Hey there,
Matteo here from Kinetyca.
A client of ours was going to a conference recently. They wanted to reach the people who would be in the room, which is a reasonable thing to want, and there was no way to do it. Nobody publishes an attendee list. You cannot buy one. Asking the organiser goes nowhere.
Two years ago I know exactly what we would have told them, because we said it all the time. Not worth it. We could build something to find those people, it would take most of a week, and it would work for that one conference and then sit there doing nothing.
So the client would not get their list, everyone would move on, and nobody would write down what had just happened.
That last part is what I want to talk about.
The cost was never the hours
When people describe what has changed in GTM data over the last two years, they describe it in hours. This used to take a day, now it takes ten minutes. True, and the least interesting version of it.
The real cost of expensive data work was the questions you stopped asking.
Every question you could not afford to answer came off the list quietly. Not with a decision, just with a shrug. Can we see which of these 400 companies runs the software we replace? Can we reach the people going to this conference? Each one had a build behind it, so each one had to justify itself, and most could not. You stopped asking, and after a while you stopped noticing that you had stopped.
That is what actually changed. Not the speed. The price of curiosity.
What we did instead
The client got their list. We searched for the people posting about the conference and took the authors of those posts. The posts were the list.
It ran on a skill we had written months earlier for something unrelated, and the instruction was one line. We ran it in July for an event in October, so the volume was thin. We will run it again in September when there is more to find.
That skill is one of about nine we keep inside Claude now. I recorded a walkthrough of the library, including one running live from a blank instruction through to a finished file.
Watch: 9 Claude Skills That Do Our ICP Research
The one thing worth stealing
If you take one thing from this, take the shape.
Every one of those nine is a defined input and a defined output. That is all.
- A lead list → the recent posts those people published
- A set of post URLs → everyone who reacted to them
- A topic → recent posts on it, and their authors
- A job title and a location → the companies hiring for it
- A list of domains → the technology those companies run
- Company names → their funding rounds and the investors behind them
- A business type and a location → the businesses that match
None of them describe a method. They describe a contract.
That is the test, and it is worth applying honestly. If you can write the input in one line and the output in one line, you have a skill, and anyone on your team can run it. If you cannot, you have a prompt, and a prompt has to be re-explained by whoever happens to be holding it that day. A lot of what gets called a skill is really a prompt with a nickname.
Two gates, before anything clever
Both of these matter more than any of the interesting parts, and both are boring.
The first is that it is allowed to come back and ask. In the recording I gave it a conference name, it found nothing useful, and instead of guessing it stopped and asked me what the conference actually was. That is the behaviour you want. A skill that guesses at a vague input does not fail loudly. It spends your money collecting the wrong data and hands you a clean-looking file, and you find out three days later when the copy makes no sense.
The second is that it prices the job before it runs. Mine came back with roughly 30 to 60 relevant posts from around 200 profiles, at about 40 cents. Then it waited. I say go, and only then does it spend anything.
Forty cents is the number that makes the rest of this letter true. When an answer costs forty cents and lands in a few minutes, you stop rationing questions.
What I am not going to pretend
You cannot open Claude and do this. Behind those skills is a system we have spent months building, and it knows our clients, our ICPs, our formats, the way we work. Without that layer, "run post search" is not an instruction, it is a wish.
The skills are the cheap part. The context underneath them is the expensive part, and that is the part that is actually ours.
On Clay, since plenty of these replaced Clay builds: it has not gone anywhere for us and still runs inside our outbound motion. What moved is the layer above it. Some of those old builds were genuinely good. They just took days to design, and only ran properly when the person who built them was the person running them.
One small thing you can do this week
Start the list of questions you are not asking.
Not a project list. A list of things you have wanted to know about your market and quietly written off, because finding out was going to cost more than the answer was worth. Including the one somebody raised in a meeting eighteen months ago that nobody picked up.
It is harder to write than you would expect, and that is the point. You are not recalling decisions, you are recalling things you trained yourself not to bother wanting. Give it twenty minutes and you will get four or five. Two of them now cost about a dollar to answer.
That list is a better roadmap than most quarterly plans, because everything on it is something you already wanted.
One of those conferences, for me, is Sculpt in San Francisco in October. I went last year and it was genuinely excellent. If you are going, reply and tell me, I would like to meet you there.
And I would like to see the list. What is a question about your market you gave up on? Hit reply, I am collecting them.
PS. If you want this built into your GTM, grab 20 minutes here.
Lawyer by training, then operations at Amazon. Runs delivery: process, documentation and the detail corporates expect.
LinkedIn ↗YouTube ↗