Case Study: The buying signals hiding inside job ads (and what a year of them revealed)

“We never have a problem closing a deal”

That was the first thing the founder said to me.

He sells software to clinical-trial companies, and his customers are the companies that drug makers hire to run their trials.

He was good once he got the meeting. Getting the meeting was the hard part.
A clear top-of-the-funnel bottleneck, and timing the buyer was part of the challenge.

Nearly all of his US leads came through a single partner, and the one outbound agency he hired ran for fifteen months and produced a single deal.

He had already tested the “spray and pray” approach.
No surprises, it didn’t work.

The question was whether we could build something smarter.

The problem with intent signals

The obvious place to start was intent.

Funding, hiring, leadership changes and other signals are useful. They tell you that something is changing inside a company.

But on their own, they’re not enough.

A company raising money doesn’t necessarily mean it’s about to buy your software. A new VP doesn’t necessarily mean they’re looking for a new vendor.

Some of those signals give names.
Some others help paint a picture, but not the full picture.

The useful question here is not just “what changed?”
It’s “what is this company actually dealing with right now, and what might that change lead to?”

That’s what I wanted to figure out.

Companies announce what they’re about to do when they hire

When a company hires, it usually tells you a lot of little things in their full job descriptions.

The job description can tell you what systems they’re using, what they’re trying to fix, what they’re building, who owns it, what deadlines they’re working towards and sometimes even what budget they have.

Of all the AI and GTM tools out there, almost no one is looking at this.

On top of that, there is a little problem: most tools give you a snapshot of what’s on the job board today. I wanted the history.

So I pulled the job ads each company had published over the previous twelve months, including the full descriptions rather than just the titles.

I'm not going to expand on how I did this.
It was hard, and used a lot of techniques and providers.

But after a few days, I had 1,087 job ads.
Enough to test whether our hypothesis would hold.

The build and using AI where it excels

AI reads every job ad. There were 1,087 of them across the test I set up.
LLMs are really good at these tasks: reading, interpreting, making summaries and answering yes/no questions.

DeepSeek does the bulk reading, ad by ad, because it is cheap. Claude checks the work independently, in a build that shares no code, so one mistake cannot slip through both. Every hit has to carry an exact quote. No quote, no signal.

Finding the right companies

For this market (US healthcare), I started with public FDA data. The FDA publishes filings that contain thousands of companies, so I used that as the starting point and had AI work through them to identify companies that were actually relevant.

Of the high-confidence candidates, only around 7% turned out to be genuine clinical research operators, in the specific segment we needed to target.

That gave me a much smaller universe to work with.

And that matters. A LOT.

If your actual market is around 500 companies, you don’t need another tool that sprays messages at thousands of people.

You can watch all 500.

The question becomes: which one should you be talking to this week?

What a job ad can actually tell you

A title like “Director of Clinical Operations” tells you almost nothing.

The description underneath it can tell you a lot.

For example, I built this composite from patterns I found across the real ads:

One ad is noise.

A year of them is a briefing.

You can see what problems the company is dealing with, which tools it already runs, who owns the decision, what it’s trying to change and what its priorities are for the coming months.

That is much more than a hiring signal.

It’s a picture of the business. And where they are heading.

The first test didn’t work

I started with the company’s own closed deals, won and lost. Job ads predicted almost nothing.

Most of those deals had almost no ads on record. One ad. Four ads. Five ads. The honest label is “not enough data”, not “failed”. I kept that result and said so.

So I set up a cleaner test: 64 companies the client had never worked with. 12 had started a new trial in the next year. 52 had not. Then I looked back 12 months at everything they had published.

What the test found

The purchase everyone expected to see in the ads never showed up. What did show up was quieter: companies getting ready for the check their own customers run before handing over a big contract.

That check happens months before the contract is awarded, and long before anything is public. So the ads did more than name a company. They put it on a clock, and showed when it was in the market while there was still time to be in the room.

The obvious purchase was not the signal. The preparation behind it was.

Companies were hiring people to deal with operational problems, replace spreadsheets, implement systems, build teams, prepare for inspections and change the way they managed trials.

They showed when a company was entering a period where a particular problem mattered, before the buying decision became obvious.

That was the useful part. This is when the software was needed. "The buying window".

The sample was small, so this points a direction rather than proving one. But it changed how I read every other signal.

Every signal gets a job

The other signals were not useless. They were being asked to do the wrong job.

• Job-ad language: Who to call, and what to say
• Leadership change: When to call
• Acquisition, with a direction: Potential threat or opening
• Funding: Whether you’re looking at the right market at all.

It’s a short briefing on each company.

— What are they dealing with?
— What do they already run?
— Who should I talk to?
— What should I say?

By the time another signal fires, the message is already written.

The competitive advantage

That’s what I mean by GTM alpha: an information advantage on a named account that a competitor using the same tools can’t simply buy.

They might be looking at today’s job board for the same company.

I’m looking at the last twelve months.

I know every project they are working on, roles in charge, the software they use and the projects they want to prioritize.

What happened next

This was a two-week sprint that I run first with my customers after one quick audit.

The two-week experiment became an ongoing engagement.


I now run the same reads monthly, feeding a database the client owns rather than another platform they rent. It's, of course, much more sophisticated, but it all started from a simple experiment.

The line I remember from the sales lead, watching the LLMs run: “frightening how powerful this stuff is already”.

What this means for your company

If you sell B2B software, your prospects’ job ads already tell you what they are about to buy, what they run today, and what they are trying to fix this year. Almost nobody reads them properly, because a snapshot of today’s openings is easier.

Two things I would do before paying for any tool. Test the signal against your own won and lost deals, even when the history is thin. And read the description, not the title.

These days I run this for other companies as a two-week sprint. If you want to know which signal actually fires in your market, book a call with me, and let's find out.