AI Marketing Automation

Clean your CRM data and stop duplicates coming back

 ·  12 min read

An operations specialist sorts contact records during a CRM data cleanup
In this article

CRM data cleanup starts with a saved copy, clear matching rules and control over the systems that create records. Identify duplicates, decide which values to keep, test a small set, then check the result before expanding. Fix repeated writes and import rules at the same time, or the next automation run can recreate the mess.

The goal goes beyond a smaller contact count. Sales needs the right history, the right company and a clear next action. Marketing needs reliable sources and outcomes. A merge that loses those details can make a tidy database less useful than the untidy one you started with.

This guide fits a business whose CRM collects forms, ad leads, spreadsheet imports or integration events. It covers both existing duplicate records and the workflow faults that keep creating them. You can apply the decision process across tools, but check the exact merge and import behaviour in your own CRM before you change live records.

Find which systems can create a record

List every route that can create or update the affected CRM object before you clean it. Include forms, integrations, imports, scheduled jobs and manual entry. Trace a sample duplicate pair back to those routes. The record count tells you the size of the problem; the write history helps explain how it happens.

Start with the object that affects a current decision. That might be contacts receiving duplicate outreach, companies splitting pipeline reports or leads entering two sales queues. Avoid combining every data problem into one first pass. A company merge and a contact merge may carry different rules and consequences.

For each route, record its input, matching field, destination and response to a failed request. Check whether two active workflows listen to the same event. An old workflow can still write records after a newer one replaces it.

In a hypothetical setup, a form integration and a separate automation both create a contact from the same submission. Cleaning yesterday’s duplicates leaves both writers active. Fix the overlap or give the two routes clearly different jobs before importing the corrected data. Our CRM automation workflow guide covers the wider workflow choices; this cleanup should focus on record identity and safe writes.

Save a copy and define the first batch

Save the records, stable identifiers, associations and fields you need to reconstruct decisions before changing data. Keep the export access-controlled and record its time. Then choose a limited first batch with clear matching evidence. A backup supports investigation, but it does not guarantee that your CRM can reverse a merge or restore every relationship.

Export the fields that drive business actions, including the current contact or company identifier, source information, lifecycle stage and relevant associations. Include the fields you plan to change. Check whether the export captures activity history or whether you need a separate supported method to retain that context.

Choose a pilot set that includes more than easy exact matches. Add a record with conflicting values, one with an open deal and one that an integration still updates. Test those cases somewhere that cannot trigger real outreach. Keep the pilot outside live sales sequences.

Agree which actions need human review. A proposed field-format correction differs from combining two people into one record. Keep a separate queue for uncertain matches. The business can continue handling known contacts while someone resolves those exceptions; uncertainty does not require an all-or-nothing cleanup.

A clerk locks a second set of customer files as red threads trace incoming records back to two desks.

Decide what counts as the same person or company

Choose matching rules for each object instead of using one rule across the whole CRM. An exact record ID identifies an existing record within that system. An email or company domain can help match an import, but neither automatically resolves every business relationship. Keep uncertain pairs for review rather than forcing a match.

Names alone make weak merge evidence. Two people may share a name, and one person may use different names or addresses over time. A shared reception email may represent several people. A group of companies may share a domain while still needing separate account records for sales and reporting.

Evidence Suggested treatment
Same existing CRM record ID Update that record through the supported route
Same source record ID replayed Check the prior write before creating anything
Exact email match with consistent person details Review against your contact-matching policy
Similar names with different emails Keep separate until stronger evidence supports a match
Same domain with different legal or buying entities Review the company structure before merging
Missing identifier Send to an exception queue instead of inventing a match

HubSpot supports import deduplication through Record ID and custom properties that require unique values. It also documents object-specific behaviour for email and company domains. Read its deduplication rules for your import path. Do not assume that an API integration uses the same matching logic as a CSV import.

Set field rules before merging records

Decide which values should survive before approving a merge. Treat current contact details, historical source, deal relationships and communication preferences as separate decisions. The newest value is not automatically the best value. A recent import can contain older data, and an empty field should not silently replace a useful one.

Write a rule for each field that affects routing, reporting or communication. For a hypothetical B2B team, a verified direct phone number may take priority over a recently imported switchboard number. The first recorded enquiry source may need to remain intact while a separate field stores the latest campaign interaction.

Do not let a cleanup expand permission to contact someone. Preserve suppression and preference records, and send unclear conflicts for review under the business’s existing policy. A merged email address does not establish permission to use it. Keep the rule explicit enough that a developer cannot interpret a blank as approval.

Check associations with open deals, companies and tasks. Ask whether a merge changes which record an integration updates next. HubSpot explains that merges combine activities, associations and most properties, with details that depend on the object. It also states that users cannot unmerge the records. Review its merge documentation before approving changes you expect an export to undo.

Separate formatting fixes from identity decisions

Standardize fields that need consistent formats, but keep those corrections separate from matching and merging. Trimming accidental spaces or applying an agreed phone format can improve usable data. Those changes do not prove that two records describe the same person. Record what each transformation changes so you can inspect unexpected results.

