Before expanding the list: one filter only your delivery record produces
Before you add volume to a prospect list, pick one fact you actually observed while delivering work and make that your filter. What makes a filter good is not that it is unusual, but whether you can say in one sentence why a company meeting the condition has someone spending budget on it now.
Contents
The answer: before adding volume, pick one filter only your delivery record can produce
A&A perspective
Before you add volume to a prospect list, pick one fact you actually observed while delivering work, and make that your filter. What makes a filter good is not that it is unusual. It is whether you can say, in one sentence, why a company that meets the condition has someone spending budget on it now. A filter that fails that test has narrowed the list without narrowing it toward a reason to buy, so what you end up with is simply an unusual list.
A&A perspective
The fact you picked is not yet a filter. You have to bring it down to a form you can judge before you send, using only information the other company publishes; write the spending reason as one sentence; and decide how you will enumerate the companies that match. Only when all three are in place does adding volume mean anything. Increase volume while one of the three is missing and, when the replies do not come, you cannot separate a wrong filter from a weak message from bad timing on their side. A failure you cannot separate does not turn into a next move, however many times you repeat it.
A&A perspective
This article does not argue for sending less. It argues about order. Nor does deciding on one filter mean turning away companies that do not meet it. Referrals arrive from outside any filter, and inbound enquiries are not something you sort by this criterion either. The scope here is narrow: how you build the list you reach out to. And if you have delivered only a handful of projects, across industries that do not yet line up, there is no population to extract a filter from, so what comes first is conversations with customers, not filter design.
What the source says: filter on size, industry and title alone and everyone sells to the same companies at once
From the sources
An article published on 22 April 2026 by Clay, a company selling a go-to-market data platform, written by Varun Anand and Mishti Sharma, states that when you and your competitors target only on company size, industry and job titles, "you're all going to be selling to the same companies at the same time". The same article treats accurate, comprehensive and timely data as table stakes and says that "Unique data, however, is your competitive edge", adding that "Unique data gives you a precision edge". It describes such data points as hidden signals competitors cannot see because they do not know to look for them.
A&A perspective
The problem does not disappear when you change the tool that builds the list. Whatever appears as a field in a company database appears there for everyone who can buy the same database. Adding fields you can extract with AI does not change this either: as long as what you are extracting is a rearrangement of public information, anyone using the same tools can reach the same rearrangement. If a difference exists, it is in the judgement about which data point to extract in the first place.
A&A perspective
For a solo or two-person services firm, this bites earlier than any question of beating competitors. If everyone you reach happens to sit inside the same loose condition, then when the replies do not come there is almost nothing left to read. The broader the condition, the more varied the situations of the people who received it, and the more the reasons for passing differ from one to the next. You can send dozens of messages and still not know what to change. Settling on one filter is less about winning than about making the result legible.
| Aspect | A filter anyone can build | A filter only delivery produces |
|---|---|---|
| Where it comes from | A field that exists in a company database | An event on the customer's side, seen before the work began |
| What it narrows | What the company looks like (size, industry, title) | A state in which spending becomes due on their side |
| Judging it before sending | Can be judged as it stands | Takes work to bring down to a proxy |
| When no reply comes | Filter, message and timing cannot be told apart | Separable by whether replies touch the original event |
| Relation to competitors | The same condition reaches the same companies at once | Invisible to anyone not thinking to look for it |
An unusual condition and a condition with a reason to buy are not the same thing
From the sources
The same article lists examples of filters its customers actually use. It records that "Certemy counts OSHA violations to find companies with compliance problems", that "Supermetrics distinguishes between brands and agencies to route leads to the right sales teams", and that "Cake.ai identifies AI engineering teams that have 2-5 developers (their sweet spot)".
A&A perspective
Put them side by side and it is clear that none was chosen because the attribute is rare. If violations have accumulated, that is a live problem inside the company right now. Whether a company is a brand or an agency changes how the buying decision is made and where the budget comes from. Two to five developers describes a specific kind of bind: too small for a dedicated owner, too large to keep doing it by hand. In every case, what the condition points at is not what the company looks like but a state in which spending becomes due on their side.
A&A perspective
So the question to ask of a candidate filter is not "which field are my competitors not using". It is "when this condition holds, what is happening on their side". A filter that answers that question usually does turn out to be one competitors are not using, because events on the customer's side are not sold as database fields. Unusualness is not the thing to aim at first; it arrives afterwards, as a consequence of having written the reason to buy first. Reverse the order and what you are left with is a condition that is unusual and that nobody is troubled by.
