Before you change anything
- Take a full export. Every step below is reversible only if you can compare against the starting state.
- Agree what a duplicate is for your business: same company domain, same contact email, or a fuzzier match on name plus employer.
- Agree which record wins a merge, and write it down. Most-recently-modified is a common default and a poor one.
- Decide what happens to records that fail every check, rather than deciding it individually five hundred times.
Pass one: standardise fields
- Normalise company names: legal suffixes, punctuation, casing, trading names.
- Normalise domains to a single form, stripping protocol and subdomain where appropriate.
- Standardise country, region and state values against one list.
- Reduce free-text roles and industries to your agreed picklists, and record the original value somewhere.
- Fix phone and postcode formats so they sort and match predictably.
Pass two: remove duplicates
- De-duplicate companies before contacts, or the same contact merges under two parents.
- Merge rather than delete, so activity history and ownership survive.
- Resolve conflicting field values using the rule you agreed, not case by case.
- Re-run the match after merging: merges create new duplicates when two parents held near-identical children.
Pass three: fill the gaps
- Identify the fields that are actually required for segmentation and routing, and ignore the rest.
- Research missing values for accounts that matter rather than for everything.
- Mark unresolvable gaps explicitly. An empty field and a confirmed-unavailable field are different facts.
Pass four: verify contact data
- Re-verify email addresses last, once each contact appears exactly once.
- Carry the result as a status on the record: Valid or Invalid.
- Keep Invalid addresses out of your sends, but keep the record and the account context it sits on.
- Where no email was found, mark the field Not found rather than leaving it blank or guessing a pattern.
Pass five: check it against how the team works
- Run your three most-used segments and confirm the counts are plausible.
- Check ownership and routing rules still resolve after merges.
- Confirm reporting that depended on the old field values has been updated.
- Agree the maintenance cadence now, while the database is clean, rather than after it drifts again.
Put it into practice