Prospecting
How do you know a contact detail is still true?
Bought data is a photograph of a moment that has passed. What a record has to carry to stay usable, and the three provenance levels that never convert.
How MarketWar handles it
Confirmed, inferred and provider NEVER convert into one another: an address we read on a page is never relabelled as one a supplier sold us, and a pattern guess is never promoted to confirmed. Records age by rule rather than by feeling — employment evidence goes stale at 365 days and freshness at 90 — a pattern needs at least 3 samples before it is used at all, and the quality gate refuses a source whose bounce rate passes 5% or whose complaint rate passes 0.1%.
Every contact database is a photograph of a moment that has already passed. People change jobs, companies are dissolved, mailboxes are retired, and nobody tells the data vendor.
So the interesting question is not how accurate a list was when it was sold. It is what each record tells you about when it was last true.
The three provenance levels, and the rule about them
Confirmed — we read it on a page. The strongest claim available.
Inferred — worked out from a pattern or a structure. Useful, weaker.
Provider — a supplier sold it to us, citing their own source.
The rule is that these never convert into one another. A bought address is never relabelled as confirmed, and a pattern guess is never promoted because it happened to deliver.
That sounds like bookkeeping until you consider how it fails. A system that collapses them shows you one "verified" column, and you then send your most careful message to the row you believed we had read on a page when actually a broker sold it. The label decides how much you should trust it, and conflating the labels destroys exactly that information.
Records age by rule, not by feeling
Two separate clocks, deliberately:
Employment evidence goes stale at 365 days. After a year, "this person works here" is a claim about last year.
Freshness goes stale at 90 days. The record as a whole needs revisiting each quarter.
They are separate because they fail differently, which is the honest limit of all verification: no check tells you whether somebody still works there. A mailbox at a company the person left is perfectly deliverable and completely useless, and address verification cannot see the difference. Only a staleness clock can flag it.
The pattern rule
A naming pattern needs at least three samples before it is used.
With one or two examples you have a coincidence. Businesses mix conventions — the founders on one, everybody hired after 2019 on another — and a pattern derived from two addresses will confidently generate a plausible address for a real person who does not exist.
The quality gate, which is about your reputation
A source is refused when its bounce rate passes 5% or its complaint rate passes 0.1%, measured on a minimum sample.
Those are not arbitrary. They are the kind of thresholds the mailbox providers themselves watch, so a source failing them is not merely low-quality — it is actively damaging your ability to reach anybody, including the people who want to hear from you.
And because sending goes out on a shared pool, that damage is not only yours, which is why the gate refuses rather than warns.
The practical habit
Re-verify before a campaign, not after a bounce. Record where every address came from, so a complaint can be traced to a source rather than to a guess. Delete the rows nobody has engaged with in a year — they cost you delivery rather than nothing.
And prefer an address you can read on a page: it is free, it is current, and the crawl will find it.
The question to ask any data vendor
For this row, where did it come from and when was it last checked?
A vendor who answers per row is selling data. A vendor who answers with an overall accuracy percentage is selling an average.
What this does not do
No verification tells you whether somebody still works there. A deliverable mailbox at a company that person left is deliverable and useless, which is why employment evidence has its own staleness clock rather than being folded into the address check.
Common questions
How fast does B2B data go out of date?
Fast enough that a list bought last year is a different list now — people move, companies close, mailboxes are retired. The useful response is not a fresher purchase but a record that says when it was last true.
What do confirmed, inferred and provider mean?
Confirmed means we read it on a page. Inferred means it was worked out. Provider means a supplier sold it. They never convert into one another, so a bought address is never displayed as something we verified.
Why does employment have a separate clock?
Because an address can be perfectly deliverable while the person has left. Verification answers whether the mailbox exists; staleness answers whether the human is still there.
What gets a source dropped?
A bounce rate over 5% or a complaint rate over 0.1%, measured on a minimum sample. Those thresholds are what the mailbox providers themselves watch, so a source failing them is damaging your sending reputation.
Related
Can you email a business you have never spoken to?
In the UK a company’s info@ and a named person at that company are different legal questions. Which you may email, on what basis, and what a sole trader changes.
How do you find business emails without buying a list?
A bought list arrives as social URLs and a name. What a free crawl can get from that, what a paid lookup is actually for, and the column people throw away.
When should you stop emailing a customer?
Sending more is the easiest growth lever and the one that destroys a list. The frequency cap, and what an unengaged contact is costing you.