Four sentences, from a real PM, timed off actual usage. It got a 400-word product spec back.
We run 21+ AI agents in production at SaaStr, and we’re building SaaStr AI Connect on top of them. That means a lot of API signups: search, enrichment, structured data, email infrastructure, inference, storage. Somewhere north of 30 this year alone.
Exactly one of those companies checked in at all.
Not one that checked in better than the others. One, total. Everybody else ran a sequence, ran a pitch, or ran nothing, and none of them ever asked whether the product actually worked for what I was doing with it.
One near-miss worth naming, because it’s the interesting case. A Coresignal rep did reach out, days after I upgraded my tier. He was smart, and he’d clearly done homework. But he was there to upsell me, not to really find out how the product was going. The trigger was my credit card, not my usage. That’s a good sales motion and it isn’t a check-in, and the difference matters: an upsell conversation tells you nothing you didn’t already know from billing.
Other than that, crickets.
Here is the entire email, sent by Alina on Exa’s product team, four days after I ran my first real batch:
Hi Jason,
I’m on the product team here at Exa. I noticed you signed up and tested Exa.
If you have a one-liner on how we did (or what could have been done better), that would be greatly appreciated (& I’m happy to drop $50 of credits in your account).
Thanks a mil, Alina
Four sentences. One PM. Fifty dollars.
I wrote back 33 minutes later with about 400 words of detailed production feedback, a benchmark number, a feature request, and my throughput and pricing constraints. She replied under three hours after that with two specific endpoint configurations and an honest admission of what they don’t support yet.
That exchange is a better onboarding and CS motion than anything the other 30 companies ran on me. Five things it got right.
#1. It asked for one line, not thirty minutes
The size of the ask sets the response rate, and most vendors get this exactly backwards. “Do you have 30 minutes for a quick call?” is a calendar negotiation with someone who has no idea if you’re worth it yet. It gets ignored, especially by the exact users you most want to hear from, because those users are busy shipping.
“A one-liner” is a 15-second commitment. It’s answerable from a phone, in line for coffee, between meetings.
And the small ask is precisely what produced the big answer. I didn’t write one line. I wrote a spec. Nobody writes a spec in response to a calendar link.
#2. It came from a named human on the product team
Not “The Exa Team.” Not no-reply@. Not an SDR working a 7-touch sequence built around a case study.
Someone who works on the product, using their own name and email, who could actually do something with the answer.
Every vendor on earth can now generate personalized-looking outreach at infinite scale. That’s the whole point of the tooling we’re all buying. Which means the signal value of a real person who can answer a real product question went up this year, not down. Automated outreach got cheap and infinite. Human attention from someone with roadmap influence got scarce.
#3. It triggered on usage, not on signup
“I noticed you signed up and tested Exa.” Tested. The trigger wasn’t account creation, it was the first meaningful batch of API calls.
Most onboarding drips fire on day 0, day 1, day 3, and day 7 no matter what happened in the account. I get “Here are 5 ways to get started” emails for products I already have in production, and I get them for products I signed up for and abandoned in four minutes. Identical sequence. The vendor is telling me they aren’t looking.
Alina hit the one window where my answer was worth anything: after a real test, before a decision. The data was fresh, the opinion was fully formed, and nothing was locked in yet. A week later I’ve either already built on it or already moved on, and either way my answer is worth less to them.
#4. The incentive was instant, small, and denominated in their own product
$50 of credits dropped in the account, with no raffle and no $100 Amazon card arriving after a 45-minute session with the user research team and an NDA.
The marginal cost to Exa is close to nothing, so this runs at volume without a budget fight. And paying a developer in API credits pays them in more usage of the exact thing you want them using more. The incentive and the activation goal are the same object. A gift card buys you an answer. Credits buy you an answer plus another batch of production traffic.
#5. The reply came from someone who knew the product cold
This is where most feedback loops die. You send real, specific, detailed feedback, and you get: “Thanks so much for sharing this, I’ve passed it along to our product team!” That response teaches you never to bother again.
