Buying signals and intent data: which ones are evidence and which are horoscopes
Most intent data is a guess about a company dressed up as a signal about a person. How to tell the two apart, and the four signals worth building a week around.
A buying signal is something that happened, with a date and a link: a role posted, a product shipped, a question asked in public. Intent data is usually an inference about a company, drawn from which articles people at that company seem to have read. The two often get sold as if they were the same thing, but only one of them holds up when you ask where it came from.
The test that separates them
Ask one question of any account in your pipeline: can I paste the sentence, and does the link still open?
That's the whole distinction, and it's the same evidence standard that separates a real lead from a plausible one, just applied a step earlier.
A job post passes. A funding announcement passes. "Intent score 78 in the Sales Automation topic" doesn't, because there's no sentence and no link, just a number from a process you're not allowed to inspect.
Where intent data comes from
It helps to understand the mechanism, because it tells you when intent data works and when it doesn't.
Most third-party intent data works roughly like this. A provider has a network of publisher sites. Someone reads a few articles on a topic, the provider records the visit, matches the IP address to an organisation, and reports that the organisation is researching that topic.
The weak step is that match. Someone on an office network maps to their employer. Someone working from home maps to their internet provider, and someone on a phone maps to a mobile carrier. So the method works best for big companies with offices, and worst for exactly the small, distributed companies that most early-stage products sell to.
That's a limitation rather than a scandal. It only becomes a problem when it's sold without the match rate. So before buying any, give the provider a hundred of your existing customers and ask what percentage they match correctly. That number is the product, and a provider that won't run the test has answered the question anyway.
Four buying signals worth building a week around
Someone asks in public. This is the strongest signal there is, and the shortest-lived: a person describing your problem in their own words, with a timestamp, somewhere you can reply. It's worth hours, not days. Finding where those questions get asked is a job of its own.
A job post describes the problem. Read the responsibilities, not the title. A post that lists "reconcile weekly reports across four systems" is a description of the problem you solve, written by the people who have it, and published on purpose.
They launched something that creates the work. A second location, a new product line, a compliance certification. Each one creates follow-on work, and some of that work might be yours.
They tried and stopped. Someone writing publicly that they evaluated your category and built it in-house instead. It's rare, but when you find one it's worth more than most other signals, because the budget conversation has already happened.
Those four are the ones we'd build a week around. The longer list of twelve sales signals covers the rest, including the first-party ones only you can see, with which kinds actually turned up in our own data.
How quickly signals go stale
The chart above is conceptual, but we've seen the real version in our own data. Of the 843 conversations Revtive's agents flagged as worth a reply between July and September 2026, only 39% came with a date at all. Among the dated ones, 23% were less than a week old when found, 42% were less than a month old, and 24% were more than eight months old. The median was 51 days.
Two lessons come out of that. First, a signal without a date is a fact of unknown age, and plenty of the "signals" in any feed will be old news. Second, age has to be part of how a signal is scored. We learned that when a five-year-old Reddit thread turned up in one workspace rated as high-impact, simply because it matched the product well. We now cap anything older than eight months at low priority.
What isn't a buying signal?
Anything you can't back with a sentence you can paste and a link that still opens.
| Looks like a signal | What it really is |
|---|---|
| Headcount grew 12 percent | An attribute, often from a data source nobody at the company would recognise |
| They're in your target industry | A filter you set, reported back to you |
| Anonymous visit from a company | An IP-based guess, often just an internet provider |
| They use a tool you integrate with | Real, but permanent, so it says nothing about timing |
| Opened your last email three times | Quite possibly not a person at all |
That last row catches people out. Apple's Mail Privacy Protection is designed to stop senders seeing whether an email was opened, and as Litmus explains, it does that by loading the email's images, tracking pixel included, whether or not anyone reads it. Some corporate security filters also open links to scan them. So a burst of opens can come from software rather than interest.
An example, worked through
Here's an illustration. Say you sell scheduling software to veterinary practices.
- A firmographic list gives you 900 practices with four to twelve vets in your region, with no reason attached to any of them.
- An intent overlay flags 140 as "in market", and you can't see why for any single one.
- A signal pass finds 31 practices that posted a receptionist or practice manager role in the last three weeks. Of those, 11 list appointment scheduling or phone handling in the responsibilities, and two more have a recent public review complaining about hold times.
That's thirteen accounts, each with a sentence and a link. It's a tenth the size of the intent list, and every one of them can be defended, which is what makes it possible to write to them at all.
How to store a signal so it's still useful next week
A lot of teams lose signals to their own CRM. The event gets recorded as a checkbox ("intent: yes") and whatever made it useful is gone.
Store three things per signal, always together: the actual sentence, copied rather than summarised, because the phrasing is what you'll quote back; the link, so the claim can be checked again by whoever picks it up in three weeks; and the date you found it. A job post found on the 4th is a different proposition on the 25th, and without the date nobody can tell which they're looking at.
A record with those three things survives being handed to someone else. A record that says "high intent, score 78" doesn't survive a single question, and that's the test any lead should pass before it gets a message.
What signals won't tell you
A signal isn't a need. A company hiring a compliance manager might be about to solve the problem without you, for good. The signal tells you the topic is live, not that a purchase is coming.
Signals aren't spread evenly, either. Some markets publish almost nothing (dentists don't post changelogs). If your buyers are quiet in public, signal-based prospecting has a low ceiling, and you want to learn that in week one, not month six. Try to find ten signals by hand before you build any process around them.
And more signals isn't better. Tracking twenty kinds produces a queue nobody works through. Two well-chosen ones, checked weekly, beat a dashboard. Revtive's agents watch a small set on purpose and have to cite a source for each one, because a signal you can't open is exactly what this article argues against.