Intent data is any observable behaviour suggesting a person or company may be in the market for something.
Simple enough as a definition. The reason the category is confusing is that it covers things with wildly different reliability, from “this person just changed jobs,” which is a fact, to “someone at this company read an article about your category,” which is an inference about an inference.
Knowing which kind you are buying matters more than knowing the definition.
The three types
First-party
Behaviour you observe directly on your own properties and in your own data.
Someone visits your pricing page. A free user hits a plan limit. A trial expires without converting. A contact in your CRM changes jobs. A past customer’s company starts hiring.
Reliability: highest. You observed it. There is no inference about whose behaviour it was.
Volume: lowest. Limited to people who have already touched you or whom you already hold.
This is the most under-used category by a wide margin. Most teams buy third-party intent while never watching their own closed-lost list for job changes.
Second-party
Behaviour observed on someone else’s property, shared with you directly.
Review site activity, a buyer comparing you against a competitor on G2 or similar. Publisher data. Partner-shared signals.
Reliability: good. The behaviour is specific and the source is identifiable.
Volume: moderate, and usually paid.
Third-party
Aggregated behaviour across a network of sites, attributed to companies by IP address or other identifiers.
“This company showed a surge in research on topic X.” Content consumption patterns, keyword surges, topic clusters.
Reliability: the question. Two inferences stack here. First, that the activity was correctly attributed to that company, IP-based attribution is imperfect and gets worse with remote work and VPNs. Second, that the activity reflects buying intent rather than someone doing research, writing a blog post, or being a competitor.
Volume: highest, which is why it sells.
Why third-party intent disappoints
Worth being direct, because it is the category most people mean when they say “intent data.”
Attribution is company-level, not person-level. You learn that someone at a 4,000-person company read about your category. You do not learn who, and the person who read it may have no relationship to the buying decision.
Remote work broke IP attribution. A meaningful share of B2B professionals work from residential connections that do not map to a corporate IP.
Topic matching is coarse. “Researching email deliverability” could be a buyer, a consultant, a competitor, a student or a journalist.
It is sold to everyone. The same surge is visible to every vendor in your category who buys the same feed, which means the prospect gets contacted by all of them in the same week.
That last point is structural and rarely mentioned. A signal available to everyone is worth proportionally less to each of them.
None of this makes it useless. It makes it a prioritisation input, a reason to work one account before another, rather than a reason to contact someone today.
What actually works
The signals with the best ratio of reliability to effort tend to be observable events rather than inferred interest.
| Signal | Type | Reliability |
|---|---|---|
| Contact changed jobs | First-party, if yours | Very high |
| Promotion or title change | First-party | Very high |
| Company hiring for a relevant role | Observable | High |
| Someone posting about the problem | Observable | High |
| Engagement with competitor content | Observable | Moderate |
| Review site comparison activity | Second-party | Moderate–high |
| Topic surge at a company | Third-party | Low–moderate |
The pattern: events beat interest. A job change happened. A job posting exists. A post was written. These are facts with dates attached. “Showing intent on topic X” is a model’s opinion.
Using it properly
Filter against your ICP first. An intent signal at a company you cannot sell to is noise, and it is the most common way intent spend gets wasted.
Respect decay. Intent signals have half-lives ranging from 48 hours to three months depending on type, and treating them uniformly means being late for some and premature for others. Signal decay windows has the table.
Stack rather than act on singles. Two independent signals pointing the same way are far less likely to be coincidence than one. Signal stacking covers which combinations work.
Do not reference it directly. “I saw your company has been researching deliverability” is both creepy and frequently wrong. Use the signal to decide who and when; write about the problem.
That last rule matters most with third-party data, because you may be wrong about the person entirely.
What to ask an intent vendor
- Is this person-level or company-level? Most third-party data is company-level, which changes what you can do with it.
- How is attribution done? IP-based attribution has known limits worth understanding.
- How fresh is it? A weekly feed on a signal with a 48-hour window is not useful.
- How many other customers see the same signal? Exclusivity, or the lack of it, affects the value directly.
- What is the false positive rate in my segment? Few vendors will answer. The question is still worth asking.
The cheapest intent data you already own
Before buying anything, three lists worth monitoring:
Closed-lost deals. People who evaluated you and did not buy. When they change jobs or get promoted, the objection that stopped them may no longer apply.
Past positive replies. “Interesting, not now” from two years ago, now with new budget or a new employer.
Churned customers. They used the product and left. If the reason was circumstantial rather than product, a move makes them reachable again.
These cost nothing, they are person-level, they are perfectly attributed, and almost nobody watches them. Compared with third-party topic surges, it is not a close contest.
The short version
Intent data spans first-party behaviour you observed directly, second-party data shared with you, and third-party aggregated inference. Reliability runs in that order and volume runs in reverse.
Third-party topic data is company-level, weakened by remote work breaking IP attribution, coarse on topic matching, and sold to every competitor simultaneously. Treat it as prioritisation, not as a reason to contact someone today.
Events beat inferred interest. A job change, a job posting, a post written this week, these are facts with dates.
And before buying anything, watch your own closed-lost list, old positive replies and churned customers. Perfectly attributed, person-level, free, and almost universally ignored.
Next: signal-based prospecting end to end, or how long each signal stays actionable.
Frequently asked questions
What is intent data in B2B sales?
Observable behaviour suggesting a person or company may be in-market. It spans first-party data you observe directly, second-party data shared by another party, and third-party aggregated inference across a network of sites.
Is third-party intent data accurate?
Less than vendors imply. Attribution is usually company-level and relies on IP matching that remote work has weakened, topic matching is coarse, and the same signals are sold to every competitor in your category.
What is the difference between first-party and third-party intent?
First-party is behaviour you observed directly, a pricing page visit, a trial expiring, a contact changing jobs. Third-party is aggregated activity across other sites, attributed to a company by inference. First-party is far more reliable and far lower volume.
What is the best intent data to start with?
Your own records: closed-lost deals, past positive replies and churned customers, monitored for job changes and promotions. It is person-level, perfectly attributed, costs nothing, and almost nobody watches it.