Ninety days ago, a multi-office brokerage in Hawaii started running its submissions through Compass. Since then the platform has processed over a thousand submissions and north of four million dollars in premium, filled 819 replacement cost estimators, and sent proposal packets that came back signed. Once, on a large commercial program, the signature came back fifty-one minutes after the packet went out. Those are the numbers. The lessons underneath them are worth more.
The ramp was word of mouth, not mandate
Monthly home submissions went 130 in April, 204 in May, 323 in June. There was no marketing push and no all-hands ultimatum. The pattern was desk to desk: a CSR who used to spend twenty minutes copying property data into a replacement cost form watches it fill itself in seconds, and tells the next desk. The first cohort trained the second. A new office came online because someone at the old one mentioned it. If your rollout plan depends on a memo, you do not have a rollout plan.
The boring feature was the wedge
We built an AI coverage check we are proud of. It did not win the first month. What won the first month was the most boring feature in the product: auto-filling a form everyone hated filling. Five minutes saved, 819 times, is 68 hours. More important, it is a visible, personal gift of time on day one. Adoption follows the feature that gives time back immediately. Trust in the ambitious features rides in on the back of the boring one.
The resistance is real, and it is not a software problem
Some share of every agency's staff – in our experience, something like a third – will not adopt new tooling voluntarily. Ever. Watching this play out confirmed what we believed going in: you do not defeat that resistance with features. The teams that get there do two things, in order. First, make the new path visibly better, not marginally better. Second, leadership makes the new path the default path. Software can only do the first half. Anyone selling you the second half inside a license agreement is lying.
The best adoption story came from the smallest office
The office we expected to adopt last was a one-person operation, an island away from every training session. That office ran the largest program of the quarter through the platform in a single afternoon: coverage checks re-run as quotes were revised, a proposal that put the coverage trade-offs in dollars, e-signature back in under an hour, carrier applications and premium finance following the same week. Adoption is not linear, and your champions are not where the org chart says they should be. Watch the edges.
We were wrong about who leads
Our thesis said producers – the revenue owners – would drive usage. Ninety days of data says CSRs drove it; producers followed, some quickly, some not yet. In hindsight the reason is obvious: the CSR does the mechanical work the platform eliminates, so the CSR feels the difference in their own day. Build for the person doing the work, not the person whose name is on the book. They are rarely the same person, and only one of them fills out forms.
The metric that matters
Weekly submission counts tell you whether adoption is happening. The number we watch hardest is different: are the coverage issues the AI surfaces being reviewed and acknowledged by a human, every time? A platform that gets used but not read is a liability with a login page. So far the discipline is holding, because the workflow enforces it: nothing moves to signature until a person clicks through what the machine found. That design decision has aged better than any model upgrade we have shipped.
Ninety days is early. The ramp could flatten, and the resistant third is still resistant. But the direction is set, and the biggest lesson is the oldest one: the gap between how work is described in meetings and how work actually happens at desks is where every useful product decision lives. Go sit at the desks.