A buyer in the room: how a seed-stage API vendor packaged and sold to large enterprises

• The company’s largest contract to date, signed in the autumn of 2026 with a large platform customer after a blind bake-off against another vendor.
• Pricing moved from a flat per-call rate to committed monthly tiers, billed monthly on an annual commitment, with trials that have written success conditions and an open-ended master agreement.
• The identity use case repositioned as fraud prevention rather than a background check, with product-page briefs written for four buyer segments.
• An ideal customer profile with the traits that qualify a buyer, and a working account of how fraud teams buy, from the cost-centre lull to the week-long sprint.
• A second platform account grew several times over in volume and moved toward a committed contract, with applicants the product passed clearing the platform’s own screening at a markedly higher rate than the unscreened stream.

The client is a seed-stage startup with an AI research API. It researches a person or a company from a few identifiers and returns a structured profile, and it was first sold to sales teams, to find and qualify leads. A handful of marketplaces and workforce platforms had started using it for something else: checking whether the person behind a résumé exists.

By spring 2026 the business was growing fast on usage-based revenue, and the founders knew what their investors would ask next: how much of it is recurring? The team was technical and shipped quickly. Nobody on it had sat on the buying side of a fraud or trust and safety team, negotiated an enterprise data contract, or been through a large company’s vendor procurement.

We had. Brendan ran the identity and trust and safety program for the contributor workforce at Scale AI, bought and ran the checks these teams buy (identity verification, phone and device signals, behavioural signals, résumé claims) and negotiated the contracts for them. Dirk built the fraud models those signals feed. The client didn’t need a sales team or an agency. It needed a buyer in the room when the founders decided who to sell to, what to charge, how to run a pilot, and what the output had to look like for a fraud analyst to act on it.

That became a monthly retainer: a standing call with the founders, every two weeks at first and monthly once the pipeline was moving, the questions in between answered in a shared chat channel, and introductions where they fit, made without any fee.

The first problem was the words. The site said data enrichment for revenue teams, and its best new customers were buying something else. Together we repositioned the identity use case as fraud prevention rather than a background check. A background check is slow, costs ten dollars or more, is regulated, and verifies what a candidate claimed. This product answers a different question, in minutes and for a fraction of that: is there a real person with years of digital history behind this résumé, or a profile that started three months ago?

That positioning does three things. It names the gap every platform in the segment feels, between a free profile link anyone can fake and a background check nobody wants to run on every applicant. It keeps the product on the fraud-prevention side of the line that the rules on hiring decisions draw, a line each buyer’s counsel has to confirm for their own use. And it tells the buyer where it goes: as a first pass before the expensive check, not instead of it.

We wrote the briefs for the product pages that followed: who each page is for (platforms onboarding contractors, companies hiring remote workers, staffing firms, and background check providers who need a faster data source), the problem in the buyer’s words, what the API checks, how it compares with the incumbent process, and the proof to show. And we gave the advice a buyer gives. Keep the self-serve sign-up, because that is how a trust and safety lead tries a tool at ten at night. But make the pages speak to the problem the buyer searches for, because that buyer is looking for “résumé fraud” or “phone number fraud”, not for a platform.

The second problem was who to call first. The founders had conversations open across banks, marketplaces, AI data vendors and government. We worked out the ideal customer profile from the buyer’s side: a handful of buckets within trust and safety, and the traits that make a company a fit. The work is high value, so the incentive to fake an identity is real. Onboarding is automated, so fraud arrives at scale. The company is growing fast, moving into new countries or launching new lines, so the problem is new and so is the budget. Two companies on the list were ruled out on that test, and the list got shorter and better.

Then how those buyers buy, which is the part a product-led team never sees. A fraud team is a cost centre. When things are quiet it minimises spend and moves slowly; when something goes wrong it signs in a week. Most of the long-running conversations in the pipeline were waiting for that moment, and the job was to be there when it came: regular check-ins so the vendor isn’t the one that gets deprioritised, a direct question about the decision process and who signs, and care not to spend months with a champion who still needs a vice-president’s approval. The big companies would also run vendor onboarding, with security questionnaires and certifications, before anyone could call the API in production, so the onboarding checklist had to be on the table before the contract, not after.

For the bank track, Brendan walked the team through how a fraud or anti-money-laundering investigation runs: the automated rules that open a case, the packet an investigator gets, the flowchart they follow to a conclusive red or green flag, and why a green flag that clears an international student’s tuition wire is worth as much as a red one. That told the founders where the product fits, as a data source inside the investigator’s case rather than a replacement for the case management system, and which buyers have money on the line.

The third problem was price, and it was the one the founders asked about most. The list price was a flat per-call rate. Prospects said it was high. It wasn’t, against a background check at ten dollars or more, or an identity check at a dollar or more, but a flat price per call with no commitment gives a buyer nothing to negotiate and gives the vendor nothing recurring. We restructured it the way the contracts we had signed as buyers were structured.

We also looked at the cost side. The agents ran on a frontier model, which left little room under the list price; an open-weight model cut the cost substantially, which opened room for a lower price at high volume without giving up the margin.

