# Visitor Match Accuracy on Known Accounts and How to Test It

What published studies show about visitor identification accuracy on known accounts, why match rates fall short of vendor claims, and a protocol to test matches on your own accounts.

Every visitor identification vendor quotes a match rate, and almost none says what it was measured against. Below, the published figures sit side by side with who produced them, the reasons they differ, and a method for measuring match accuracy on companies you can verify.

- No independent test of visitor match accuracy on known accounts has been published with its method and data.
- The largest disclosed dataset, Abmatic's vendor-run study, matched about half of B2B sessions to a company.
- Only about one in five sessions reached the high-confidence band in that same study.
- Remote work, private relays, bots and mobile networks all pull real match rates below vendor claims.
- Measure accuracy yourself with logged visits from known accounts, counted against a fixed denominator.

These figures apply to company-level identification from IP and network signals on B2B websites. Person-level matching is far lower in every published figure. Every published match-rate figure comes from a vendor or from a vendor's own traffic, so treat each one as that vendor's claim until a test on your own known accounts confirms it.

## What Visitor Match Accuracy On Known Accounts Means

Visitor match accuracy on known accounts is the share of visits from companies you can verify that a visitor identification tool attributes to the right company, which is a different number from the match rate vendors usually quote, and [Leadpipe warns that a plausible contact profile is not proof of a correct visitor match](https://leadpipe.com/blog/visitor-identification-accuracy-independent-test-results).

Buyers tend to fold three different measures into the one word "accuracy". Keeping them apart is the first step in reading any figure about [website visitor identification](https://growthtribune.com/website-conversion/visitor-identification/). The three measures are:

- **Match rate:** The share of all sessions a tool attributes to any company, right or wrong.
- **Precision:** The share of those matches that name the right company.
- **Known-account recall:** The share of visits from verifiable companies that the tool names correctly.

> **Known account:** A company whose visits you can verify outside the tool, such as your own company, a customer or a partner whose people agree to take part and log each visit.

A tool's dashboard cannot tell you whether a match is right, because the tool produced the match. Visits whose company you know independently, logged outside the tool, are the way to score it.

Match rate alone says nothing about correctness. A tool that assigns a company to every session reaches a high match rate whatever its precision, and a tool that withholds doubtful matches looks weaker on match rate while sending fewer wrong companies to sales.

Before running any test, it helps to know [how company matching works](https://growthtribune.com/what-is-website-visitor-identification/), from the IP lookup to the company database behind it.

## How Accurate Is Website Visitor Identification?

In the largest study that publishes its sample and dates, [Abmatic matched 46.8% of 1,204,258 B2B website sessions to a named company](https://abmatic.ai/blog/visitor-identification-match-rate-study-2026) over a 90-day window ending July 6, 2026, yet the same vendor-run study put only 20.9% of sessions in its high-confidence band.

**46.8%** (90 days ending July 6, 2026, 1,204,258 sessions on B2B websites): The share of real human sessions that Abmatic's own platform matched to a named company, with simulated and internal test traffic excluded. Source: [Visitor Identification Match Rate Study 2026](https://abmatic.ai/blog/visitor-identification-match-rate-study-2026).

The same study shows how much the headline depends on the denominator. Counted by resolvable domain instead of company name, the company match rises to 51.2%. Inferred role or title reached 15.4% of sessions, and a named individual 7.0%. Nearly 60% of matches sat in the lowest confidence bands.

The average also hides a wide spread between sites. Across the 8 sites with more than 1,000 sessions, Abmatic's company match ran from 13.3% to 64.5%, with a median of 47.3%. A site at the low end of that range matched less than a third of the headline rate.

Two limits apply to this study. Abmatic sells a visitor identification product and ran the study on its own platform's traffic. The study also does not describe how matches were checked against known identities, so it measures how often the tool named a company, not how often that company was right.

Abmatic's study put company match at 46.8% and named individuals at 7.0%, which shows the gap between [company-level versus person-level identification](https://growthtribune.com/company-level-vs-person-level-visitor-identification/). Person-level figures are lower in every source below.

The rest of the published evidence is thinner. Happierleads and Leadpipe publish benchmarks or guidance, and Warmly, Leadfeeder (formerly Dealfront) and Leadinfo publish figures for their own tools. None states a sample, dates and ground truth together.

