We check our own email data at random and publish what we find.
“How accurate is your data?” is the first question we get asked, and everywhere else in this industry the answer is an adjective. So here is ours as a number instead: a random sample of our email library, drawn separately from each trust tier, put through an independent verification service, and published with the sample size, the date, the margin of error and the method that produced it. Nothing on this page is estimated, and nothing is rounded in our favour. If the answer were 91%, the page would say 91%.
No measurement has been published yet.
This page has no number on it today, and we would rather leave it empty than fill it with one nobody took. The measurement is built and scheduled to run monthly; it needs an account with an email verification service, which we are in the middle of arranging. The first run will appear here with its date, its sample size and its margin of error, whatever it says.
The method is set out below. In the meantime you can measure us yourself, on your own list, without paying anything or talking to anybody: test your list and we will tell you how much of it we hold and how full each field is.
A deliverable address is not a correct address.
The limit, in full
A deliverable address is not a correct one. This measures whether mail arrives at the mailbox, and nothing else: an address can be deliverable and still belong to the wrong person, or be a shared departmental mailbox rather than the clinician. It says nothing about whether a specialty, an employer or a licence on the record is right, and nothing about whether a phone number is answered. It is one measurement of one field.
What it is still good for
It is the number that decides whether a campaign lands or gets your sending domain in trouble, and it is the one figure in this industry that is usually asserted rather than shown. Read it as what it is: evidence about one field, taken the same way every month, that you can watch move rather than take on trust.
For the accuracy of the rest of a record - specialty, employer, address, licence - the honest answer is a different one: every field carries the date it was last confirmed, and you can check any of it against the public record yourself. That is on what is in the data, and anything you find wrong can be reported and will be fixed for everybody.
Written down, so you can argue with it.
A number without a method is a claim. This is the method, in the order it happens, and it is the same one every month - which is what makes one run comparable with the next.
The sample is random and reproducible
Each run records the seed it drew with, the size it drew, and the population it drew from. The same seed over the same population draws the same addresses - not a similar set, the same rows - so any figure here can be redrawn and recounted rather than taken on faith.
The same sample size from every tier
The same number from each tier rather than a share of each, so every published figure carries the same precision. Sampling in proportion to size would put nearly the whole sample into the largest tier and leave a smaller one with a handful of addresses and an interval too wide to mean anything - and the small tiers are exactly where somebody needs to know what they are getting.
Stated as points
Every figure carries the margin of error that goes with its sample size, computed by a method that cannot produce an impossible bound - no interval here reaches above 100%, which the textbook formula does at the rates the trusted tier measures. A measurement without its interval is a number pretending to be exact.
Suppression wins, before anything is sent anywhere
An address covered by a suppression is never drawn and never submitted to anybody, and it is checked twice - once when the sample is drawn, and again immediately before each check, so a request filed in between is still honoured. Addresses we have taken out of use are excluded too. Somebody who asked not to be contacted is not quietly included in our accuracy measurement.
The result is a count, never a list
What we keep from a run is how many addresses were checked and how many of them each answer applied to. No address appears in the record, on this page, or in any log. An exported list of “addresses that bounced” would be a contact file with a tidier name, so it does not exist.
Monthly, and dated
It runs once a month, every month, so the figure on this page has an age you can see rather than one you have to ask about. A measurement does not get quietly kept because it was flattering: the date is beside the number, and the runs before it are below.
Do not take this page's word for it.
Paste your own identifiers into test your list and we will tell you, before you pay anything or speak to anybody, how much of your list we hold and how full each field is on it. Coverage per specialty and per state is published on the list pages, and what a record actually contains is on what is in the data.