CRM Data Quality Metrics: What to Measure and What to Ignore

Every team that gets serious about CRM data quality eventually asks the same question: what do we actually measure? The bad answer — and it’s the popular one — is a single “health score” out of 100, computed by a vendor’s opaque formula. The good answer is five specific metrics, each with a formula you can write on a whiteboard, a threshold you chose on purpose, and a list of offending records behind every number.

We run data quality audits on client CRMs for a living. These are the metrics we actually compute, why we compute them this way, and the ones we’ve stopped bothering with.

The five dimensions, translated into CRM terms

Data quality theory gives you the classic dimensions — completeness, uniqueness, freshness, validity, consistency. They’re the right frame, but only after you translate each one into a concrete metric on a concrete object. “Completeness” means nothing; “percentage of open deals with an amount, a close date, and a next step” is something you can audit, trend, and argue about productively.

The thresholds below are our starting defaults — the values we set on day one of an engagement and then tune. They are not industry benchmarks, because honest industry benchmarks for this don’t exist. Your right numbers depend on your sales cycle, your tooling, and how much of your pipeline actually feeds the forecast.

1. Completeness: are the fields that drive decisions filled in?

Not all fields matter equally, so don’t measure completeness across all of them. Measure the fields the forecast and the funnel actually run on:

• Open-deal completeness = open deals with amount AND close date AND next step ÷ all open deals. Starting default: 90%. These three fields are the minimum for a deal to participate in a forecast at all.
• Contact lifecycle completeness = contacts with a lifecycle stage set ÷ all contacts. Starting default: 95%. A contact without a lifecycle stage is invisible to funnel reporting and usually to routing.

The trap here is measuring completeness on fields nobody uses. A 60% fill rate on “industry” might be fine; a 60% fill rate on close date means nearly half your pipeline can’t be forecast. Pick five to eight fields per object that decisions actually depend on, and ignore the rest.

2. Uniqueness: new duplicates per week, by source system

Most teams measure total duplicate count, which is a backlog number — useful once, during cleanup, and then mostly noise. The metric that earns its place on a weekly report is:

• Duplicate creation rate = new duplicate pairs created this week, segmented by the source system that created the newer record. Starting default: effectively zero from automated sources — any sustained nonzero rate from a single integration is a defect, not a data-entry problem.

The segmentation is the whole point. Duplicates don’t appear uniformly across a CRM; they pour out of one place — an outreach tool with different matching logic than your marketing platform, an enrichment vendor that ignores your dedupe key, a webform with no validation. Reported in total, the metric says “we have a duplicates problem.” Reported by source, it says “the LinkedIn sync is creating duplicates because it matches on name instead of email.” One of those is a complaint; the other is a bug report you can close.

3. Freshness: is the pipeline alive?

Stale data is the quality problem that most directly corrupts the forecast, and it has two good metrics:

• Pipeline freshness = open deals with any logged activity inside your staleness budget ÷ all open deals. The budget should scale with your cycle — our starting default is 45 days for a roughly 60-day cycle, and we expect 85%+ of open deals inside it.
• Median days-since-last-activity across open deals. No threshold needed; this one exists for its trend. A median that creeps from 12 days to 19 over a quarter is telling you the pipeline is filling with deals nobody is working, even while the count of “stale” deals looks stable.

Use both. The percentage catches the tail; the median catches the drift. The mechanics of acting on staleness — triage lists, evidence-of-life, archiving — are covered in our pipeline hygiene guide.

4. Validity: do the values follow the rules?

Validity is the least glamorous dimension and the cheapest to automate:

• Validity rate = records passing all format and picklist rules ÷ all records, computed per object. Starting default: 98%. Rules include: emails that parse, phone numbers in a consistent format, picklist fields containing actual picklist values (not free-text leakage from an old import), dates that aren’t obviously wrong (close dates in 1970, founded dates in the future), and required-field rules per stage.

Validity violations are almost never created by humans typing — they arrive in bulk from imports and integrations. That makes the per-rule breakdown more useful than the aggregate: a sudden batch of 400 contacts failing the email format rule has one cause, and the created-date plus source field will name it. Our full rule set is in the CRM audit checklist.

5. Consistency: does the CRM agree with the systems that don’t lie?

This is the metric most teams skip and the one we’d defend hardest. Your billing system is ground truth — invoices either went out or they didn’t. So:

• Closed-won count agreement = closed-won deals in the CRM with a matching invoice or subscription in billing ÷ all closed-won deals, per period. Starting default: 100%, because anything less means the CRM contains revenue that doesn’t exist.
• Closed-won amount agreement = how closely deal amounts match invoiced amounts on the matched set. Our starting default is matched amounts within 5% of invoiced; discounting and proration explain small gaps, but persistent large ones mean reps are closing deals at aspirational numbers.

Consistency checks are powerful precisely because they’re external. Every other metric on this list can be gamed by editing the CRM. This one can only be fixed by making the CRM true.

Why we don’t use a blended health score

The single-number health score is seductive — executives like one number, vendors like one dial — and it destroys exactly the information you need. A blended score of 85 can hold perfectly steady while duplicate creation triples, as long as completeness improved at the same time. The averaging is the problem: it converts five answerable questions into one unanswerable one.

What we report instead, per rule: the count of violations, the trend against last week, and the evidence — the actual records that fired the rule, with the field values that triggered it. A findings list with evidence ends arguments; a score starts them. When someone asks “why did the number drop,” the answer should be a list of records, not a shrug at a formula.

Trends beat absolutes

The absolute value of any of these metrics matters far less than its movement. An 88% freshness rate is neither good nor bad in isolation — it depends on your cycle and your definition. But a freshness rate that fell six points this week is unambiguous: something changed. A spike in any single metric is best read as a process bug report. New duplicates spiked? An integration changed its matching behavior. Validity dropped? Someone ran an import. Completeness fell on one team? A new rep wasn’t onboarded onto the deal-entry rules.

This is also why measurement has to be automated and scheduled — weekly, in our practice. Trends require a baseline and a cadence, and a quarterly manual audit gives you neither. The same logic drives the continuous re-audit step in our CRM cleanup process: you’re not measuring to admire the number, you’re measuring to detect drift while it’s still one week’s worth.

Give one or two metrics teeth: tie them to the forecast meeting

Metrics nobody is accountable to are decoration. The fix is to wire one or two of them directly into a meeting that already matters — for sales-led teams, the weekly forecast call.

The metric we’d choose: percentage of open pipeline passing all hygiene checks — current next step, future close date, activity inside the staleness budget, valid stage. In our experience this composite is the strongest single indicator of forecast quality, because a deal that passes every check simultaneously is a deal someone is actually working with real information, and a deal that fails any of them is a guess wearing a stage label. Run the check the day before the forecast call, and make the rule explicit: deals failing hygiene get discussed as risks, not counted as commit.

Once one metric has consequences, the others improve as a side effect, because the underlying behavior — keeping deals honest — is the same behavior every metric on this list is measuring from a different angle.

Where to start

Don’t build a dashboard with all five dimensions on day one. Pick open-deal completeness and pipeline freshness, automate them this week, and add the rest as each one stabilizes. If you want the measurement layer without building it yourself, our open-source toolkit computes these checks against HubSpot and Salesforce with evidence attached to every finding — but the discipline transfers regardless of tooling: explicit rules, weekly cadence, trends over absolutes, and evidence behind every number.