Walk through the pricing pages in this category and you will find per-address prices, credit terms, integration counts and one accuracy percentage. You will not find the unresolved rate, and you will not find the false positive rate.
The unresolved rate is what share of your list the verifier declined to judge. It is the denominator of the accuracy claim, and publishing it would show immediately that a 99.6% figure describes two thirds of a business list rather than all of it.
The false positive rate is the share of addresses called valid that then bounced. This is the error that costs you something: a false negative loses you a contact, a false positive damages your sending reputation. They are not equivalent and they are always averaged together.
Neither number is hard to produce. A vendor running any internal quality process already has both. They are absent from pricing pages because they are the two figures that would make the accuracy claim readable.
How to get them anyway
Take a sample of a few thousand addresses from your real list, including the business domains. Run it through two or three vendors.
Count the unresolved results from each. That is the coverage comparison, and the spread between vendors is usually much wider than their accuracy claims suggest.
Then send to the addresses each one called valid and count the bounces. That is the false positive rate, measured on your data rather than on a benchmark somebody else designed.
Do it on your own list rather than a test file. A consumer list of Gmail addresses and a B2B list behind security gateways produce completely different numbers from the same vendor.
Where this argument costs us something
The short version
- Compare vendors on unresolved rate rather than on accuracy claims.
- Measure false positives by sending, because that is the only way anyone can.
- Use your own list for the test, since vendor performance varies enormously by list composition.
Questions people ask
Why is a false positive worse than a false negative?
A false negative removes a contact you could have reached. A false positive puts a bounce on your sending record, which affects placement for every other address on the list.
How large should a test sample be?
A few thousand addresses drawn from your real list, weighted the way your list actually is. A sample that leaves out the hard domains measures the easy part of the problem.