub-100 combined ratios. They all love that.

That has to be the main goal. Hiring 50 data scientists straight out of Cambridge University won’t give you an underwriting profit. Experience and judgement cannot be automated (yet).

Because of this, I was slightly alarmed to read earlier this week that some wag has counted up the number of times Hiscox people mentioned “artificial intelligence” in the company’s earnings calls since Q1, 2021. It’s 98 times. The article said AI mentions during the annual earnings call were up 617%.

I downloaded the transcripts and did some counting of my own. I found that Hiscox mentioned “AI” or “GenAI” 28 times in its year-end call with analysts. They said “LLM” six times, and “automation” four. That’s 38 AI words in total. In comparison, “underwrit-” was said 46 times, but “risk” was mentioned only 29 times, and “profit” only 22. “Product” had 24 mentions, “customers” 29, and brokers garnered only 12 mentions. During the Q1 call the AI-related words had another 14 mentions, but underwriting merited only five, brokers 9, risk 4, and profit just 2.

What should really matter, from an investor perspective, is an insurance underwriting company’s ability to deliver sub-100 combined ratios, plus their cost of capital (or, at least, an investment result that makes up the difference). Can AI really be more important than profit, product, customers, or brokers?

I am certain that the savvy IR team at Hiscox is telling investment analysts and shareholders the information they want to know about, and will make sure they send the message clearly. But investors themselves should not be quite so interested in AI adoption and use. Underwriting quality and the insurance pricing cycle are far more relevant. Profits matter most when valuing a business. When AI fails to increase profitability, it becomes a drag on profits.  

Soon – if not already – saying you use AI in your business will be the equivalent of saying you’re using Word, the internet, or electricity. Underwriting companies must articulate instead the ways in which AI has reduced their costs or increased their sales, or both. Equally (or perhaps more importantly), they should explain how continued implementation will deliver long-term shareholder benefits. (Always mentioned in the plural, in shareholder benefits are in practice only singular: profit.)

It doesn’t matter how many times you say AI, or how many specialists you have working on it. If it doesn’t add quickly to the bottom line, it’s just another cost someone will have to trim later. If your title is AI Underwriting Support Lead, watch out!

* Like every Insurance Technology Diary entry about AI, this one is accurate only to the best of my knowledge at the time of writing. The pace of AI progress is so great that I cannot guarantee it remains so now that it’s finished, let alone when you read it.

Guillaume Bonnissent is CEO of Quotech.