• Anchor high and let the buyer push back. Don’t offer every discount up front; keep them for the negotiation, and ask whether the buyer cares about unit cost or total spend, because different teams watch different numbers.
• Tiered per-call prices for monthly commitments, so a bigger commitment buys a lower unit price and the vendor gets recurring revenue.
• An annual commitment billed monthly, rather than one large payment up front. A monthly invoice clears a lower approval threshold, renews without a finance review, and is easier to grow.
• Credits issued monthly and expiring, overages billed at the same rate, and never a service cut-off when the credits run out.
• A trial with a success condition agreed in writing (“as long as the trial shows this, it continues”), ramping commitments over the first quarter, and automatic rollover into the standard contract so nothing needs a second signature.
• A letter of intent with two tiers for the buyer to react to, and an open-ended master agreement so the next product can be switched on without a new legal process.

In the summer a large, fast-growing platform customer came back to the table after months of quiet, with a competitor in the running and a request for a blind bake-off: a couple of thousand queries over two weeks, scored by the platform’s own quality team on precision and recall. The platform hadn’t set its benchmarks yet. That was the opportunity, and the playbook had five moves.

The contract was signed in the autumn of 2026, the client’s largest to date.

• Guide the buyer to the number that reflects what they want. On a similar evaluation we ran as buyers, the team spent its time checking every link a vendor found, when what mattered was how often the vendor was right on “definitely real” and “definitely fake”. We helped frame a rollup metric around the decision, not the evidence.
• Make the output a decision, not a score: approve, decline or escalate for review, with a risk score and citations, mapped to what the platform’s system would do with each. Confirm the buyer’s appetite for false positives against false negatives with the person who owns the pilot, before the first batch.
• Standardise the inputs. Vendors in a bake-off rarely get the same fields. The team asked for more, and got them.
• Review the data with the buyer twice a week. Disagreements with the platform’s earlier labels were worked through case by case, and some of those labels changed. A buyer trusts a vendor that engages with the buyer’s own labels.
• Lead with quality and let the commercial terms follow. The pilot fee was waived against a cheaper competitor; the anchor on the contract stayed where it was.

A research dossier is a research result until a fraud analyst can act on it, and the distance between the two is where most signal vendors lose the trial. We spent a good share of the calls on the output. Binary signals sell (“fake or not”, “in the country or not”); nuanced ones need a data scientist on the buyer’s side. The reason in plain words next to every flag. False positives explained by segment (degree verification works for large universities and struggles with small or international colleges) so the buyer can fix patterns instead of filing tickets. And hands-on implementation support in the first weeks, because a signed contract that never gets integrated churns at renewal.

On a second platform account the same shape produced measurable lift: applicants the product marked good went on to pass the platform’s own screening at a markedly higher rate than the unscreened stream, and the review tier was tightened so more good candidates went straight through and the operations team reviewed fewer cases. The next step we recommended there is the one buyers ask for next: linking repeat accounts, because a rejected applicant who comes back with a new email, a new phone and a lightly edited résumé wastes an interview slot every time. The client already had a product for linking related accounts, and it became the next bolt-on.

That is the growth pattern we described from the buying side. Land with one product on a credit card. Add bolt-ons until the bill is noticeable. Then bundle into a committed contract, anchored to the sustained baseline rather than the spike. We watched an identity vendor’s account grow from a basic check to a full stack in a year by doing exactly that, from the buying side.

Between March and October 2026 the identity use case was repositioned as fraud prevention, with product-page briefs for four buyer segments. The founders had an ideal customer profile with a handful of trust and safety buckets and the traits that qualify a company. Pricing moved from a flat per-call rate to committed tiers with monthly billing, trials with success conditions, and an open master agreement. The largest contract in the company’s history was signed with a large platform customer after a blind bake-off. A second platform account grew several times over in volume and moved toward a committed contract. Risk leaders at two large banks and a risk-decisioning platform took introductions. And the product roadmap reads like a fraud team’s wish list (repeat-account linking, continuous monitoring, skills matching) because the buyer’s seat was in the room when it was written.

What we didn’t do: sell for them, take a commission or run their pipeline. The founders closed every deal.

A monthly retainer since March 2026. A standing call with the founders and their operations lead, every two weeks through the spring and monthly after that, and the questions in between answered in a shared channel within a working day. Brendan is on every call; Dirk joins when the question is the model or the evaluation. Introductions to buyers are made where they fit, and both sides are told we advise the vendor. We take no commission, finder’s fee or success fee from the vendor or from anyone we introduce them to. When a client of ours evaluates a vendor we advise, we pause that vendor’s retainer until the client has decided, and tell both sides.

Eight things, in the order we would say them on a call.

• Sell the problem the buyer searches for, not the platform you built. The buyer is looking for “résumé fraud”, not for a category name.
• Position below the regulated product, not against it. A first pass for fraud at a fraction of the cost sells faster than a better background check.
• Qualify on incentive, automation and growth. High-value work, automated onboarding and a company in hypergrowth is a buyer with a problem and a budget.
• Fraud teams buy slowly until they buy in a week. Stay in touch so you’re the vendor they call.
• Price for commitment, bill monthly, and never cut the service off. Anchor high, keep discounts for the negotiation, and write the trial’s success condition down.
• In a bake-off, shape the metric and own the review. The buyer’s labels are wrong more often than they think.
• Ship a decision, not a score, and support the integration. The contract you win is lost at renewal if nobody integrated it.
• Get the master agreement open-ended. The second product should be a line item, not a legal process.

If you sell an API or a data signal into fraud, risk, trust and safety or compliance teams, your revenue is usage-based, and your investors are asking about contracts, this is the seat we take. It starts with a free thirty-minute call. Bring your pricing page and the deal that is stuck, and we’ll tell you how the buyer on the other side is reading both. Book the free call below.