I didn't find out my CRM data was unreliable from a dashboard. I found it out on a board call, when the pipeline number I'd rehearsed didn't match the number my co-founder had in his head, and neither of us could say which one was right.
That's the trap with CRM data: it looks authoritative. Rows, fields, stages, timestamps. It has the shape of truth even when the content is wrong, and by the time you notice, you've usually already built a forecast, a hiring plan, or a board deck on top of it.
Why the number you care about is the wrong place to look
The instinct is to audit the metric you actually care about, pipeline value, win rate, average deal size, and see if it feels right. Don't start there. Those totals are downstream of a dozen smaller inputs, and a bad input can move them in either direction: it can overstate a shrinking pipeline just as easily as it understates a healthy one. Staring at the total won't tell you which.
What you can check in about twenty minutes are the structural signals that predict whether the totals are trustworthy at all, before you build a forecast, a hiring plan, or a board deck on top of them.
Five signals that predict trustworthy data
Duplicate rate. Pull a count of contacts or companies sharing an email domain and a near-identical name. Above roughly 5 to 8% of total records, reps are creating new entries instead of finding existing ones, and that means real activity is getting split across two records instead of rolled into one. Split activity looks like less activity than you actually have.
Field-completion rate on the fields you'd actually use. Not every field, just the two or three you'd pull into a forecast: stage, close date, deal amount. Query what share of open deals have all three populated. Below 70% and your forecast total is built on guesses for a third of the pipeline, even though the CRM displays one clean-looking number.
Stale record age. Check how many "open" deals haven't had a logged call, email, or stage change in 30-plus days. Past roughly 15% of open pipeline, your total includes deals that are functionally dead but still counted as live, usually the single biggest reason a pipeline looks healthier than the business actually is.
Unassigned or wrong-owner records. Count active deals with no owner, or an owner who no longer works there. Every one of those is a deal nobody is accountable for closing, sitting in your total as though someone's working it.
Bounce and invalid-contact rate. Pull a list for outbound or a renewal push and check what share of emails bounce. A rate creeping above 10 to 15% on active accounts usually means contact records are aging faster than anyone is updating them, which quietly inflates what you think is addressable.
What good looks like
None of these thresholds are precise science, they're proxies, not proof. But in practice, a CRM under roughly 5% duplicates, above 70% field completion on the fields that matter, under 15% stale open deals, near-zero unassigned active deals, and under 10% bounce rate is one I'd trust for a forecast or a board number. Miss two or three of those and I'd treat anything downstream, pipeline value, projected close rate, even headcount plans tied to a sales number, as directionally interesting, not something to bet a decision on.
The fix is usually process, not a new tool
When I ran this check the first time, the problem wasn't that we had the wrong CRM. It was that nobody owned data hygiene, so every rep's shortcuts compounded quietly for months. The fix was boring: whoever runs the weekly pipeline review got a standing fifteen-minute slot to run these five checks and clean up what they found, before any number left the room. Fifteen minutes a week is cheap. A board update built on a phantom pipeline is not.
The 15-minute weekly check
- Run a duplicate-contact query filtered by matching email domain and near-identical name; merge anything over your threshold.
- Pull open deals missing stage, close date, or amount; backfill or assign an owner to fix each one within the week.
- Sort open deals by last-activity date; flag anything past 30 days for a manual check-in, not another automated follow-up.
- Reassign any deal with no owner or a departed owner's name still attached.
- Spot-check bounce rate on your most recent outbound or renewal list; scrub anything that bounced twice.
If you're still running deals out of a spreadsheet, this check matters less; spreadsheets are usually small enough that bad data is visible on sight. But the moment a CRM total drives a real decision, how many reps to hire, what to tell investors, whether a forecast is on track, run these five checks first. It takes less time than building the forecast, and it's the difference between trusting a number and just hoping it's right.
Frequently asked questions
How often should I check my CRM data quality?
Weekly, for the five signals above, takes about fifteen minutes and catches problems before they reach a forecast or board deck. A deeper audit, checking every field rather than just the ones you use for reporting, is worth doing quarterly.
What duplicate rate is normal for a CRM?
Under roughly 5% of total records is healthy for most early-stage teams. Above 8%, reps are likely creating new records instead of searching for existing ones, and your activity counts are probably understated.
Does a low bounce rate mean my contact data is accurate?
It means the emails are deliverable, not that the job titles, ownership, or deal status attached to them are current. Pair bounce rate with field-completion and stale-record checks rather than relying on it alone.
Should I buy a data-quality tool instead of checking manually?
Not at seed stage. A weekly fifteen-minute manual check on these five signals catches most of what a paid tool would catch, without adding another subscription or onboarding curve. Revisit that decision once you have enough reps that a manual check no longer fits in fifteen minutes.