A CRM record is good enough when the process reading it does the right thing. A missing job title barely matters to billing. A wrong billing contact matters a great deal. Quality is judged against use, which is why a single "data quality score" for a whole CRM tells an operations leader very little.
What dimensions define CRM data quality?
In October 2013 a DAMA UK working group published six core dimensions for assessing data quality. Each one asks a different question of a CRM.
| Dimension | The question for a CRM | How operations feels a failure |
|---|---|---|
| Completeness | Are the fields a process needs filled in? | Orders stall waiting for a billing contact |
| Uniqueness | Does each customer exist once? | Two reps call the same account; credits land on the wrong record |
| Timeliness | Does the record reflect the customer now? | Invoices go to a person who left last year |
| Validity | Do values follow the rules and formats? | Records fail integration into billing or ERP |
| Accuracy | Does the record match reality? | Service goes to the wrong address |
| Consistency | Does the CRM agree with other systems? | Billing and CRM show different customers for one account |
Which dimensions matter most to operations?
Three come up again and again.
Staleness is timeliness in practice. Records decay as people change jobs and companies restructure, and a record that was right at entry is wrong a year later without anyone touching it.
Duplicates are uniqueness failures, and they spread damage: split history, double outreach, payments and credits applied to the wrong account.
Ownership is not one of the six dimensions, but operations depends on it. A record with no living owner has nobody to confirm a change, chase a missing field or approve a merge. Track it next to the six.
All three surface first in the back office, where billing, collections and service teams work from the CRM's version of the customer and absorb the cost when it is wrong.
How do you run a CRM data quality assessment?
The DAMA UK paper outlines a typical approach, which maps well onto a CRM:
- Identify the data items critical to operations and reporting, such as billing contact, legal entity, account owner and status.
- Choose the dimensions that apply to each, and how much each one weighs.
- Define what good and bad look like for each dimension.
- Apply the checks to the data.
- Review the results and decide whether quality is acceptable.
- Correct the data and the process that produced the errors.
- Repeat on a schedule to see the trend.
One point from the paper is worth holding onto: accuracy can only be assessed against the real thing or an authoritative reference. A CRM cannot check its own accuracy. Compare samples against the billing system, signed contracts or the customer.
Where does AI change the assessment?
Checking accuracy and consistency at scale used to mean people comparing records by hand. AI can compare every CRM record with the billing system and the contract, read free-text notes and emails for signs of change, and score each record on the dimensions above. Fixing what it finds is a separate job: the running practice is CRM data hygiene, and the engineered system that does it continuously is CRM data cleansing.