Cake & Arrow says insurance AI needs better workflow fit

Cake & Arrow has released a report on how insurance agents and brokers are using artificial intelligence at work, based on interviews with 16 professionals across 13 US states.

The report, titled The Connective Thread: From Agent and Broker Research to a New Design Vision for AI-Enabled Insurance Work, examines a gap between insurers' AI spending and adoption by the people expected to use the tools in their daily work.

The findings suggest AI is already taking hold on the front line of insurance distribution, but often through individual trial and error rather than formal company programmes. Agents and brokers described working with fragmented systems, uneven guidance and limited organisational support as they tested tools for tasks such as drafting, preparation and follow-up.

The research drew on in-depth qualitative interviews with participants in a range of roles, business settings and experience levels, from six months in the sector to more than 30 years. The group included people working at firms of different sizes and in different operating environments.

The study identified four recurring themes: AI is spreading without a clear roadmap; agents are seeing some efficiency gains, but use remains relatively shallow; users want better integration rather than more automation; and human expertise remains central.

Those findings reflect a wider issue for insurers, brokerages and agencies trying to embed AI into established workflows. While many firms have increased investment in AI tools, employees on the distribution side still appear to be relying heavily on their own judgement about where and how to use them.

Josh Levine, Founder & CEO, Cake & Arrow, said the interviews did not show resistance to the technology itself.

"The agents we spoke with are not resistant to AI," said Josh Levine, Founder & CEO, Cake & Arrow. "They're curious, resourceful, and already finding ways to use it. But too often, they're being left to figure it out on their own. The opportunity for the industry is not simply to build more AI tools. It's to design AI that fits how agents actually work, earns their trust and strengthens the human relationships at the center of insurance."

Design approach

Alongside the research, the report sets out a proposed design model called Adjacent. The concept is framed as an AI sidebar that sits alongside the systems agents and brokers already use, including agency management systems, customer relationship management software, email, meetings, documents and carrier portals.

Rather than replacing those systems, the idea is to create a layer that links them more closely. This, Cake & Arrow argues, could help account teams prepare for interactions, make decisions, draft communications, follow up on tasks and update records while keeping staff in control of the process.

The emphasis on integration reflects one of the report's central arguments: adoption may depend less on adding standalone AI products and more on reducing friction across existing workflows. In insurance distribution, work often moves across multiple systems, making it harder to introduce another separate tool.

The report also outlines five principles for AI design in this area. These include designing for integration to address fragmentation, building tools around a "trust but verify" approach, keeping the human role at the forefront, designing around the full workflow rather than isolated moments, and supporting knowledge transfer as well as efficiency.

That focus suggests the agency sees AI as an aid to professional judgement rather than a substitute for it. The research argues that experienced staff remain essential in insurance work, particularly where client relationships, product interpretation and decision-making are concerned.

Levine said the main challenge is not simply deploying the technology.

"Technology doesn't roll itself out," said Levine. "For AI to make a meaningful difference in insurance distribution, it has to solve real problems inside real workflows. That requires more than technical capability. It requires intentional design."

The findings add to a broader debate across financial services and insurance about whether AI projects are being built around operational realities. In many cases, firms have focused on the promise of automation, while users on the ground have had to work around legacy systems, compliance demands and long-established ways of dealing with customers.

For insurers and intermediaries, that raises a practical question: whether current AI spending is aligned with the daily needs of account teams and advisers. The report argues that trust, usability and workflow fit will shape adoption more than the presence of AI alone.

The study found that agents and brokers were generally willing to experiment, but that willingness had not always been matched by structured support from employers. That leaves organisations at risk of uneven use, inconsistent processes and a gap between strategy set at the top and practice on the ground.

Research based on 16 interviews is limited in scale, but it offers a snapshot of how AI is entering insurance work through everyday use rather than formal transformation programmes. Participants' accounts point to a sector where adoption is already under way, but the structure around that adoption remains unsettled.