What came back instead, in under three hours:
- The specific configuration that solves my problem today, by endpoint and parameter, including the per-field grounding confidence I hadn’t been using
- A second path I didn’t know about, with structured schema support and citations
- A direct “we don’t yet expose that” on the one thing I asked for that they don’t have, plus a note that it’s interesting enough to look at
- An offer to get on a call and run test examples together to close it out
The “we don’t have that yet” is the part most CS teams are trained out of, and it’s the part that builds the most trust. A clear no from someone who understands the ask is worth more than an enthusiastic maybe from someone who doesn’t. I know exactly what to build around now.
What $50 and one email actually bought
The ask was a one-liner. What they got back was a design partner conversation:
- The use case in production terms. An AI matching product on ~145k B2B and AI executives, where CEOs post roles, candidates get matched, and we broker warm intros. The hard part isn’t the matching, it’s that profile data goes stale constantly, and a recruiting product running on 18-month-old employers is worthless.
- Two concrete jobs. Company enrichment for HQ, funding stage, and headcount across ~3k companies to feed stage-fit ranking. Plus job-change detection sweeping profiles our structured-data vendor returned not_found on.
- A benchmark on the adversarial set. That not_found pool is the worst possible input: misspellings, common names, thin footprints. It hit 89% clean person matching there and surfaced real job moves we’d have missed for months.
- A specified feature gap. We run our own model pass on top to verify identity, because the search and extract split doesn’t tell us “is this the same human.” A match-confidence signal would let us skip that layer on the easy cases. That’s a roadmap item written by a paying user with a use case attached.
- The constraint that made the deal. 10 QPS is what made the batch viable at all. The alternative we tested was roughly 5x the cost and took hours instead of minutes.
- Direction of spend. It’s a scheduled weekly job now, running continuously.
No NPS survey produces that. And a QBR is too late, because by then I’ve already routed around whatever was broken.
One email did three jobs most companies split across three teams
Activation check. Did the user get to value, and if not, where did they stall. Usually owned by growth, usually measured with a dashboard instead of a question.
Product research. What should we build next, from someone with a live production workload. Usually owned by product, usually run as a scheduled interview cycle two months behind the market.
Expansion. Turn a test into recurring spend. Usually owned by CS or sales, usually starting after the contract, which is after the architecture is already frozen.
Three orgs, three timelines, usually three quarters. Or one PM and $50.
What the other 30 did instead
Not one of them asked a question. The full range:
- Nothing at all. By a wide margin the most common outcome.
- An automated drip: “5 ways to get started,” “Did you know,” links to quickstart docs I’d finished reading before the first email landed.
- An SDR asking to book a demo of the product I had already implemented and was already paying for.
- A sharp rep who appeared days after I upgraded a tier, to sell me the next one. Triggered by revenue, not by usage.
- A CSM assigned after contract signature, checking in monthly on adoption, at the exact point where nothing I say can change what got built.
The version to steal
- Trigger on first meaningful usage, not on signup. If they never made a real call, that’s a different email.
- Send from a named human on the product or engineering team. Personal address, no template chrome, no logo header, no unsubscribe footer stapled to a one-line note.
- Ask for one line. Say the words “a one-liner.” The explicit smallness is the mechanism.
- Attach a small instant incentive denominated in your own product. No hoops, no call required.
- Reply personally within hours, with specifics, including a clear no on what you can’t do.
- Offer to debug live, as an option and not a condition.
A PM can run 20 of these a week in a couple of hours. Do that for a quarter and you have 250 conversations with people who just touched the product for real, plus a feature backlog written by paying users, plus a set of accounts where a human being with roadmap influence knows your name.
It Compounds
Four days later I’m writing this post about them, and it’ll be read by a lot of B2B founders who buy APIs.
They didn’t engineer that. It’s just what happens when a company treats the first week of a new account as a conversation instead of a sequence. The cheapest product research available in B2B right now is one honest question, sent by a real person, at the moment someone has just finished testing your product and has an opinion nobody has bothered to ask for.
Almost nobody sends it.
