DemandCompassby automate rev.ops.
From The Demand Compass

How accurate is third-party intent data?

Vendors publish match rates without methods. In our own reviews, the placements that failed most were the ones resting on one kind of signal, not the ones with a wrong company.

Nobody can tell you how accurate third-party intent data is, because no independent study with a published method exists, and the figures vendors and their rivals report for matching a visit to the right company run from near 0% to 80%. The better question is how often an account flagged as hot holds up. In our own reviews of 2,221 hot placements across 23 DemandCompass tenants, the main failure was thin evidence, not the wrong company, and placements resting on one kind of signal were ruled wrong 47% of the time, against 26% when the evidence was mixed. Accuracy comes from corroboration, not from the feed.

What does accuracy mean for intent data?

Accuracy in intent data has two parts: identity, whether the signal is pinned to the right company, and meaning, whether that company is actually moving toward a purchase in your category, and a feed can get the first right and the second wrong. Vendors talk mostly about identity. Buyers lose money on the second, because a correctly identified company researching a topic can still be a student project, a competitor, or a deal already promised to someone else.

Third-party intent is the kind bought from a network of publishers. One large provider describes its method plainly: it watches what companies read across a co-op of sites and flags a company when the last three weeks of reading on a topic rise above that company's own 12-week baseline. [source: Bombora, Company Surge product page, 2026] That is a measure of change in reading, at the company level. It does not say who read, or whether they know you. See what intent data is for the three kinds and where each one is blind.

What do the published accuracy figures say?

The published figures on third-party intent accuracy disagree with each other by a wide margin, none of them comes with a method you could repeat, and almost all of them measure identity, matching a visit or a reading session to a company, as reported by vendors about their own products or their rivals'.

Who says itWhat they claimMethod published
Vendors, as reported in 202660 to 80% company matchNo
A visitor identification vendor, 202630 to 65% of US B2B traffic, citing "independent testing"No
An intent vendor, 2020, about IP-only rivalsabout 20%, falling to 0% with remote workNo

[source: Doug Camplejohn, Coffee, Website visitor identification accuracy, 2026] (Source: Russell Martin, Demandbase, Your Target Accounts Are Working from Home, 2020)

Remote work is the reason the vendors give for the gap: a person reading from home shows up on a residential connection, not on their employer's network.

The analysts do not supply a number either. Forrester's guide to evaluating providers asks buyers to judge relative accuracy themselves, against what their own sources and sales activity already show, and to "focus on areas you know best." (Source: Brett Kahnke and Nora Conklin, Forrester, How To Evaluate Intent Data Providers, 2025)

There is no accuracy rate for third-party intent you can look up. There is only the one you measure on your own accounts.

What did our own placement reviews show?

Our own placement reviews show that most hot placements in DemandCompass failed on the evidence behind them, not on the company: across 23 tenants from October 12 to November 13, 2026, a reviewer read the evidence behind 2,221 accounts and leads placed as hot, and flagged 76% of them as thin.

The placement review is a daily check that reads every piece of evidence behind a recently hot account or lead and rules the placement defensible, weak or wrong, by asking whether it would survive a rep's objection. Its verdicts across all 23 tenants:

2,221hot placements reviewed, 23 tenants
9.9%ruled defensible
55.3%ruled weak
34.8%ruled wrong

The reasons it gave, out of the same 2,221 (a placement can carry several): thin evidence in 1,692 (76%), stale evidence in 727 (33%), evidence that did not support the quadrant in 670 (30%), and a person outside the buying role in 340 (15%).

The sharpest cut is concentration. Where one signal type made up 80% or more of the evidence, 47.4% of 893 placements were ruled wrong. Where the evidence was mixed, 26.4% of 1,328 were. Fresh evidence did not rescue a single-type placement: with the newest signal under 30 days old, single-type placements were still ruled wrong half the time.