**Published evidence on visitor match rates**

| Vendor source | Sample | Headline figure | Ground truth |
| --- | --- | --- | --- |
| Abmatic match-rate study | 1,204,258 B2B sessions, 90 days | 46.8% company by name | Not described |
| Happierleads benchmark | US B2B sessions, undisclosed | 30-65% top tier | Not disclosed |
| Leadpipe evaluation guide | No test run | Scores withdrawn | Recommended only |
| Warmly data overview | Not disclosed | About 60% company | Not disclosed |
| Leadfeeder claim | B2B visitors, undisclosed | Up to 45% company | Not disclosed |
| Leadinfo claim | European business visitors, undisclosed | 35-40% in Europe | Not disclosed |

None of these sources discloses a known-account ground truth. Warmly adds its own caveat in its blog, giving company-level match rates of 30-65% and person-level rates of 5-20% depending on traffic, and international traffic at 20-40% company-level.

## Why Match Rates Differ From Vendor Claims

IP-based visitor matching assumes people browse from an office network, but [Gallup found that 54% of remote-capable US employees work hybrid and 25% work exclusively remote](https://gallup.com/401384/indicator-hybrid-work.aspx?os=a) in August 2026, so a large share of visits from known accounts arrive from home connections that resolve to an internet provider, not an employer.

Six drivers pull production match rates away from vendor claims. They are listed in the order a team can act on them.

**What lowers match rates in production**

| Driver | Effect on the match | Evidence |
| --- | --- | --- |
| Hybrid and remote work | Home ISP, not employer | Gallup, Leadfeeder, Snitcher |
| iCloud Private Relay | Temporary relay IP shown | Apple |
| Bot traffic | Inflates the denominator | Imperva Bad Bot Report |
| Mobile carrier networks | Often unmatched | Leadinfo, Snitcher |
| Visitor geography | Rates vary by country | Abmatic study |
| Demo versus production | Demo rates run higher | Warmly |

**Remote work:** Leadfeeder says identification depends on the visitor's internet connection rather than the device, hides visits from internet service providers and finds dynamic home IPs harder to track. Snitcher describes accuracy on residential networks as limited. An employee of a known account working from home is often invisible to both.

**Private relays:** With [iCloud Private Relay, a second relay run by a third-party content provider gives the site a temporary IP address](https://support.apple.com/102602), while the user's real address is visible only to the network provider and Apple's first relay. A tool matching that temporary address sees the relay, not the visitor's company.

**Bots:** The [Imperva 2026 Bad Bot Report found that automated traffic made up over 53% of all web traffic](https://community-cybersec.thalesgroup.com/blogs/tim-chang/2026/05/05/the-2026-bad-bot-report-is-here). Unless bots are filtered first, they swell the session count that a match rate is divided by. Leadfeeder says it removes bots and other low-quality traffic automatically, and Warmly says it filters bots before matching.

**Mobile networks:** Leadinfo states that it will not recognise a visit from a mobile network such as 4G, and Snitcher says mobile carrier visits often cannot be identified. A phone on office Wi-Fi is a different case, because the connection decides the match, not the device.

**Geography:** Abmatic's company match by visitor country ran from 58.8% in the United States and 55.5% in Canada to 34.1% in the United Kingdom, 13.9% in Germany and 4.0% in China. The study offers explanations for the spread but does not test them. Warmly says non-US traffic may identify at lower rates, and Leadinfo quotes its own figure for Europe only.

**Demo versus production:** Warmly's blog says demo match rates run 3-5x higher than what a buyer sees in production. A vendor demo on favourable traffic is a ceiling, not a forecast.

Not every matched visit is a buyer either. Customers, candidates and early researchers all show up in the same reports, which is why [a website visit does not always mean buying intent](https://growthtribune.com/why-a-website-visit-does-not-always-mean-buying-intent/).

## What Each Vendor Says About Its Match Accuracy

None of the five vendors covered here publishes a known-account accuracy test, and their own statements range from Leadfeeder's view that no honest single percentage of visitors identified exists to Warmly's estimate that its IP matching is about 60% accurate before ICP filtering. Every figure below is the vendor's own statement.

**Vendor statements on match accuracy**

| Tool | Published figure | Matching basis | Scope |
| --- | --- | --- | --- |
| Leadfeeder | Up to 45% of B2B visitors | Daily-rebuilt IP database | Companies |
| Snitcher | No published figure | Business IP range database | Companies, people via email |
| Leadinfo | 35-40% in Europe | IP, network metadata, ASN | Companies only |
| Albacross | No published figure | IP, cookies, network parameters | Companies only |
| Warmly | About 60% company | Cookie, fingerprint, IP waterfall | Companies and contacts |

### Leadfeeder

Leadfeeder is the current name of the company formerly called Dealfront, and its visitor product is now called Web Visitors. Its help centre says there is no way to honestly give a percentage of visitors identified, because rates depend on traffic, audience and industry. Its marketing still says it identifies up to 45% of B2B visitors at company level.