The second source: a filter does not work until it reaches who decides to pay
From the sources
A Growth AMA by Y Combinator partner Gustaf Alströmer, published by Y Combinator on 29 October 2018, answers a B2B SaaS founder asking about lead generation by advising them to "Figure out who is the actual buyer of your product. Who is the decision-maker that will pull out the credit card / sign the bill?" It then describes putting a hundred LinkedIn profiles matching that buyer into a spreadsheet, finding their email addresses and writing to them individually.
From the sources
Asked whether to go after large corporations or mid-sized companies first, the same AMA says it is nearly always easier to start with smaller ones. The reason given is to "Start with companies that move fast and need your product but don't have a 'procurement team' yet" — that is, a company where the person who likes and uses the product can decide to pay for it themselves. It adds that you want companies that can afford the product and have a lower amount of internal friction in getting it done.
A&A perspective
The second piece of advice reads as an example of a condition built on how spending arises, not on an attribute. "No procurement team yet" is neither an industry nor a size; it is a statement about who inside the company decides to pay. Translated to a small services firm, the condition becomes "companies where the person who uses the work and the person who approves the spend are the same person". That distinction itself is covered in "When an AI services deal stalls: separate the user from the buyer", but there it is a check you run once a deal has started. To put it into a list filter, you have to be able to judge it from outside, before you send anything.
Where the filter comes from: with no sales team to interview, the population is your own delivery record
From the sources
Clay's article says the first step in deciding which unique data points you need is to understand your customers deeply, and advises you to "Interview your sales team about what they manually research before calls and why those signals matter". It also says to "Study your best customers to know what unique data points to look for".
A&A perspective
A solo or two-person firm has nobody to interview. The person doing the manual research before a call is you. The substitute population is the record of work you have delivered. This is not something the source states; it is A&A's transposition of the source's procedure to a company of a different size. The transposition holds because what the original step is trying to extract is not a salesperson's opinion but a fact — what was actually being looked at before the deals that closed. In a one-person company, that fact lives in your own records.
A&A perspective
The procedure runs like this. List the last five to ten pieces of work you delivered, as far as you can recall them. For each, write one line about what was happening on the customer's side before the work started: a staff member resigned; the number of trading partners grew; an audit raised a finding; they brought in another tool and could not keep it running. The point is to write the event on their side, not the fee or the industry. Write the event in the words the customer used themselves. Rewrite it into your own tidied-up phrasing and you lose the ability to notice that the same event also occurred in another project.
A&A perspective
List five to ten and one or two events will recur. Those are your candidates. An event that appears only once cannot yet be told apart from coincidence. Note that what you have at this point is not a filter but an event on the customer's side. Such events are normally invisible before you make contact. The next section brings the event down to something you can judge before sending. An event you cannot bring down that far has to leave the candidate list, however promising it looks.
Three tests that turn a candidate into a filter, and one disconfirmation
A&A perspective
The first test is whether you can judge it before you send. Using only what the company publishes, or material you can feed to a tool, can you decide whether it meets the condition? If you cannot, bring it down to a proxy you can judge. The proxy does not have to point at exactly the same set of companies as the original event. But you do have to be able to say, in your own words, in which direction and how far it drifts. Drift you cannot describe is what bites later, when you try to read the results.
A&A perspective
The second test is whether it has come down to a reason for spending. Write one sentence: when this condition holds, such-and-such is increasing on their side, so somebody's budget now pays to move it outside the company. If the subject of that sentence is what you offer, rewrite it. The only thing that should appear in it is their situation; nothing you can do belongs there. A condition that will not yield such a sentence has narrowed the list without narrowing it toward a reason to buy.
A&A perspective
The third test is whether you can count them. Using a route available to you today — an industry association roster, job postings, published lists of trading partners, a municipal register of businesses — can you enumerate companies that meet the condition? If you find ten, that is not a list filter; it is a list of companies to approach individually. In that case, do not build a sending system; go to the ten directly, which is faster. Designing a filter and automating outreach are separate decisions.
A&A perspective
Once a candidate passes all three, look for disconfirmation. Among the customers you have delivered to, is there one that met the condition and did not buy, or one that did not meet it and bought anyway? If so, the filter is still coarse. A coarse filter is still usable, but using it while knowing it is coarse changes how you read silence afterwards. And if no counterexample appears at all, suspect that the filter is merely tracing your own past rather than pointing at anyone new.
A hypothetical: a two-person services firm compares two candidate filters
Hypothetical example
As a hypothetical, consider a two-person services firm that modifies business systems and automates PDF paperwork. Work has arrived through referrals until now, but the contact at the referring company moved on and the flow thinned. So for the first time they decide to build an outbound list. Two candidates emerge.
