AI prompt engines create risks for underwriters and agents

• Usefulness of AI prompt engines depends on application
• Insurance agents stand to lose from adoption of prompt engines
• Accenture recommends oversight for underwriting applications

As AI prompt engines such as Claude and ChatGPT become increasingly popular for consumer use, it remains to be seen how they may catch on in business and industry – especially in insurance.

"When people ask about AI in insurance, they really mean, 'am I going to sit down at my desk and have a prompt engine in front of me, or am I going to have a drop down menu?'" said Nathan Golia, senior analyst, North America P&C insurance at Celent. Prompt engines are probably going to be "how we all interact with computing," he added.

The prospects for prompt engines at insurers – both customer-facing and for employees to work with – will depend on their application, said Andrew Dalton, chief product officer at MSI, an MGA owned by The Baldwin Group. 

"There's certainly a ton of opportunity to leverage AI to augment processes within the underwriting space," he said. "That's something that we're doing in a number of our programs here at MSI. What exactly that looks like in terms of the UI [user interface] is going to vary. But something that remains really important is at the end of the day, you have a person checking all that work."

Moving to AI prompt engines is "the next logical step in the evolution of the interface," said Ellen Carney, independent insurance market analyst and strategist formerly of Forrester Research. Carney expects voice recognition engines to become a primary means of interacting with the AI prompt engines. "We're going to be speaking a lot more than we have in the past," she said.

Of all participants in the insurance industry, insurance agents would be the most impacted by a pairing of voice recognition with an AI prompt engine, according to Carney.

"If you're going to be able to speak to some technology interface, like you speak to an agent now, why do you need those agents?" she said. "Agents who are just in the middle under the bell curve, are going to be the ones at most risk. A mainstream agent that didn't work very hard because they knew the renewals were coming in – this is a huge risk for those kinds of distributors."

Agentic AI, however, according to Matthew Madsen, global commercial insurance lead at Accenture, actually increases the need for humans who know the business and can solve problems. "The future is not humans just writing prompts or interacting with chatbot agents," he stated in a written response to questions. "Rather, I expect a combination of conversational interaction, AI-driven recommendations, and core systems working together." 

Madsen points to underwriting workbenches that use agentic AI as an example. Greater, more complex risks, increased broker and customer expectations, and geopolitical volatility all increase the pressure on underwriters to get faster and more precise, he said. As a result, the underwriting workbenches are moving away from user interfaces and task managers and toward AI agents deciding what systems to use and what actions to take in response to unexpected events.

How much a user interface for underwriting should be augmented by AI depends on nuances in the underwriting itself, according to MSI's Dalton. "Regardless of what the user interface looks like, there's a ton of opportunity to leverage AI for underwriting to accelerate processes and help present the information to underwriters in a much faster way," he said. "There's less manual work to do, but what's really important is you still have the underwriter in the loop to make sure the right decisions are being made, and you're not ceding decision making to the AI itself."

The more various entities use AI, the more everyone else must sign on to it, according to Celent's Golia. "The impact of AI development as a component of the greater macroeconomic climate is a circuitous thing," he said. "In an industry where you're looking at data and trying to establish risk, changes happen faster. So you need AI to fight the world created by AI."