The insurtech graveyard is full of technically impressive products that nobody uses. The pattern is consistent: a team builds something smart in isolation, sells it to an agency principal, and then watches adoption flatline because the people who actually do the work — CSRs and producers — never asked for it, don't trust it, and have workarounds that feel faster even when they aren't.

We've taken a different approach. Before we wrote a line of production code for Compass, we spent months physically inside a brokerage — sitting next to CSRs, watching producers quote, timing every workflow, and mapping every pain point. The lessons that came out of that process shape how we build today.

Lesson 1: The Problem Isn't What Leadership Thinks It Is

When we talk to agency management, the priorities are predictable: faster quoting, better reporting, AI that can handle phone calls. All reasonable. But when we sit with the people doing the actual work, a different picture emerges.

CSRs aren't slow because they lack motivation — they're slow because they're toggling between data portals, manually copying property data into replacement cost forms, and re-keying the same fields across multiple carrier systems. Producers aren't missing coverage gaps because they don't care — they're missing them because no system surfaces the comparison between what's expiring and what's being quoted.

The first lesson: build for the workflow you observe, not the workflow that gets described in a meeting room.

Lesson 2: Adoption Requires a Better Path, Not Just a New One

This is the uncomfortable truth that most SaaS companies avoid. If users can still do the job the old way, a meaningful percentage will — especially in insurance, where average staff tenure is measured in decades and change resistance is cultural.

The teams that successfully roll out new tools tend to do two things in sequence. First, they make the new path observably better than the old one — not marginally better, but visibly faster or more accurate from the first use. Second, they consolidate access so that the new path becomes the natural one, with leadership alignment on what the standard workflow looks like.

The order matters. If you force people onto a worse tool, you get resentment. If you make a better tool the natural path, you get adoption that sticks.

Lesson 3: The Adoption Curve Tells You Everything

When Compass rolled out, we watched the submissions count weekly. The early weeks were measured. The team was learning the interface, asking questions, finding rough edges. Our team was present in person to answer every one.

Then the growth happened. The CSRs who started in week one were now helping train the next wave. New offices came online. By the time the system was running at full cadence, weekly submission volume had multiplied many times over.

That growth wasn't driven by a marketing campaign or a mandate from the top. It was driven by the tool being observably better than what it replaced. When a CSR who used to spend 20 minutes filling a DRCE sees it auto-populate in seconds, they tell the CSR at the next desk. That's the only adoption engine that actually works.

Lesson 4: AI Has to Earn Trust Through Transparency

Compass includes an AI coverage analysis engine that reads policy documents and flags gaps. The metric that matters isn't how many analyses it runs — it's the rate at which staff actually review and engage with what the AI surfaces.

That engagement only happens when the AI's output is legible. Every flag requires a human acknowledgment. Every coverage check produces a detailed analysis that the CSR or producer can read, question, and override. The system never makes decisions for them.

This is the balance that most AI products get wrong. They either automate too aggressively — making decisions that humans should make — or they're so passive that they become notification noise. The sweet spot is an AI that does the research a human doesn't have time for, then presents its findings in a way that makes the human's judgment call faster and better-informed.

What We'd Tell Other Founders

If you're building for insurance — or any industry where the end users didn't choose your product and have 20 years of muscle memory working against you — here's what we'd say:

Go sit in the office. Not for a day. For months. The gap between how work is described and how work actually happens is where every good product decision lives.

Build for the person doing the work, not the person signing the check. They're almost never the same person, and their priorities are different.

Make the new way better before you make it the default. In that order.

And measure adoption weekly. If the curve isn't climbing, nothing else matters.