We have studied how well 17 email verifiers perform. Among them, world-famous solutions such as NeverBounce, ZeroBounce, BriteVerify, etc.
To do so, we have created a test file of 100 email addresses.
The file has been processed by all the email verifiers. Finally, we measured the accuracy rate of each competitor.
Disclaimer: this study has been conducted by Icypeas, a new Email verifier. To establish our good faith and prove the reliability of this study, we have fully documented our work. Everything was recorded on video. To know more about this, see the "Open Data" section.
17
100
18
0
We created a sample of 100 email addresses. 50 were valid email addresses, 50 were invalid email addresses. To find valid email addresses, we extracted email addresses from our corporate mailboxes. We picked emails of people we discussed with recently (<5 days), and we verified on Linkedin that these people did not move to a new company. This gave us a list of ~120 emails.
We selected 50 emails out of those 120 randomly. To do so, we used the RAND() function on Excel to give each email address a random numerical ID. Then, we sorted these IDs and we kept the 50 smallest. Also,we added 50 invalid email addresses. We created various kinds of invalid email addresses: some of them are built with an invalid domain name, some others have a valid domain name and a non-existing username. We invented plausible usernames, with realistic first and last names, avoiding gibberish usernames.To make sure that those email addresses were truly invalid, we sent actual messages to them and received hard bounce notifications.
We have uploaded this 100-email list on the platforms we wanted to assess. Once the file was processed, we downloaded the verified list. We spotted the errors,i.e. the false positives (when an Email verifier says an email is deliverable while it is not) and the false negatives (when an Email verifier says an email is non-deliverable while it is). This gave us the accuracy rate, computed as follows: Accuracy rate = (number of emails – number of errors)/100.
Here is how the 50 real persons in our test file are distributed along several criteria: geography, gender, seniority, company size,etc. We had no constraint on those criteria. That’s why the sample is not evenly distributed.
