FIELD NOTE · Database Marketing
Measure a cleanup project without hiding the hard cases
A smaller duplicate count is useful only if the same test was applied.
Freeze the starting definition
Record what counts as a duplicate candidate, a missing required value, and an unresolved exception. Save the extraction date and population size. Without that baseline, a later percentage can improve simply because difficult records were removed from the report or the matching rule became less demanding.
Separate detected, reviewed, and resolved
Imagine 120 candidate pairs. After review, 70 are confirmed duplicates, 30 are distinct, and 20 remain uncertain. Reporting “120 duplicates removed” would be wrong. Report the candidate volume, review decisions, and actual changes separately. Pair counts are not always unique record counts: one record can appear in multiple pairs.
Track the work that remains
Keep an exception queue with an owner and next action. Some issues need a customer response, another source, or a business decision. A responsible project can end with documented unresolved cases rather than an artificial perfect score. Estimate the ongoing effort so that the next import does not recreate the same problem.
Connect improvements to a real workflow
Measure whether a user can complete the intended task with fewer corrections or less review time. A test sample can be compared before and after using the same procedure, while acknowledging other changes that may affect the result. Avoid claiming revenue impact from a data-quality percentage alone. Publish only synthetic or properly permissioned examples, not identifiable customer records.
Prepared with AI assistance. Examples are constructed for explanation, not reported customer results. Corrections are welcome through the contact page.
← All guides