Buy ads to learn the words, not to buy the customers
Advertising is the wrong acquisition channel for most small B2B companies and an extremely good research instrument. The deliverable is not customers. It is four words that make the right person click.
Small B2B companies mostly cannot afford paid acquisition, and mostly know it. The maths is unforgiving: you are bidding against better-funded companies for clicks from a buyer who needs several touches before they will talk to anyone, and the payback arrives quarters later, if it arrives.
So the usual advice is to skip ads. That advice throws away the most useful thing ads do, which has nothing to do with acquisition:
ROI doesn't have to be the reason you're running ads. It could be to learn how to position in ways that really get people interested. So you could run the ads only long enough to get a sense of, my God, this is how you get someone to click in four words.
Jason Cohen
It costs me a known amount of money to quickly iterate into the best positioning I can get — which you can then use in any way.
Jason Cohen
A known amount of money. That is the part that changes the decision. As an acquisition channel, ads are an open-ended commitment against an unknown payback. As a laboratory, they are a fixed invoice for a specific answer, and the answer is words you then use everywhere — in cold email, on the homepage, on a call, in the first line of a proposal.
Why nothing else answers this question
You could try to learn positioning from your other channels. Each has a reason it cannot:
SEO is not how you figure that out. It takes too long, changes take too long, and it's very hard to say exactly what did what.
Jason Cohen
| Channel | Time to a readable signal | Can you attribute it? |
|---|---|---|
| SEO / content | Months to years | Barely — rankings move for reasons you did not cause |
| Cold email | Days, but low volume | Yes, though a hundred sends is a thin sample |
| Sales calls | Immediate, richest detail | Yes — but you only reach people who already replied |
| Paid ads | Hours | Precisely, and at whatever volume you buy |
Sales calls are the deepest instrument and they have a selection problem: everyone on a call already responded to something. Ads test the sentence on people who have not agreed to anything yet, which is the population your cold outreach is actually addressing.
Running it as an experiment
The design matters, because an experiment run like a campaign produces a campaign's uninformative result.
Write hypotheses, not variations. Not five headlines — three claims that imply different buyers or different problems. The career claim against the utility claim. The expensive-problem framing against the time-saving framing. A named role against a named industry. If two headlines could be true of the same reader, they are one hypothesis with two haircuts.
Fix the budget and the end date before you start. This is a purchase, not a channel. Decide the number, decide the date, and let it end on the date whatever happens.
Measure the click, not the conversion. You are buying an answer about attention, and your landing page and follow-up are separate variables you have not controlled. Click-through rate at a fixed audience is the reading. Conversions in this volume are noise.
Keep the audience identical across arms. Same targeting, same placement, same spend. The only thing changing is the sentence.
Write down what would surprise you before the numbers come in. It is the cheapest way to stop yourself explaining any result as a confirmation.
Reading the result honestly
A clear winner is the good case: you now have a sentence with evidence behind it, and it goes into the cold email opener, the homepage headline and the first line of the deck the same week.
Three results that are not "no signal", though they get filed that way:
- All arms perform the same. Your claims were closer together than you thought. They are variations on one position, and you have not yet found a second one to test against it.
- Everything performs badly. The most useful outcome available, and the cheapest version of a lesson usually learned over two quarters of quiet cold email. Either the audience is wrong or the category is not something anyone is looking for.
- The winner is the one you least wanted. Take it seriously — that is the entire reason to run the test rather than argue in a meeting. Then check who clicked before you rebuild the company around it.
The one trap: a headline can win clicks by promising something you do not sell. Always read the winner back against what you actually do before adopting it.
What this is not
It is not a plan to become an ads-driven business. It is not a reason to keep spending after the answer arrives — the moment you have the sentence, the experiment is over and the budget stops. And it does not replace talking to buyers. A hundred clicks tell you which words earn attention; they cannot tell you why, and only a conversation does that.
Doing it
- Write three genuinely different claims about who this is for and what it changes for them.
- Set a budget you would spend on a research project — the fixed sum you would pay a consultant for an answer — and an end date.
- Run one audience, three arms, identical everything else.
- Take the winning words into cold email and the homepage the same week. The ads were the instrument; the copy is the product.
- Stop on the date. The failure mode is a laboratory quietly becoming a channel you cannot afford.
Keep reading
- How to answer a hard question on a call while you are still thinking Someone asks the question you did not prepare for. There is a four-part shape that gets you through it sounding certain — and a rule about what goes in the middle that decides whether anyone believes you.
- Personalisation that works, and personalisation that costs you trust Using someone's first name is not personalisation — and the people you most want to reach read it as proof that a machine sent the email. Here is the line between the two.
- The first salesperson you hire should be you, for longer than you want There is a revenue number below which nothing sales-related should leave the founder, and it is higher than most people expect. The gate before any hire is not a budget — it is a count.