Matching runs on an IP-to-company database rebuilt daily from multiple data sources. Visits from internet service providers are not shown or charged, and bots and other low-quality traffic are removed automatically. Leadfeeder says dynamic home IPs are harder to track, while visits through a company VPN can be identified.

### Snitcher

Snitcher publishes no identification-rate or accuracy figure. It matches each visitor's IP address against a proprietary, continuously updated database of business IP ranges and describes accuracy by network type. Office networks rate very high, corporate VPNs high when routed through company infrastructure, and residential and mobile networks limited.

Consumer ISP, VPN and mobile carrier visits often cannot be identified and appear as "Anonymous" with an approximate IP-based location. When a visitor has given an email and the site calls Snitcher's identify method, the session is linked to that person.

### Leadinfo

Leadinfo says the identification rate for quality platforms is between 35% and 40% in Europe, about one-third of business visitors, and lower outside Europe. It reads IP address and network metadata, analyses ASNs, searches 220 million European business profiles and checks each identification against multiple data sources. When in doubt, it does not show a visitor as identified.

Leadinfo identifies companies only, needs the visitor on a business IP address and does not recognise visits from mobile networks such as 4G. A lead linked to the wrong company can be corrected by sending the IP address and the Chamber of Commerce number by email or chat. Leadinfo acquired Canada-based Visitor Queue on January 21, 2026 to extend its reach into North America.

### Albacross

Albacross publishes no match-rate or accuracy figure in its help documentation or on its comparison page. It maps IP addresses, cookies and other public parameters, such as country and network, to company records in its internal database using matching algorithms. It identifies companies only, not individuals.

Albacross adds Bombora off-site research intent and scores each identified company on on-site behaviour, off-site research and firmographic fit.

### Warmly

Warmly states company identification of around 60% and contact identification of around 45%, and says its data is not 100% coverage or 100% accurate. After filtering bots, data centres and VPNs, it runs a matching waterfall. Cookies come first and give a verified match, though they are the rarest. Browser and device fingerprinting follows, which Warmly believes is 80% accurate, then IP matching at about 60%, each rising to about 95% when filtered to the customer's ICP.

Matches the customer contributes, through Warmly UTM parameters, form autocompletion or synced HubSpot cookies, are rated high accuracy. Data comes from several providers, including Demandbase, Clearbit and Bombora, and agreement between them sets the confidence score. Warmly says niche industries and non-US traffic may identify at lower rates.

No two of these figures share a denominator. Leadfeeder counts B2B visitors, Leadinfo business visitors in Europe, and Warmly gives separate company, contact and per-method figures, so they cannot be ranked against each other. Readers choosing a tool can compare [company-level visitor identification tools](https://growthtribune.com/best-company-level-visitor-identification-tools/) on matching evidence and fit, then test the shortlist with the method below.

## How To Test Visitor Match Accuracy On Known Accounts

To measure visitor match accuracy on known accounts, send logged, controlled visits from companies you can verify through each network type, then divide the visits each tool attributes to the right company by all controlled visits, the approach Leadpipe describes as testing with participating visitors whose identities and visits are independently logged.

The protocol needs a panel of participants, a shared log and access to each tool's list of matched companies. Work through it in this order:

- [ ] Build a known-account panel of companies you can verify, such as colleagues, customers or partners who agree to take part, and log each participant's company domain.
- [ ] Plan visits from every network type, including office network, corporate VPN, home broadband, mobile carrier and iCloud Private Relay, because each resolves differently.
- [ ] Give each visit a unique landing URL or UTM tag, and log its time, network and participant.
- [ ] Run every tool on the same pages over the same observation window.
- [ ] Compare each tool's matched company domain with the logged domain, and record a correct match, no match or wrong company.
- [ ] Report known-account recall and precision for each network type, with every denominator shown.
- [ ] Sample live traffic separately for 30 days to measure each tool's match rate on real visitors.

**Panel:** Spread participants across several companies rather than recruiting many people from one, because one employer's office network would dominate the result. Record the company domain each participant should match to before any visit takes place.

**Network types:** The mix of networks matters more than the number of visits. Snitcher rates office networks very high and residential and mobile networks limited, Leadinfo does not recognise mobile network visits, and Apple's relay hands the site a temporary address. A panel that only visits from the office will overstate accuracy.

