The minimal match keys and a simple confidence score
For small teams you want a handful of reliable match keys that cover the usual duplicate causes (forms, imports, manual entry) without creating noise. Start with 3–4 keys and a points-based score: canonical email, normalised phone, postcode + surname, and company domain when B2B applies.
- Canonical email = 50 points (lowercase, strip tags like +, trim). If email exact match → strong signal. Works across HubSpot, Salesforce, Marketo and Pardot.
- Normalised phone = 30 points (remove spaces, +44→0 or vice versa; compare last 9 digits). Good fallback for contacts without email.
- Postcode (UK normalised) + last name = 25 points (normalise spacing and upper/lower). Useful for household duplicates in small local lists.
- Company name + domain = 40 points (trim common suffixes like Ltd, Ltd.). Use for company-level merges in B2B.
Score example: treat a candidate pair with ≥90 as an automatic merge, 60–89 as human review, <60 as no action. That simple bucketed confidence model balances safety and effort without heavy tooling.
Safe merge policy, rollback and lightweight monitoring
Define a clear merging policy before you touch live records: which record becomes the canonical one (most recent activity or most complete fields), which fields are merged vs preserved, and how to record provenance. Always add a small provenance field (e.g. "merged_from: yyyy-mm-dd;rule=X") so you can trace what happened.
Before any automated merge, snapshot the pre-merge state: export the two records to a CSV or save a short JSON blob into a sheet. That export is your rollback copy—if a merge goes wrong you can use it to reconstruct or manually split records. Many CRMs (HubSpot, Salesforce) let you view merge history; keep the exported snapshot anyway for spreadsheets and middleware merges.
Lightweight monitoring: create a saved view that shows recently merged records and a daily duplicate-count metric (simple SQL not required—use a list or a sheet). Run a weekly sample audit of 20 recent auto-merged pairs; if >1 in 20 are incorrect, tighten thresholds or pause auto-merges. If you operate on the South Coast or locally in Fareham, this is a quick way for small teams to keep CRM optimisation visible to the team that uses the data daily.
Afternoon‑ready implementation recipes and quick tests
Spreadsheet-only (no dev): export contacts into a sheet, add columns for canonical_email, phone_normalised, postcode_norm, company_domain. Use formulas to compute points per rule and a total score. Add a "recommended_action" column with simple IF logic: >=90 "auto-merge", 60–89 "review", else "skip". For merges, create a second sheet that lists merge pairs and use it as a manual task list to apply in your CRM.
CRM automation or middleware: in HubSpot or Salesforce you can run automations in "audit-only" mode—create a workflow that writes a "merge_suggested" tag instead of performing the merge. In Zapier/Make or a lightweight script, implement the same scoring and only call the CRM merge API when score≥90. For Marketo/Pardot, use matching rules to flag duplicates and export suggested merges for a human reviewer.
Quick tests before production: 1) Dry run on a 100-record sample: label suggested actions but don’t merge; measure precision (true positive rate) by manually checking 30 suggested auto-merges. 2) Run an audit-only week: let workflows flag "merge_suggested" and let users review the queue—track time-to-review and false positives. 3) Do a controlled auto-merge of 50 low-risk pairs, snapshot before, and run your weekly sample audit. If you prefer not to do this alone, Optira can help set the rules, run the dry runs and hand you a safe, documented process for ongoing CRM data cleanup (see our HubSpot CRM data cleanup service for an example approach).