A customer database is not a static asset. Roughly a fifth of it is wrong within a year through nobody's fault. Here is what to automate, and the one operation to keep firmly in human hands.
A customer database is not a static asset. It degrades continuously, and it does so whether or not anybody touches it.
The number, and how much weight to put on it
The figure quoted everywhere is that business-to-business contact data decays at roughly 2.1 percent a month, which compounds to somewhere around 22 percent a year. It traces to MarketingSherpa research, and it is operationalised in HubSpot's widely used database decay model.
It is one dataset, repeated across hundreds of pages until it looks like consensus, which I mention because that is exactly how a single number becomes an apparent fact. Treat it as directional. The mechanism is not in doubt: people change jobs, companies rebrand, domains lapse.
You do not need the precise rate. You need to accept that doing nothing means roughly a fifth of your list is wrong within a year.
How to build it
1. Decide what a good record looks like
Write down the required fields before automating anything. Usually: name, verified email, company, role, source, owner, and date of last meaningful contact. You cannot flag incompleteness without a definition of complete.
2. Flag duplicates rather than merging them
Detect on email, then on name plus company, then on domain plus similar name. Present candidates for review. Automated merging is the single most destructive thing you can do to a CRM, because the wrong record wins and history disappears.
3. Process bounces automatically
A hard bounce should mark the record immediately. Continuing to send to a bounced address damages your sending reputation, which is a slow problem that takes months to undo.
4. Flag decay by silence, not by age
A record with no meaningful interaction in twelve months is a decay candidate. A record created two years ago that you spoke to last week is fine. Age alone is the wrong signal and it is the one most tools use.
5. Enrich at the point of entry
Fill company, size and industry when a record is created rather than in a periodic batch. Enriching at entry keeps the data fresh by definition and avoids a rolling clean-up project.
6. Send a quarterly ownership review
A scheduled list to each record owner: your records flagged as stale, duplicated or incomplete. Fifteen minutes each, four times a year, beats an annual clean-up nobody finishes.
Tools and what they cost
| Option | What it costs | Honest trade-off |
|---|---|---|
| Native CRM duplicate management | Included in most paid tiers. | Adequate duplicate detection. Decay flagging by last meaningful contact usually needs building. |
| Enrichment services (Clearbit, Apollo, Cognism) | Per contact or per seat, varying widely. | Removes manual research at entry. Coverage outside large markets is patchy and worth testing before committing. |
| Email verification (NeverBounce, ZeroBounce) | Per verification, typically fractions of a cent. | Cheap and effective before any campaign. Verification is a point-in-time check, not a subscription to accuracy. |
| Apps Script over a CRM export | Free with Google Workspace. | Full control over what counts as stale and how flags are presented. You build the detection logic. |
What it is actually worth
The value here is defensive and it shows up in three places.
Deliverability. Sending to dead addresses damages sender reputation, which quietly reduces delivery to everyone else on your list, including your real customers. This is the cost people notice last and it is the most expensive.
Wasted effort. Time spent contacting people who left two years ago.
Trust in the system. Once people believe the CRM is wrong, they keep their own private lists, and at that point you have no customer database at all, only a shared illusion of one.
Your own measure: take 50 records at random and check them. The share that are wrong is your current data quality, it takes an hour, and it is worth more than any published decay rate.
How it breaks
Automated merging destroys history. The most serious failure and the reason for the flag-not-merge rule.
Aggressive archiving loses dormant relationships. A contact silent for eighteen months may still be a warm relationship. Archive, do not delete, and keep it retrievable.
The flags are ignored. A flagged record that nobody reviews is just a field. The quarterly owner review is what converts flags into action.
Enrichment introduces errors. Third-party enrichment is confident and sometimes wrong. Never let enrichment overwrite a field a human entered.
How to tell whether it worked
Bounce rate on sends, which is the clearest external signal. Share of records with a verified email and a named owner. And the number of duplicates resolved per quarter, which should fall over time as entry-point discipline improves.