A team with budget for one of these usually buys the other, discovers in month four that it does not do what they needed, and concludes that the whole category is overhyped. The category is not overhyped. It contains two genuinely different products that share a vocabulary.
This post is the distinction, mechanically, so you can tell which one you are about to sign for.
How third-party intent is actually built
Worth understanding in detail, because the mechanism explains every one of its strengths and limits.
A provider (Bombora is the reference implementation) operates a consent-based data co-operative of publisher websites. Bombora's runs to over 5,000 B2B sites. Those publishers share anonymised records of what content their visitors consume. The provider maps that consumption back to companies, primarily through IP and identity resolution, and builds a topic profile per company.
Then the important part: it does not report raw volume. It reports surge. An account's research on a topic over roughly a three-week window is compared against that same account's own twelve-week baseline, and a spike is flagged. Comparing each company to itself corrects for the fact that a large enterprise reads a lot about everything and a ten-person startup reads almost nothing.
The output is: this account is researching this topic more than it usually does.
That is genuinely valuable. It is also, precisely, all of it. There is no person. There is no quote. There is no artefact you can open.
What accuracy means here
An independent test by Brixon Group found roughly 81% accuracy for Bombora's intent signals, which is to say about one in five surging accounts is not meaningfully in market.
How bad that is depends entirely on how you use it.
As a prior, it is fine. If you have 4,000 accounts in your ICP and you use surge to decide which 200 get worked this quarter, a 19% error rate still leaves you far better off than alphabetical order. Prioritisation is exactly the job it was built for.
As a trigger, it is poor. If surge causes a rep to call someone with "I noticed you're researching this" (which is both a common play and a slightly creepy one), you are wrong one time in five, in a sentence that claims knowledge you do not have. And because there is no evidence behind the claim, you cannot check before you send.
Nearly every disappointment with third-party intent traces to using a prior as a trigger.
How a buying signal is built
The opposite path, with the opposite trade-offs.
You start from what you sell rather than from a topic taxonomy. A system reads public sources (conversations, job postings, reviews), retrieves candidate documents, and judges each one against your business. What survives is a specific artefact with a URL, an author, a date and a body of text.
The output is: this person, in this post, on this date, said this.
You can read it. You can quote it. You can decide the model was wrong by looking at the same evidence it looked at. And when you reach out, the reason for the timing is something the other person actually wrote, which is a different conversation from one that opens by describing their browsing behaviour.
What you give up is coverage. Signals only exist where somebody wrote something down. A company can be deep in an evaluation and say nothing in public for the entire cycle, and a signal system will never see it. Third-party intent would.
Side by side
| Third-party intent | Buying signals | |
|---|---|---|
| Unit | Account plus topic score | A post, review or job ad |
| Resolution | Account | Person, and the thing they wrote |
| Evidence | None readable | The full text |
| Coverage | Broad; sees silent research | Only what is said in public |
| Freshness | Weekly-ish, baselined over 12 weeks | As fast as the source updates |
| Auditable | No, you cannot check a surge | Yes, read the post |
| Best used as | Prioritisation | Timing and opening line |
| Typical price | Quote-only, enterprise | $49–$499/mo self-serve |
| Fails when | Used as a trigger | The buyer never posts |
The honest recommendation
If you are an enterprise running ABM against a defined account list, buy third-party intent. Coverage of silent research is exactly what you need and nothing else provides it. A signal tool is a poor substitute because your accounts are large, careful, and mostly not posting on Reddit.
If you are under about 50 people selling to a market you cannot enumerate, intent data is the wrong purchase, not because it is bad, but because you have no account list to prioritise and the pricing assumes you do. What you need is to find out who exists at all, which requires evidence rather than scores.
If you are somewhere in the middle, the combination is genuinely better than either, and in a specific order: intent narrows the account list, signals supply the timing and the opening line for the accounts on it. Most teams who report that signals "work" are describing that combination without naming it.
Where Openpulse sits, and what it will not do
We are on the evidence side, entirely. Openpulse reads Reddit, Hacker News, X, LinkedIn, Facebook, the major job boards, G2, Trustpilot, YouTube and Google Reviews, pulls full post text where the platform allows it, judges each candidate against what you sell, and shows the reasoning and the rejections.
It does not do any of the intent-data job. No co-operative, no anonymised browsing, no account-level topic surges, no coverage of research that leaves no public trace. If your buyers evaluate quietly and buy through procurement, this product will show you very little and we would rather you knew that before the trial than after it.
What it does instead is described in anatomy of a buying signal, and the category-level view is on buying signals.
See it on your own market
Paste your website, review the plan it proposes, and read what comes back tomorrow morning.
Questions about anything here? Email support@openpulse.cloud.