How we sold AI underwriting to a 200-year-old insurance carrier

The eighteen-month sales cycle nearly ended us. By month twelve — September 2024 — we'd burned roughly $3.1M of the seed and had four months of runway left. The carrier kept asking for additional pilots in adjacent lines of business. Each pilot was unpaid. Each pilot consumed eight engineering weeks. Their procurement team had a 23-page security questionnaire that we filled out three separate times because each version was reviewed by a different department.

In November 2024 their head of underwriting called and told me, very politely, that the deal was being deprioritized to Q3 2025. I was on a hike in Sausalito when I took the call. I sat on a rock for an hour afterwards.

§The cofounder fight that didn't break us

That night Idris and I had the worst conversation in the company's history. He thought we should pivot to MGAs. The MGA path would close $400k deals in 90 days. It would also turn us into the company we had explicitly chosen not to build. I was paralyzed.

What saved us was a single phone call. Our advisor — a former chief underwriting officer at a top-five carrier — picked up at 9pm and listened to me describe the situation. Then he said: "She told you Q3 2025 because someone above her told her to push it. The actual decision is being made above her by a person who has not yet seen your product. Find that person." He gave me a name.

I cold-emailed the name on a Wednesday. We had a forty-minute call the next Tuesday. The deal closed five months later. The framing — that the named buyer is sometimes not the actual decision-maker — was the same insight I'd later see in Switchboard's procurement-classification fight with their Fortune 500 customer.

The breakthrough was identifying that the carrier's procurement timeline and the carrier's actual decision timeline were two different things. The 23-page security questionnaire wasn't gatekeeping the deal — it was the cover story for a more political conversation happening inside their executive team about whether AI in underwriting was a 2025 priority or a 2026 priority.

Once I understood that, I changed our entire approach to enterprise sales. We started writing what we called "executive briefings" — six-page documents tailored to a specific named executive, anticipating their political constraints inside the carrier, and proposing a deal structure that gave them a clean win. We did not send these through procurement. We sent them directly, by email, to the named executive.

The technique worked five times in 2025. Average deal size: $3.2M. Average sales cycle: 11 months, down from the initial 18.

Idris and I, by the way, are still cofounders. The MGA conversation happened. We didn't pivot. He told me later that what mattered to him in that moment wasn't the strategy decision, it was that I had taken his pivot suggestion seriously enough to lose sleep over it.

§From PoC to production at the carrier

The nine months after the carrier moved Quorum from PoC to a single line of business — small-commercial property — were the most operationally intense of my career. Loss ratio on Quorum-underwritten policies came in at 41.3% versus the carrier's book-wide 58.7%. Quote-to-bind time dropped from 11 days to under six hours. The carrier added a second line, then a third. By April 2026 we were underwriting roughly 14% of their new commercial premium.

We made two hires that decided whether we survived production. The first was Daniyar, a former Munich Re actuary who joined as Head of Risk Modeling — his existence ended every "but how do you handle adverse selection?" conversation in carrier meetings inside fifteen minutes. The second was a former Verisk product lead, Marisol, who built our integration layer to Guidewire, Duck Creek, and Majesco — the three policy-admin systems that decide whether a carrier can adopt your product at all. Total team is now twenty-three people; ARR crossed $9.2M in March 2026. The "domain credibility beats AI" pattern shows up in the synthetic-biology story from this cohort, where Onyx Labs hired ex-Ginkgo scientists ahead of more ML engineers for the same reason.

1. 1
   MGAs are a tempting trap.

   They will close fast and pay you small. The product roadmap they pull you toward will diverge from the one that wins carriers. Pick the harder path early or you can't get back.

2. 2
   Procurement timelines and decision timelines are different documents.

   The first is what your champion shows you. The second is what's happening in an executive team three floors above your champion. Find that team.

3. 3
   Write for the named executive.

   A six-page document aimed at one person, by name, anticipating their political constraints, beats a fifty-slide deck written for "the customer."

4. 4
   Insurance pilots are not free.

   Each unpaid pilot consumes engineering weeks you cannot recover. Charge for pilots — even nominally — as a forcing function on commitment.

5. 5
   Your cofounder's "let's pivot" is sometimes a request for you to take their fear seriously, not a strategy proposal.

   Treat it as the emotional event it is. Then make the strategy decision separately.

What this looked like in practice.

The thing nobody tells you about *selling to a 200-year-old carrier* is that the cycle time is not the product of bureaucracy — it is the product of regret. Every senior person in the building has watched a vendor over-promise and has scar tissue. The slowness is rational. We learned to absorb it instead of fighting it.

The thing nobody tells you about *AI underwriting* is that the regulators care about model-explainability artifacts more than model accuracy. Our SHAP value reports for every declined policy were what got us through the NAIC model regulation review, not our loss-ratio numbers.

The thing nobody tells you about *hiring a former Munich Re actuary* is that they cost twice what your investors think they should cost and are worth four times that. The credibility transfer in customer meetings is immediate.

The thing nobody tells you about *Guidewire integrations* is that they are not technically hard — they are politically hard. The carrier's IT organization has to accept your product running inside their PolicyCenter instance. We spent eight months on architectural review for what was, technically, a four-week build.

The thing nobody tells you about *the line-by-line expansion motion* is that each new line of business is a new sales cycle internally, even after you are in the carrier. Treat every line as a fresh logo and the math gets honest.