Heffernan Insurance Brokers is using AI to reshape how brokers work
Kate Grasman, chief information officer at Heffernan Insurance Brokers, arrived three and a half years ago with a mandate beyond maintaining systems. Her focus was transformation-and she's backing it with a structured experimentation program that tests six new technologies annually.
The approach has positioned the brokerage to move early on both traditional AI and what Grasman calls agentic AI, systems designed to orchestrate entire workflows while keeping humans involved. The results are measurable: $400,000 in savings from a single automation platform, and a 50% improvement in placement success rates from an experimental carrier database.
From task automation to operational savings
Grasman's initial strategy targeted repetitive work that consumed producer and account manager time. She adopted ChatGPT over two and a half years ago-early for a traditional insurance firm-and licensed it directly to the company.
The team then embedded AI into specific operational processes. A platform called Fulcrum automates policy checks, proposal generation, and policy comparisons-work that previously went offshore. The system now produces proposals in seconds and checks policy documents with 96 to 98% accuracy, trained through collaboration with an innovation startup.
"We've already quantified it. We've already canceled our work with an outsourced provider in India," Grasman said. The result is faster turnaround for frontline teams.
Building workflows, not just automating tasks
Grasman distinguishes between AI-which makes individual tasks more efficient-and agentic AI, which orchestrates multiple tasks while keeping humans in control. One example currently in development is a proprietary carrier appetite database built with Stitch Studio.
The system helps producers identify insurers most likely to write specific risks by automatically gathering data from internal communications and submissions. Account managers and producers provide minimal input; if something looks promising, they send an email and the system ingests it through AI.
Early testing found a 50% improvement in placement success rates. The project began as an experimental initiative run by high school interns.
A staged approach to testing and adoption
Heffernan's innovation program follows a structured framework. Tests and learns start with two or three users to answer a basic question: does this solution work? Many fail at this stage.
Promising concepts move to a broader pilot involving roughly 20 employees across different technology comfort levels. An internal AI committee of about 25 employees evaluates new tools.
Adoption often becomes the biggest barrier. The brokerage spent two years developing a proposal tool Grasman considered a clear winner, but discovered the problem: generating initial proposals from scratch required more work than the old process. The team temporarily added staff to help generate early proposals until workflows stabilized and staff became comfortable with the system.
"The change, adoption, and figuring out why the tech isn't skyrocketing was critical," Grasman said.
Evaluating vendors with hands-on rigor
Grasman describes herself as a "buy" CIO rather than a builder. Her team evaluates roughly 50 AI companies annually and pilots six per year.
Vendor selection relies on hands-on demonstrations. "If they only have a video of their demo, I'm done," she said. "They must show me the solution."
The team also assesses the people behind the technology. Industry networks help surface new startups and ideas. This approach keeps Heffernan focused on its core business-selling insurance-rather than building technology products.
AI Agents & Automation are reshaping how insurance operations function, but success depends on disciplined pilots, clear adoption strategies, and the right vendor partnerships.
Your membership also unlocks:
