Every gate we run throws away companies we would have been glad to talk to. That is not a defect on a roadmap somewhere, it is the price of the method. B2B prospect qualification that never costs you a good company is not qualification at all, it is a search box with better manners. What follows is the honest bill for each of our four gates, taken from the first quarter of 2026: 12,410 Read at the top, 388 Qualified at the bottom.
The order is not arbitrary. Each gate runs where it saves the most work for the one behind it, so the cheapest and broadest cut goes first and the narrowest goes last. Nothing here reorders itself per tenant.
01Gate 1: are they a buyer, or do they just say the word
The buyer test asks one thing: is data the product, or a side activity. A company whose cost of goods is collection has a budget line that grows when collection gets harder. A company that merely mentions collection somewhere has an inconvenience, and an inconvenience does not sign a contract.
This gate turned down 10,818 of 12,410 in the quarter, which is 87 percent of everything we read. It catches four groups: vendors who sell collection themselves and are therefore competitors, agencies that bill time rather than volume, resellers who pass someone else's data through and carry no cost of goods at all, and the very large category of companies where a job advert mentioned scraping once and the actual business is something else entirely.
What it wrongly throws away is specific and we can name it. Companies mid-pivot whose public site still describes the business they are leaving. Companies young enough that the site is a waitlist form. Companies writing in a language where our reading is weaker and our confidence is therefore lower, which biases the whole gate toward English-language businesses in a way we have not solved.
We keep it because the alternative is worse than a false rejection. A message to a company that has no budget line for this is not neutral, it costs their reader a minute and it costs us the only clean attempt we will ever get at that domain.
02Gate 2: is there a reason to write this week
The timing test asks whether something is true today that was not true a while ago. A hiring advert for a data engineer who will own collection. A competitor SDK appearing in a public repository. A new marketplace or region announced in the last few weeks. A status page that has been apologising about collection failures. A funding round that changes what they can spend.
It turned down 988 of the 1,592 companies that cleared Gate 1. Every one of those was a real buyer by the first test, which makes this the most expensive gate we run in terms of good companies discarded.
It is also the gate we are least confident in, and we would rather say so plainly. It is biased toward companies that publish. A business with a genuine collection problem that never writes about it, never hires publicly and keeps its repositories private will be turned down here every single time, and it may be the best prospect in the whole quarter. We know the bias exists. We have not found a way to close it that does not involve guessing.
We keep it because without it the message has no first sentence. A message that opens with a specific thing the reader recognises about their own week is a different object from a message that opens with a description of us. Removing this gate would not produce more replies, it would produce more messages that sound like every other message they receive.
03Gate 3: is there a Door
A Door is a usable public contact route that the company itself published and intends strangers to use. A contact form on their own domain, a published enquiries address, a support route that reaches a human. No Door, no Knock. It turned down 183 companies in the quarter.
Those 183 include some of the highest-scoring companies of the entire quarter. Several were textbook fits: data is the product, the targets are hostile, the timing signal was loud and recent. We turned them down anyway, and we want to be clear that this is a choice rather than a limitation we are waiting to fix.
The obvious workaround exists and we refuse it. We do not run email-finding tools, we do not guess address patterns from a name and a domain, and we do not read the personal profiles of the people who work there in order to manufacture a route. Any of those would clear Gate 3 tomorrow. All of them turn a message the recipient asked to be reachable for into a message that arrives somewhere they never advertised, and that is precisely the behaviour that made outbound something people block on sight.
- A CAPTCHA on the form is not a Gate 3 failure. The Door exists, so the job parks in needs_review for a person. We do not solve or bypass bot detection, and that is the designed behaviour rather than a fallback we are apologising for.
- A contact form that silently fails is a Gate 3 failure the second time we see it, and the domain is suppressed rather than retried.
- A shared inbox published on the company's own site is a Door. An address inferred from a pattern is not, no matter how confident the inference.
We keep it because the Door is what makes the whole thing defensible. Everything else we do could be described as clever research. Gate 3 is the part that says the recipient chose to be contactable, and losing 183 good companies a quarter is what that sentence costs.
04Gate 4: does the jurisdiction allow the approach
The permission test asks whether an unsolicited business approach is acceptable in the recipient's jurisdiction without prior consent, and it is deliberately conservative. It turned down 33 companies in the quarter, the smallest cut of the four by a wide margin.
Thirty-three sounds trivial until you notice which ones. This gate holds back whole jurisdictions rather than individual companies, so a bad ruling here does not cost us one prospect, it costs us a country. That asymmetry is the reason the default leans toward turning down rather than proceeding.
What it wrongly throws away is real. Some of those 33 would have welcomed the message, and in a few cases a more permissive reading of the same rules would have let it through. We are not offering that reading as legal advice and we are not going to pretend the position is settled. The setting is per tenant, it is visible, and a tenant who has taken their own advice can move it.
We keep it because the downside is not symmetrical with the upside. The upside of a message that should not have been sent is one conversation. The downside is a complaint that attaches to a real named person at a real address, which is how we send everything, on purpose.
05Why B2B prospect qualification has to hurt
Three of these four gates reject companies that would have bought. That is not a rhetorical flourish, it is arithmetic from the quarter: 988 real buyers turned down on timing, 183 turned down for want of a Door, 33 turned down on jurisdiction. Twelve hundred companies, roughly, that a more relaxed system would have written to.
The reason we sleep at night is that we tested the relaxed version and it was worse, on a real month, with real numbers we published. Loosening one gate produced more delivered messages and fewer total replies. The gates are not a moral posture. They are the cheapest thing we found that works.
The other reason is auditability. Every one of the 12,022 turned-down companies has a written reason attached to the gate that stopped it, which means the method can be argued with. A prospecting system that only shows you its winners is telling you nothing you can check.
None of this involves email, which is worth restating because most outbound discussions assume it. Everything here is delivered through the target company's own public contact form, and email sending remains scheduled for Q4 2026. Gate 3 would be considerably easier to pass if we relaxed it for email, and that is one of several reasons we are in no hurry.