Here's the trap. Three vendors tell you they are 95, 95 and 99 percent accurate. You buy from one of them, send, and bounce at 12 percent. Somebody is lying, obviously, and you have no way to work out who.
Probably nobody is lying. Two different quantities are being reported under one word, the most-cited independent test was run by a company that sells data, and every claim is measured on the day the record was created rather than the day you send to it.
It is a bit like a fruit stall advertising that its produce is 95 percent fresh. True at the market, and not a claim about the state of your fridge in three weeks.
So this page sets out what the published figures actually measure, who ran the one benchmark everyone cites, what decay does to an accuracy claim over a year, and how to get a number for your own list instead.
Who has actually tested these databases?
Nobody without a stake in the answer, as far as we can find. The test cited most often put nine databases against the same 500-record sample and produced a range of 65 to 98 percent. It was run by Cleanlist, which sells data.
That does not make the result wrong. Vendor-run benchmarks are sometimes the only thing anyone bothers to build, and a 65 to 98 percent spread is a more useful finding than any single vendor's number. It does mean the one apparently neutral data point in this category is not neutral, and there is nothing behind it.
The same applies further down. Dropcontact publishes a 99 percent validity claim supported by its own benchmark across 20,000 contacts, in which it declares itself the most effective solution on the market. That is a vendor testing itself and reporting that it won.
What do the published numbers actually measure?
Two different things, and vendors quote whichever is higher without always saying which.
Coverage, or match rate, is how much of your list comes back with a result. Accuracy is how often that result is correct. A database can have excellent coverage and poor accuracy, or the reverse, and the two numbers move independently.
| Vendor | Published figure | What it is |
|---|---|---|
| Dropcontact | 99% validity | Accuracy, on its own benchmark of itself |
| UpLead | 95% data accuracy rate | Accuracy, method not published |
| FullEnrich | ~80% email coverage | Coverage, plus ~30% of raw data removed by verification |
| RB2B | 15 to 20% basic, 35 to 45% premium | Match rate, and unflattering |
| Vector | ~10% help centre, 25 to 35% CEO blog | Match rate, disagreeing with itself |
| Common Room | 60 to 75% | The partner's number for the partner's engine |
RB2B is the interesting one, because its number is low and it publishes it anyway. Fifteen to twenty percent on basic resolution is not a marketing figure, and it is the only one on this page that cost the vendor something to disclose. We looked at that whole distinction separately in what visitor identification actually identifies.
Why does a 95% list bounce at 12%?
Because the claim was measured when the record was verified, and you sent to it later. B2B contact data decays as people change jobs, at a rate usually quoted between 2.1 and 3 percent a month. That range circulates widely across vendor blogs, the figures conflict, and we have not found a traceable primary study behind any version of it, so treat it as a working assumption rather than a measurement.
Run it forward on a list sold as 95 percent accurate, at 2.5 percent a month:
| Months after purchase | Still accurate |
|---|---|
| 0 | 95.0% |
| 3 | 88.1% |
| 6 | 81.6% |
| 12 | 70.1% |
| 24 | 51.7% |
At the low end of the decay range a year takes you to 73.6 percent, and at the high end to 65.9 percent. So a list sold as 95 percent accurate is roughly a 70 percent list twelve months later, and the vendor's claim was true when they made it.
That single piece of arithmetic reconciles most of the gap between what these companies advertise and what people experience, without requiring anyone to be dishonest. It also explains why re-verifying before a send is worth more than choosing between two vendors' accuracy claims.
Can you compare accuracy claims at all?
Not usefully, for three reasons that have nothing to do with honesty.
- Different samples. A US mid-market list and a European SMB list produce very different results from the same database. Every published figure was measured on somebody else's list.
- Different definitions of a hit. FullEnrich states that a landline returned instead of a mobile counts as no result and costs nothing. Other vendors do not say how they count that, and it moves the number.
- Catch-alls. This is the big one. A vendor that treats an undecidable accept-all address as valid will report higher accuracy than one that does not, on identical data. Clearout publishes its rule; NeverBounce publishes none and separately excludes accept-all and unknown results from its bounce guarantee. We covered that split in the verifier comparison.
So how do you get a real number?
Test on your own list, because that is the only sample whose results apply to you.
- Take 200 records from your actual target segment, including the awkward parts of it. Not the vendor's sample, and not your best accounts.
- Run them through two or three tools and count three things separately: how many returned anything, how many returned a mobile, and how many bounced when you sent. Those are three different numbers and only the third one is accuracy.
- Mind the free-tier limits while you do it. Only three of fifteen free tiers permit export, and fifty free credits buys five mobile numbers at a ten-times multiplier, which is not a sample. We set out which free tiers do real work separately.
- Re-verify before every send on anything older than three months. On the decay arithmetic above, that is where the list has already lost seven points, and verification costs less than a cent an address.
The answer to the question in the title is that no independent party has measured it, the closest thing to a neutral test was run by a company selling data, and the honest range from that test was 65 to 98 percent across nine databases. Your own number is the only one that will predict your bounce rate. We set out how to get it for about $52, and it costs about a morning to get.