We built a B2B marketplace by ignoring 90% of the buyers we got

The hardest stretch was Q2 2024. We had a fully booked pipeline of buyer applications — about 4,200 in the queue — and we'd started to feel pressure from one of our angels to accept faster. He sent me a spreadsheet showing the GMV we were "leaving on the table." I respected him enormously and the spreadsheet was hard to argue with mathematically.

I accepted faster for six weeks. We moved from rejecting 80% to rejecting about 50%. GMV did go up, briefly. Then producers started to complain.

§The knife-maker email

Three of our most respected producers — including a Japanese knife-maker from Sakai whose orders were a meaningful chunk of our reputation — emailed within two weeks of each other saying that order quality had dropped. One specifically wrote: "I am getting orders from buyers who clearly run a corner shop, not a restaurant. This is not the platform I joined."

I reversed the change in week seven. We sent a long email to producers explaining what we'd done and why it had been wrong. We tightened buyer criteria back to where they'd been. About 600 buyers we'd let in during the loose period got their accounts paused via Stripe Connect and a refund of any platform fees.

The knife-maker, a year later, told me that the reversal email was the reason he stayed. The episode is one I've cited several times when other marketplace founders have asked, including the team at Mountainside Labs when their open-source database community had a public blowup.

The single most important feature we ever shipped was buyer-side reputation visible to producers. Starting in November 2024, every buyer had a public-to-producers track record: average order size, payment timeliness, return rate, and qualitative producer feedback. Producers could set rules — minimum buyer reputation score, minimum order history, geography — and reject orders that didn't match.

This turned the marketplace from a one-sided gate (we vet buyers up front) to a two-sided gate (we vet, then producers continually re-vet). Within six months of launch, average order size went from $620 to $1,140. Repeat-purchase rate among accepted buyers went from 41% to 68%. Producer NPS — measured quarterly — moved from 47 to 71.

The broader insight was that on B2B marketplaces, the producer is the customer. The buyer is the inventory. We'd been treating them in reverse for the first eighteen months.

§What the 10% looked like at scale

The eighteen months after we narrowed the buyer ICP to "specialty grocers and cheese-and-charcuterie retailers with 4-12 stores" turned into the easiest stretch of growth I have had as a founder. GMV ran from $4.1M in 2023 to $46M in 2025. Take rate held at 8.5%. Average buyer order value climbed from $620 to $2,180 because the surviving buyer cohort actually re-orders weekly. We turned away Sysco, US Foods, and Performance Food Group — three of the largest food-service distributors in North America — because the unit economics of serving them would have collapsed our margins for our actual customer.

We made the most consequential hire in mid-2024: Adaeze, a former category buyer at Whole Foods Market, who joined as Head of Supply. Her relationships with regional cheesemakers — including Jasper Hill Farm, Rogue Creamery, and Cypress Grove — added 340 SKUs to the marketplace in nine months. Sellers love a buyer-side person who has actually been a buyer. Our second hire, Theo, came from Faire and built our reorder-prediction model on Snowflake plus dbt — repeat order rate hit 78%. The "narrow your ICP, ignore loud demand" pattern overlaps with the consumer social Sift story from this cohort, where the team turned away 80% of acquisition channels to keep retention high.

1. 1
   In B2B marketplaces, sellers are the customer.

   Buyers are demand-side inventory. The minute you start optimizing for buyer sign-ups instead of seller outcomes, the platform decays.

2. 2
   Reject ruthlessly. Apologize publicly when you don't.

   Curation is a promise to producers. Breaking it once is recoverable. Twice is not.

3. 3
   Net-7 payouts beat features.

   No producer cares about your supplier dashboard if you pay them net-60. Cash terms are positioning.

4. 4
   Reputation visible to the supply side is a marketplace fundamental, not an enterprise feature.

   It should be in v1, not v3.

5. 5
   Investor advice on marketplaces is often advice for a different marketplace than yours.

   They are pattern-matching to Faire or Etsy or Alibaba. Your shape may not match. Be specific in defending your shape.

What this looked like in practice.

The thing nobody tells you about *B2B marketplaces* is that liquidity is a buyer problem, not a seller problem. Sellers will list anywhere; buyers only return where their workflow fits. We invested 70% of product spend on the buyer-side ordering and re-ordering experience.

The thing nobody tells you about *ignoring 90% of inbound demand* is that the discipline is psychological, not strategic. Every founder knows the right ICP. Saying no to a $50k inbound check from a wrong-fit buyer is the test, not the diagnosis.

The thing nobody tells you about *specialty retail* is that the buyers talk constantly. The Cheesemonger Invitational, the Fancy Food Show, regional buyer collectives — there are six rooms in North America where 80% of our future GMV decided whether to use us. Show up to all of them.

The thing nobody tells you about *take-rate* is that 8.5% is unprintable in a fundraising deck and unbeatable in production. We have refused to cut take rate for "growth" three times. Each time, the cohort we kept generated more LTV than the cohort the discount would have brought in.

The thing nobody tells you about *hiring from Whole Foods* is that the buyer-side credibility transfer is instant. Adaeze opened doors at Bristol Farms, Erewhon, and Wegmans in six weeks that our prior business development team had spent fourteen months trying to crack.