How do we actually measure whether our CRM data is any good?
Score the fields your decisions depend on, not every field. For each one, measure fill rate, validity, consistency with the same fact in another system, and freshness. Track them monthly and break them out by how the record was created, because office staff, technicians on mobile, web forms and bulk imports fail in different ways. A single aggregate quality score hides the one broken field that matters.
Work backwards from decisions
Nobody has the appetite to fix every field, and most fields do not matter. Start from the decisions you make with data: where to spend marketing budget, who to coach, which customers to contact, how to staff next month. List the fields each decision depends on. That list is usually under twenty fields, and it is the only list worth scoring.
The advantage of deriving it this way is that it also tells you what a failure costs, which is what gets a cleanup project funded.
Four measures per field
- Fill rate. What share of records have any value. The easiest to measure and the least informative on its own.
- Validity. What share of values are well-formed: a phone that has the right number of digits, an email that could deliver, a date that is not in 1900, a picklist value that is still on the list.
- Consistency. What share agree with the same fact held elsewhere. Customer name and address in the CRM versus the field service system is the standard check, and disagreement rates above a few percent usually mean a sync problem rather than a typing problem.
- Freshness. How long since the value was last confirmed. A phone number nobody has dialed in four years is unverified, whatever the fill rate says.
Segment by entry path or the number tells you nothing
A blended sixty percent fill rate could be ninety percent from the office and ten percent from mobile, or it could be uniform. Those call for completely different interventions: a form change, a training session, or an import mapping fix.
Break every measure out by creation source. It roughly triples the size of the report and it is the difference between knowing you have a problem and knowing what to do on Monday.
A field can be fully populated and useless
Fill rate is the metric that gets gamed fastest. Make a field required and it will be one hundred percent filled by the end of the week, largely with the first option in the list, N/A, or whatever the last record had.
Add a distinctness check: how many unique values, and what share sits in the single most common value. A required field where eighty percent of records share one value is not populated, it is defaulted. Watching that ratio over time catches the degradation early, and it belongs in the same routine monitoring that any integrated platform should be doing on the data it depends on. Feeding AI on fields like that is where most disappointing results in AI integration actually originate.
Topics: data quality · metrics · data hygiene · reporting
Have a version of this question about your own business?
The useful answer usually depends on which systems you run and how they're connected. That's a conversation, not a blog post.