Worked example: duplicate CRM records Fictional example for sales and data teams. The goal is to prevent the same active customer from being counted more than once in reports. 1. Define the rule Each active customer record should represent a distinct customer. Agree matching criteria with Sales. Use a normalised tax identifier where available and reliable. Send name or email matches for manual review. Track records that cannot be evaluated separately. 2. Measure the baseline In a fictional sample of 1,000 evaluable records, 30 excess rows are confirmed. In each duplicate group, count only rows beyond the one retained. Uniqueness = (1,000 − 30) / 1,000 × 100 = 97%. This example uses a threshold of at least 99%. 3. Act when the rule fails The data steward logs an issue, retains the candidate matches and checks ambiguous cases with Sales. The data owner approves which record to keep. The technical team merges confirmed duplicates, preserves order references and logs the change. If an import caused the problem, add a check before the next load. 4. Measure again After removing the 30 excess rows, rerun the same rule. This example leaves 970 evaluable records and zero excess rows: (970 − 0) / 970 × 100 = 100%. Schedule another check after the next import. Record how many records were excluded and why. 5. Keep the evidence Copy and complete: dataset · date · rule and version · evaluable records · excluded records · excess rows · result and threshold · issue · owner · corrective action · retest result · next review.