None of our tenants ran a third-party topic feed through the system in this window, so this is not a score of any vendor. It is what happens to readiness evidence of every kind when it is read one placement at a time. These are our tenants, B2B companies selling to technical buyers, not a sample of the market.

What do our data and the research show together?

Together, our placement reviews and the published research show that the public argument about intent accuracy is aimed at the wrong half: the research argues about identity, with ranges nobody can test, while our reviews show placements failing mostly on meaning, on evidence too thin or too old, and one kind of signal failing about twice as often as a mix.

A recent Forrester piece explains why identity is not enough. Sixty-eight percent of buyers start with a preferred vendor in mind, and that vendor wins 55% of the time, so intent signals are often the effect of a preference already formed. [source: Ian Bruce and Kelvin Gee, Forrester, Preference Matters More Than In-Market Intent Alone, 2026] A topic surge at the right company can be perfectly matched and still be research toward someone else.

Our calls point the same way. In 6 of 88 external calls between August and November 2026, the other side raised whether data could be trusted. In none of the 6 did anyone cite a measured accuracy figure from any vendor. Two were about contact data, emails and phone numbers. Two were about signals that felt noisy without a person checking them.

The book's answer to "your signals are noisy" is to agree, and then never fire on one. Readiness is a composite that needs corroboration across levels, the industry, the company and the person, before a score means anything, and old evidence loses weight on a curve. The In-Market quadrant exists for exactly this case: an account showing readiness that has not yet earned a rep.

How do you measure intent accuracy yourself?

You measure intent accuracy yourself by testing a feed against accounts whose outcome you already know, measuring what it adds on top of the signals you already have, and keeping a holdout so you can see whether acting on the signal changes anything. Four steps cover it.

  1. Back-test on accounts you know. Take last quarter's closed-won and closed-lost. Ask the vendor for its signals on them over the prior months. A feed that lit up the losses as brightly as the wins is telling you about reading, not buying.
  2. Compare the feed's surging accounts with what your own site, product and events already showed. What it adds is only what you did not already have.
  3. Never route on one kind. Require a second, independent and recent kind of evidence on the same account before a rep sees it. In our reviews, that alone separated a 47% wrong rate from a 26% one.
  4. Hold out. The book's rule is to hold back 5 to 10 percent of qualifying accounts at random and not act on them. If the worked accounts convert no better than the ignored ones, the score is fooling you.

Weight every signal by its age as well. The decay rule says a signal is evidence with a half-life, so a surge from two months ago should count for little. See the decay rule and how the pieces add up in the scoring engine.

Questions readers ask next

Is first-party intent data more accurate than third-party?

It is more specific, because you know which pages and product events produced it and often who did. Third-party data reaches accounts earlier, before they visit you, but stops at the company. Neither is accurate alone; the reliable read is the two agreeing on the same account within a few weeks.

Should I ask an intent vendor for its accuracy rate?

Ask, but treat the answer as a claim. No vendor we found publishes the method behind its figure. A trial on accounts you already know, over a few weeks, will tell you more than any number on a sales slide.

Does remote work make intent data less accurate?

It makes company matching harder for any method that relies on office networks, because people reading from home appear on residential connections. Publisher co-ops and first-party signals tied to a known person are less affected.

Why were so few of your placements ruled defensible?

The reviewer asks whether a rep could defend the placement on the evidence alone, and most placements rested on too little. That is the point of running it: a placement that would not survive a rep's objection gets flagged with the reason before anyone acts on it.

Do this Monday

Pull your intent list from last month, from any source. Next to each account, write how many independent kinds of evidence it had: the feed, your own site or product, a company event, a conversation. Count the accounts with only one. That share is your exposure, and it is a better accuracy number than any vendor's.

This note draws on the chapter In-Market of The Demand Compass: The Signal-Based GTM Framework for B2B Marketing Leaders in Complex Sales, by David Moreira and Marcos Stubrin.

Get the book →