**Tagging:** A unique URL or UTM parameter lets you find the same visit in every tool. Warmly counts its own UTM parameters as a high-accuracy match route, so keep test tags separate from any tag a tool reads as identity, or the test measures the tag rather than the network match.

**Window:** Leadpipe's evaluation guide recommends a common observation window. Run every tool at the same time on the same pages, so database updates and traffic shifts affect every tool equally.

**Scoring:** Count each controlled visit once per tool, then calculate the measures below.

**How to calculate each measure**

| Measure | Numerator | Denominator |
| --- | --- | --- |
| Known-account recall | Correct company matches | All controlled visits |
| Precision | Correct company matches | All matched controlled visits |
| Wrong-company rate | Wrong company matches | All matched controlled visits |
| Match rate | Sessions given any company | Live sessions after bot filtering |

Report each measure by network type as well as in total. A single blended figure hides the fact that office visits and home visits behave like two different products.

**Live sample:** Warmly advises buyers to ask vendors to prove results on their own traffic for 30 days and to validate accuracy, not just match volume. Run the live sample alongside the panel. It gives the match rate on real traffic, while the panel gives precision and recall.

> **Warning:** Every participant should agree to take part before any controlled visit, and the test should run under the site's existing consent setup, with no tool loaded outside it. This is general information, not legal advice.

**Limits of the method:** A panel is small, its mix of networks is chosen rather than random, and the result holds only for that site's traffic over that window. Treat it as a check on a vendor's claim for your own accounts, not as a market-wide figure.

## How To Classify Visitor Match Errors

One visitor match error a vendor names itself is a wrong company caused by an outdated IP registration. Leadinfo gives it as the reason a lead can be linked to the incorrect company, and fixes the link manually when sent the IP address and company registration number.

Each wrong or missing match in a test falls into one of seven types:

**Visitor match error types**

| Error type | What it looks like | Likely cause | How to spot it |
| --- | --- | --- | --- |
| Stale IP registration | Wrong company named | Outdated IP record | Logged domain differs |
| ISP or hosting network | Provider named as visitor | Home or cloud connection | Telecom or cloud firm named |
| VPN or relay exit | Wrong company or none | VPN or Private Relay | Fails on that network only |
| Shared network | Host company named | Co-working or partner office | Same company for many people |
| Parent or subsidiary | Right group, wrong entity | Group-level IP registration | Parent or sister domain |
| Bot visit | Company with no human | Automated traffic | Visit missing from log |
| Low-confidence match | Plausible but unconfirmed | Probabilistic fingerprint or IP | Low confidence score |

The vendors' own documentation explains several rows. Leadfeeder hides internet provider visits and Snitcher shows them as Anonymous, so an internet provider named as a company points to a tool without that filter. Warmly filters data centres and VPNs before matching and treats fingerprint and IP matches as less certain than cookies. Apple's relay replaces the visitor's address with a temporary one.

Shared networks and parent-subsidiary errors are judgement calls rather than published failure modes. A co-working space or agency office carries many companies' people on one connection, and a group's IP block may be registered to the parent. Count them anyway, because they decide whether a match reaches the right sales owner.

Count each error type separately for each tool. The largest category is the question to put to the vendor. A wrong company can be fixed at the source in some cases, as Leadinfo does manually for stale registrations, rather than only in your CRM.

## What A Visitor Identification Study Should Disclose

A visitor identification accuracy figure is only usable when it names who ran the test, its sample, dates, denominator and ground truth, and [Happierleads' reverse IP accuracy benchmark](https://happierleads.com/blog/reverse-ip-lookup-accuracy-benchmark-2026) gives match and precision ranges from blind tests without disclosing its sample, dates or ground truth. Check any figure against the list below before it shapes a buying decision.

- [ ] Who ran the study, and whether they sell a visitor identification tool
- [ ] Sample size and the denominator behind each figure
- [ ] Start and end dates of the data
- [ ] Geography and traffic mix, including mobile and remote visits
- [ ] How ground truth was established for each match
- [ ] The confidence threshold for counting a match
- [ ] Separate company-level and person-level figures
- [ ] The spread across sites, not only the average
- [ ] Access to the raw data or an audit report

**Abmatic:** Its study discloses a sample of 1,204,258 sessions, a 90-day window ending July 6, 2026, separate denominators, confidence bands, and per-site and per-country spreads. It does not describe how matches were verified against known identities, and it runs on the vendor's own platform.

