Insurance organizations are pouring money into artificial intelligence, with AI software spending forecast to reach $297 billion by 2027, but a new report from experience design firm Cake & Arrow finds that investment is failing to reach the agents who need it most.
Only 8% of independent agents are using AI on a daily basis, according to Liberty Mutual data cited in the report. Based on in-depth interviews with 16 agents and brokers across 13 states, the report finds that AI adoption in the agent and broker channel is largely ungoverned, inconsistently implemented, and producing far less value than the technology is capable of delivering.
The picture that emerges is one of improvisation at scale. Most agents interviewed fell somewhere between “ungoverned” and “informal” on a spectrum of AI governance maturity, the report found. In practice, that meant agents were handed access to tools with little guidance, left to learn from peers, and in some cases uncertain whether they were even permitted to use the AI tools already on their desktops. One participant, a sales manager at a captive agency, told researchers that more training around use cases and prompting “would go a long way.”
Workarounds Are Creating Compliance Risk
The governance gap is not just an efficiency problem. With limited organizational direction, agents are increasingly turning to unsanctioned tools to get the job done. The report cites Gartner research finding that 88% of employees with enterprise AI access also use personal AI tools for work tasks, a pattern the agent and broker interviews confirmed. Participants described routing work through personal ChatGPT accounts and consumer applications their employers have no visibility into, less out of recklessness than out of a lack of guidance.
The exposure this creates runs in two directions, the report notes. Sensitive client data flowing through uncontrolled channels creates compliance and security risk. At the same time, the agents most likely to be using unauthorized tools are often among the most productive, making the behavior difficult to discourage.
One commercial account manager told researchers she was unsure whether using her personal ChatGPT account was permitted but added: “I’m honestly not sure how I’d manage a book this large without it, so I’m not asking a lot of questions.”
Shallow Use and a Looming Knowledge Gap
Even among agents actively using AI, the report finds that adoption tends to be shallow. Most are using AI for routine tasks: drafting emails, summarizing policies, comparing quotes, rather than the deeper workflow integration that produces compounding productivity gains.
Cake & Arrow cited Gartner data showing that employees who use AI across nine to 12 use cases are five times more likely to report high productivity than those using it for just one to three. Most agents, the report notes, are nowhere near that threshold.
The agents who are getting the most from AI, researchers found, are not necessarily the most tech-savvy. They are the ones with enough domain expertise to know when AI output should be questioned.
An account executive with 31 years of experience told researchers: “You need to be able to look at an AI output and know something’s not quite right.”
That judgment, the report notes, is built through experience, not prompting, which creates a specific risk for newer agents who may rely on AI as a primary source of truth before they have the expertise to catch its errors. One veteran agent described that scenario plainly as “an E&O claim waiting to happen.”
The longer-term stakes are equally significant. With 50% of the insurance workforce projected to retire over the next 10 to 15 years, the report identifies a looming knowledge drain that AI is uniquely positioned to help address, but only if organizations build tools and governance structures capable of capturing and transferring institutional knowledge before it walks out the door.
The report frames the adoption gap not as a technology failure or a people problem, but as a design and organizational investment failure, arguing that the organizations pulling ahead are those doing the slower work of understanding how agents actually work before deploying tools meant to help them.
Obtain the full report here . &