Hypothetical example
Candidate A is "manufacturers with 50 to 200 employees". It passes two of the three tests: you can judge it before sending, and you can count them, both from public information. But the spending sentence will not come. You can get as far as "a manufacturer with over 50 staff probably has a lot of clerical work", and then nothing explains why the spend arises now. Try to make their situation the subject of the sentence and there is nothing to write. Candidate A drops out here.
Hypothetical example
Candidate B comes from the delivery record. In four of the last seven projects, the state before the work started was: the number of trading partners had grown, and someone was re-keying PDF forms whose layout differed per partner. The spending sentence writes itself. Each new trading partner adds one more layout, and re-keying time grows in a way that does not scale with headcount, so once partners increase, a decision arises between hiring clerical staff and sending the work outside. But there is no way to judge this state from outside before sending.
Hypothetical example
So they bring it down to a proxy: "companies whose careers page is advertising a clerical role whose description contains order-processing or EDI". Job postings are public, so the proxy can be judged and counted. The drift can be described too. Companies in the same state that are not hiring will be missed, so it drifts toward under-inclusion; and some that are hiring are replacing a departure rather than absorbing more re-keying, so false positives are mixed in. Knowing both, they send to companies matching the proxy and then look not at how many replied but at whether the replies touch the original event. If not a single reply touches it, they conclude the proxy drifted too far.
A&A perspective
What does the work in this example is not that candidate B is unusual. It differs from A in that it fixes, in advance, what you should suspect when the replies do not come. Rejection under A tells you nothing about whether the filter, the message or the timing was wrong. Under B, the way of reading — what the replies touch — was decided first, so whatever you send becomes material for the next decision. That is the practical return on settling for one filter.
What the sources do not say, and where this order does not apply
From the sources
The same YC AMA says, of this cold email motion, that "The biggest mistake people use when doing this is sending too few emails." The sentence sits in a passage about writing to a hundred profiles and, if it works, repeating the same thing twenty times to reach a hundred users.
A&A perspective
This is in tension with the argument here, which says to decide the filter before adding volume. It does not, however, say to send less. The reading that reconciles the two is "decide the filter, then go to volume" — but that reconciliation is A&A's reading, not something either source states. If you are unsure which to put first, we think the practical split is whether you have delivered more than ten projects. Below ten there is no population to extract a filter from, so going to volume to learn who your buyers are is faster. That too is A&A's judgement, not an instruction from the sources.
A&A perspective
The limits of the sources themselves are worth stating. Clay's article is written by the vendor selling the go-to-market product it describes; it is not an independent study or an effect measurement, and no population or method is published for any of the customer examples it lists. Its assumed reader is an organisation with a dedicated go-to-market team, not a solo or two-person services firm. The YC AMA assumes US startups in 2018, and the motion it describes is largely about getting people to try a product. Neither source claims any of this transfers unchanged to the Japanese contract-services market in 2026. Neither establishes that filtering on a unique condition causes more won work.
A&A perspective
There are situations where this order is unnecessary. With only a handful of delivered projects across unaligned industries, there is no population to extract from. And while referrals keep the schedule full and no outbound list is needed, this work stays low priority. The filter applies to the list only; it is not a reason to close the door on referrals, which arrive from outside any filter. Finally, this article publishes no A&A reply rate, win rate or list size, because no measurement we could publish has been carried out.
Next step: list five past projects and write one sentence
A&A perspective
There is one thing to do today. Write down the last five pieces of work you delivered and, for each, one line about what was happening on the customer's side before it started. It takes about half an hour and needs no tools. Do not write their industry or the fee. If you do, you will want to sort by those, and the events will stop being visible.
A&A perspective
If one event recurs, write beneath it the single sentence explaining why spending arises, with their situation as the subject. If you cannot write it, that event is not yet a filter: try another one, or wait until you have delivered more work. Expanding the list can wait until that sentence exists.
A&A perspective
The overall arrangement, from acquisition through retention, is set out in "AI-native GTM: a practical guide for solo founders and small teams". If you are stuck because no filter emerges, or because the filter you chose does not match your past delivery, our free initial consultation can start by sorting out where you are. We would not decide the filter for you; we would start by working out together where in your delivery record the candidates are likely to be.
What comes first is the filter, not the volume. A filter earns its place by reaching down to why spending arises on the customer's side, and its source is your own delivery record rather than a company database.
Sources & editorial note
Primary pages read for this article. Publication dates below belong to the sources; access dates record our research.
- GTM Alpha: How Winning Teams Build a Competitive Edge
Clay · 2026-04-22 (page shows "Apr 22, 2026"; authors Varun Anand and Mishti Sharma)
Accessed 2026-10-19 - Growth AMA with YC Partner Gustaf Alstromer
Y Combinator · 2018-10-29 (page shows "October 29, 2018")
Accessed 2026-10-19