**Happierleads:** Its benchmark, published June 23, 2026, claims that top-tier vendors identify 30-65% of US B2B sessions and bottom-tier vendors 10-25%, across blind tests on identical traffic. It names no vendor per figure and discloses no sample, dates or ground truth.

**Leadpipe:** Its page titled as independent test results says its evaluation steps are recommendations, not reported test results. Earlier vendor scores on that page were removed because no inspectable audit report or dataset existed.

Any figure credited to an unnamed independent audit fails the first two checks on the list.

## What Match Accuracy Means For Routing Known Accounts To Sales

Because most company matches carry low confidence, with only about one in five sessions reaching the high-confidence band in Abmatic's study, a known account should reach sales automatically only when the match is high confidence or confirmed by a second signal such as a form fill or tagged email click.

Four rules turn the evidence into a routing policy:

- **Route strong matches automatically:** Send a company match to sales without review when it is high confidence. Repeat visits alone do not count as confirmation.
- **Ask for a second signal on weak matches:** A form fill, an email identify call, a UTM click or a match to an account owner in the CRM should confirm a single low-confidence match.
- **Review shared networks and hosting providers first:** Hold matches from co-working spaces, agencies and cloud providers for a person to check before routing.
- **Re-run the known-account test after a change:** A new vendor, a new market or a large shift in traffic mix can move accuracy.

The vendors support the second-signal rule. Snitcher links a session to a person only after the visitor gives an email and the site calls its identify method. Warmly rates its own UTM and cookie matches above its probabilistic IP matches. Leadinfo withholds a visitor when in doubt.

A higher match rate is not always better. A tool that names more companies may name more wrong ones, and Happierleads describes the same trade-off in its own ranges, published without a disclosed method. For routing, a wrong company sent to a rep costs more than a visit left unmatched.

> **Tip:** Store the match confidence and matching method next to every routed account in the CRM. When a rep reports a wrong company, the record shows whether the error came from a low-confidence match that should have waited for a second signal.

The playbook on how to [route identified accounts to sales](https://growthtribune.com/how-to-route-identified-accounts-to-sales/) covers ownership, qualification rules and follow-up times once the match itself can be trusted.

## Frequently Asked Questions

**Is match rate the same as accuracy?**

Match rate is not accuracy, because match rate counts every session a tool attributes to any company while accuracy counts only the attributions that name the right company. Abmatic's study matched 46.8% of sessions by company name without describing how matches were checked against known identities, so its figure is a match rate.

**Does visitor identification work on mobile traffic?**

Visitor identification works poorly on mobile carrier traffic. Leadinfo states it will not recognise a visit from a mobile network such as 4G, and Snitcher says mobile carrier visits often cannot be identified. A phone on office Wi-Fi is a different case, because the connection decides the match, not the device.

**Can I correct a visit matched to the wrong company?**

Yes, with Leadinfo at least, which says a lead can be linked to the incorrect company when an IP address is still incorrectly registered. Send Leadinfo the IP address and the Chamber of Commerce number by email or chat, and it can fix the link manually. Log each case so repeated errors show up.

**Is a higher match rate always better?**

A higher match rate is not always better, because it can come with lower precision. Happierleads claims, without a disclosed sample or ground truth, that high-recall tools match 50-70% of sessions at 60-80% precision, while high-precision tools match 30-45% at over 90%. For sales routing, a wrong company costs more than a missed one.

## Sources

- [Visitor Identification Accuracy: Independent Test Results](https://leadpipe.com/blog/visitor-identification-accuracy-independent-test-results), leadpipe.com
- [Visitor Identification Match Rate Study 2026](https://abmatic.ai/blog/visitor-identification-match-rate-study-2026), abmatic.ai
- [Global Indicator: Hybrid Work - Gallup](https://gallup.com/401384/indicator-hybrid-work.aspx?os=a), gallup.com
- [About iCloud Private Relay - Apple Support](https://support.apple.com/102602), support.apple.com
- [The 2026 Bad Bot Report is Here!](https://community-cybersec.thalesgroup.com/blogs/tim-chang/2026/05/05/the-2026-bad-bot-report-is-here), community-cybersec.thalesgroup.com
- [Reverse IP Lookup Accuracy Benchmark 2026](https://happierleads.com/blog/reverse-ip-lookup-accuracy-benchmark-2026), happierleads.com

---
Source: https://growthtribune.com/original-study-visitor-match-accuracy-on-known-accounts/