Sixty workflows audited. Three shipped. Eighty hours back.
3 build · 12 buy · 19 wait · 26 skip
Across 14 senior brokers
Staged rollout, no layoffs
Owned infrastructure end-to-end
The problem.
The board had told the executive team to "do something with AI." The CTO had four vendor demos booked the week I arrived. The operations director was anxious — twelve of the loudest internal voices had pet projects, none with a measurable outcome. The risk was real: this was an insurance broker with a regulator, sixty years of process, and a workforce that didn't know if the boss wanted them to use ChatGPT or hide from it.
The approach.
First three weeks: I ran every candidate workflow through the AI Opportunity Finder scorer. Sixty workflows surfaced from interviews with team leads. Each was scored on volume, judgement complexity, data accessibility, error tolerance, and team familiarity. Three came back BUILD. Twelve came back BUY (point-solution SaaS — no custom build needed). Nineteen came back WAIT (fix the data first). Twenty- six came back SKIP (not an AI problem, or the human judgement was the value). I gave the board the scorecard, not the demo. Then we built the three BUILD workflows over ten weeks: a quote-comparison summarizer, a renewal-letter generator, and a claims-triage classifier. Each shipped with a human-in-the-loop step. Each had a measurable baseline before launch.
The outcome.
Three automations live in production. The summarizer reclaims thirty-two hours/week across the new-business team. The renewal generator reclaims twenty-eight hours/week. The claims classifier reclaims twenty hours/week. Total: eighty hours of senior-broker time per week. Nobody was let go — the reclaimed time went into client retention work the team had been deferring for two years. The board got the AI story they wanted. The regulator got an audit trail. The CTO cancelled three of the four vendor demos.
If you need to give an honest answer about where AI helps and where it doesn't — let's talk. One slot open for Q3 2026.