Define the decision and grain
Start with the report or workflow that needs trustworthy data. An account coverage question requires reliable company relationships. An enquiry conversion question requires a consistent event and duplicate rule. A revenue report needs appropriate opportunity and financial definitions.
Salesforce’s duplicate-management guidance explains why duplicate records can undermine CRM use. Duplication is one dimension of quality, alongside missingness, validity, consistency and timeliness. Inspect the dimensions that could change the decision rather than pursuing abstract perfection across every field.
Use a focused quality register
| Check | Example failure | Decision risk |
|---|---|---|
| Identity | One company has several unlinked records | Account coverage is understated |
| Completeness | Dispositions are missing | Rejections appear to be inactivity |
| Validity | Impossible stage dates | Cycle-time measures are distorted |
| Consistency | Teams use different stage meanings | Conversion rates are not comparable |
| Freshness | Former contacts remain active | Outreach and coverage are misleading |
| Lineage | Source fields are overwritten | Acquisition history cannot be reconstructed |
Choose a representative sample and trace it to source events. Automated checks can reveal patterns, but manual inspection is often needed to understand what a status actually means.
Distinguish missing from negative
An empty buying-role field may mean unknown, not absent. A blank disposition may mean no follow-up, an integration failure or incomplete recordkeeping. If the report treats all blanks as one outcome, it can suggest a commercial explanation that the data does not support.
Repair causes as well as records
A one-time cleanup can improve the current dataset while new errors continue entering it. Inspect form rules, integrations, matching, training and ownership. Define when a duplicate warning should block creation and when a legitimate exception requires review. Preserve records and relationships during any merge process.
Report quality with the result
Show material coverage gaps and exclusions alongside the metric. If a large part of a cohort lacks dispositions, say that progression is unknown for those records. Do not imply that an incomplete dataset represents all activity.
Assign an operating owner
Define who can correct data, approve a definition change and monitor recurrence. Keep a change log so historical comparisons remain interpretable. A quality program becomes useful when a team can trace a questionable number, resolve the issue and prevent it from silently returning.