Every visitor identification vendor has a number on its homepage. 60% match rate. 70%. “Identify up to 80% of your anonymous traffic.” The number is always big, always round, and almost never comes with a methodology attached.
Over the past 18 months, a handful of studies have tried to measure what these tools actually get right. The results are not flattering for the category. This post walks through the evidence, explains why “match rate” is the wrong thing to shop on, and gives you a short audit you can run on your own traffic before you trust any of it.
The report that put “accuracy” on the table
In March 2025, Customers.ai published its “2025 State of the Website Visitor Identification Industry Report.” The headline finding: when ecommerce brands and first-party data providers graded popular visitor ID providers, two of the three tested landed at roughly 5% to 30% accuracy. Put differently, 70% to 95% of the “identified” visitors were the wrong person.
One caveat you should hold onto. Customers.ai wrote the report, and Customers.ai came out on top (it scored itself at 65% to 85%). That is a vendor benchmark, not an independent audit, and it should be read that way. But it did something the category had avoided for years: it separated “we returned a name” from “we returned the right name.” That distinction turned out to matter more than anyone was admitting.
What 1.2 million B2B sessions showed in 2026
A more useful data point arrived this summer. In July 2026, Abmatic AI published a match rate study built on 1,204,258 real B2B website sessions over 90 days, with bot and internal traffic stripped out. Also a vendor, but this one published its methodology and its unflattering numbers alongside the good ones.
Here is what the data showed:
- Company-level match: about 47% of sessions resolved to a named company (51% if a resolvable domain counts). More than half of B2B traffic produced no company match at all.
- Person-level match: 7%. Only 7% of sessions resolved to an actual named individual. Another 15% carried an inferred job title, which is a persona guess, not a person.
- Only 1 in 5 matches was high confidence. When company matches were graded by confidence score, 20.9% landed in the high band. Nearly 60% sat in the lowest bands.
- Same vendor, 5x spread. Across sites with 1,000+ sessions, company match rate ranged from 13.3% to 64.5%. Person-level ranged from 2.2% to 9.2%.
- Geography changes everything. U.S. traffic matched at 58.8%. Germany matched at 13.9%. China and Hong Kong matched at 4% to 5%, mostly because that traffic is datacenter and proxy noise, not buyers.
Independent write-ups have started converging on similar ranges. Common Room’s September 2026 guide puts realistic company-level identification at roughly 30% to 65% and person-level at 5% to 20% for U.S. B2B traffic, “well below the 60–80% often advertised.” That is now the honest ballpark, and any pitch that ignores it is quoting a best case.
Why “match rate” is the wrong number to shop on
Match rate answers one question: how often did the tool return something? It says nothing about whether the something was correct. A tool that guesses on every session will post a spectacular match rate and a terrible accuracy rate, and the dashboard will look identical either way.
That is the part that makes false positives so expensive. They do not show up as blank records. They show up as confident, fully populated ones: a name, a title, an email, a company. Knock2’s false-positive audit guide breaks the failure modes into three patterns worth memorizing:
- Wrong company, right traffic. The visit was real, but the resolved company has no plausible connection to it.
- Wrong person, right company. The account is defensible, but the named contact left a year ago or sits nowhere near the buying decision.
- Stale identity, right graph. The match was correct when the identity graph last refreshed. The person has since changed jobs and the record never caught up.
Pattern three is a data-decay problem, and it is structural. ZoomInfo’s own published benchmark puts B2B contact decay at roughly 25% to 30% a year, and most other estimates land in the same neighborhood or higher. An identity graph that refreshes quarterly is, by definition, working from partially expired information. Multiply that by the 60% of matches already sitting in low-confidence bands, and the “identified visitor” list your sales team gets on Monday morning is a lot thinner than the row count suggests.
What bad matches actually cost you
The cost of a wrong identification is not the record. It is everything downstream of the record.
Email a stranger about pages they never viewed and you burn deliverability on a domain you need. Push the record into your CRM and you pollute the source of truth your whole team reports from. Retarget the wrong household and you pay for impressions that had zero chance of converting.
Then there is the quiet cost: after one bad week, your reps stop trusting the feed entirely, and a tool you are still paying for becomes a spreadsheet nobody opens.
This is why quality of match beats quantity of matches every time. Thirty correct identifications a week that your team acts on are worth more than three hundred guesses they have learned to ignore.
How to audit any visitor ID tool on your own traffic
You do not need a lab for this. You need a sample, a spreadsheet, and about two hours a quarter.
- Ask “company or person?” first. Get the vendor to state which level their headline number refers to. Company-level and person-level are different products with different price tags, and a 60% figure is almost always the former.
- Demand confidence scores. If the tool cannot tell you how sure it is about each match, it cannot be filtered, and you will be routing coin-flips to sales.
- Verify against known truth. Take 50 identified visitors who also filled out a form, booked a call, or exist in your CRM. Compare the tool’s identity to the real one. That percentage is your actual accuracy on your actual traffic.
- Check the geography of your traffic. If a meaningful share of your visitors sit in Germany, Australia, or behind corporate VPNs, expect the U.S.-centric benchmark to be badly optimistic.
- Test for freshness. Spot-check 20 matched contacts on LinkedIn. Count how many still work where the tool says they do.
- Ask what happens after the match. A correct identification that sits in a CSV is worth nothing. Find out what the system does in the minutes after a qualified visitor is recognized.
Run this before you sign, and again every quarter after.
Vendors change data sources. Identity graphs drift. Your traffic mix shifts when a new campaign launches. The audit that passed in January can fail in June.
Where we stand on this
We run visitor identification for our clients, so it would be easy to end this post with a number of our own. We do not publish one, and that is deliberate. As the studies above show, accuracy is a property of your traffic as much as the vendor, and any single percentage quoted before seeing your site is a guess dressed up as a fact.
What we will say: match accuracy depends on the underlying data, which is why we are open about where our data comes from. And instead of a benchmark slide, we offer a Traffic Intelligence Evaluation.
14 days. Your actual visitors. No obligation. You judge the quality on the traffic you already paid for.
If you want the background before that, our explainer on person-level vs. company-level visitor ID covers why the 47%-versus-7% gap exists in the first place, and Identity Resolution Explained covers the technology that determines which side of it a tool lands on.
The category has a real accuracy problem. The fix is not a better sales deck. It is a buyer who asks for the methodology.
