Commercial Insurance GTM Engineering
A broker-facing GTM system for a construction and energy insurer: three plays that find the moment a broker is about to place risk, make the carrier familiar at that office, and measure whether answer engines name it, all pointed at one number, submission volume.
Overview
Commercial construction and energy insurance is sold entirely through brokers, so growth means more brokerage offices sending risk and the brokers who already do sending more. This system runs three plays against that. Play 1 is signal-led outbound: a broker universe pulled from Apollo and scored on four axes (fit, timing, access, intent), a weekday sweep that reads the trade press and classifies nine signal types with a written thesis for each, cadences that enroll the right brokers when a signal fires, a do-not-contact list checked at enrollment, and an agent that drafts every step in the underwriter's voice for a person to approve. Play 2 is targeting: the scored universe sliced into a LinkedIn-shaped audience so paid media and outbound point at one list. Play 3 is AI visibility: an audit that asks eleven answer engines the questions brokers actually ask, records who gets named and cited, ranks which prompts to win first, and drafts the page that closes each gap. A spend dashboard prices every play in cost per submission, and a console holds the appetite, voice, underwriters, sources, and prompts. Every figure carries a provenance label and a test suite asserts those labels against the code, so a number cannot quietly become a claim. The broker universe is public data, revealed contacts are masked, and the carrier is described by what it writes rather than by name.
What it involved
- Signal thesis, not a feed: nine signal types ranked into act, aim, and context tiers, each with its sourcing, inclusion criteria, and confidence written down, and loss events deliberately excluded from outreach.
- Four-axis broker scoring (fit, timing, access, intent, 25 points each) summed in code, with intent hard-capped at zero until real engagement is logged, so nothing scores hot on a guess.
- Cadences that fire from signals: a project award, a practice hire, or a carrier pullback enrolls the brokers at that firm, the agent writes each step, and a human reply stops the machine.
- A broker one-pager built from live web research, with its sources, before an underwriter picks up the phone.
- A targeting cockpit that turns the scored universe into a LinkedIn company list and title set, sized against the real universe, so paid and outbound stay on one list.
- An answer-engine audit across eleven engines with Profound-style visibility, rank, share of voice, and citation share, plus page drafts with FAQ schema for every gap.
- Provenance on every number (live, stored, by hand, configured, needs a key, illustrative, planned), asserted by tests against the code that backs it.
Stack
- Next.js 16
- React 19
- TypeScript
- Tailwind v4
- Supabase Postgres
- Claude (scoring, drafting, briefs, audit)
- Groq
- Apollo
- Hetzner cron