Start with simple rules that the business can explain. Define a country field consistently. Keep an original source value beside a normalized reporting category when that history matters. Decide how unknown values should appear instead of replacing them with guesses that make the completion rate look better.

Avoid inferring a person’s role, company size or buying status from weak clues. Enrichment can suggest information, but your process still needs a source and a review rule for consequential changes. A clean-looking field filled with an unsupported value can send a lead to the wrong queue.

Our lead-tracking guide explains why consistent source records matter before you scale spend. During cleanup, check a few complete enquiry paths rather than only the spreadsheet. Confirm that the original receipt, current CRM record and reported outcome still describe the same event.

A magnifier reveals different portraits on two contact cards that show the same storefront.

Use imports that update the intended records

Build the import around stable identifiers and the exact object you intend to change. Test a small file, inspect errors and compare the resulting records before loading the rest. Make the creation policy explicit. A cleanup import should not accidentally create new contacts because a row failed to match an existing record.

HubSpot’s import-file guidance explains how to include Record ID or a custom unique-value property when updating records. Use the identifier from the destination account. A row number from a spreadsheet or an ID from another CRM does not automatically identify the same record there.

Map every column deliberately. Check that a company field does not land on a contact property and that a text value does not overwrite a date or selection field. Retain a copy of the exact uploaded file and the mapping so a later reviewer can reproduce the change.

Treat unmatched rows as a result to inspect. HubSpot documents an update-only error when identifiers do not match existing records. Its import error guide explains that those affected rows do not create new records. Investigate the mismatch before switching the job to create-and-update just to remove the error message.

Stop the same event from creating another record

Give each incoming event a stable identity and check what happened to it before creating another record. A retry should find or complete the original operation. It should not assume that a timeout means the destination saved nothing. Test this behaviour across every active route that can handle the same event.

Zapier distinguishes trigger deduplication from action behaviour. Its duplicate-data guidance says action behaviour depends on the receiving app. A trigger that recognizes old events does not guarantee that a create action or a second workflow will avoid duplicate records.

For a custom integration, ask for a documented source-event key and a destination-record reference. Keep the key stable during a retry. Where the platform supports a unique field or an atomic update-or-create operation, use it carefully. A separate search followed by create can still race when two workers process the same event together.

Suppose an example form submission has event ID FORM-1042. The CRM saves the contact, but the integration loses the response. A safe recovery checks the stored mapping or unique key before creating anything else. The expected result is one business record with a recorded retry.

If several tools write to the CRM, agree which system controls each field. Our custom software service covers integrations built around existing operations. A clear write policy matters more than adding another automation tool to an already unclear process.

Test the awkward cases before expanding

Test normal creation, updates, duplicate events, conflicting values and destination failures before expanding the cleanup. Check the final records and relationships, not only a green success label in the automation tool. Include a concurrent retry test when two jobs can run together. Keep test records clearly separate from real sales activity.

Test Expected business result
New valid enquiry One correct record and the intended association
Same event arrives again No extra business record
Existing person submits a new enquiry Correct record update and separate enquiry history where needed
Destination times out after saving Recovery finds the prior write
Two workers receive the same event The identity rule prevents duplicate creation
Import contains an uncertain match Review queue or clear error, without an automatic merge

Check alerts as well as successful writes. Someone needs to know when a record waits for review or a destination rejects it. Keep the original event available through an approved retention process so the team can investigate without asking the buyer to submit again.

Our guide to scoping an AI project shows how to turn these expectations into acceptance checks. If the workflow needs outside help, define the tests in the scope before the build starts. Do not accept “deduplication enabled” as a substitute for a working test.

An operator sends two red tokens down separate paths that lead to a single contact card.

Measure whether the problem stays fixed

Track new duplicate candidates, confirmed duplicate creations, unresolved exceptions and failed writes after the cleanup. Keep each measure separate. A falling contact count only shows that records disappeared or merged; it does not prove that sales has better data. Check real follow-up and reporting decisions alongside the technical counts.

Use a fixed review period and a clear denominator. For example, a hypothetical team processing 200 incoming events finds four confirmed duplicate creations. That is a 2% event-level duplicate creation rate for that sample. A second run with zero duplicates still needs the event count and observation period before anyone claims lasting improvement.

Keep measuring the original pain. If two salespeople still contact the same buyer, inspect assignment and notification rules even when the contact table has no duplicates. If source reports still disagree, check attribution and enquiry history rather than merging more records to force a match.

Before approving outside work, use our automation cost guide to separate a one-time repair from ongoing support. Bring a redacted example pair, the list of record writers and the failed path to a tracking and funnel audit. That gives the review a specific problem to solve without handing over your whole database at the first conversation.

FAQ

What should a CRM data cleanup include?

Include a saved baseline, an inventory of record writers, matching rules, field-survival rules and a tested change plan. Check associations and reporting after the changes. Then repair the imports or integrations that created the problem. A one-time merge can reduce existing clutter while leaving the same fault active.

Can we merge contacts that share a name?

A shared name alone does not establish that two records describe the same person. Compare stable identifiers and relevant history under an agreed matching policy. Keep uncertain pairs separate until someone resolves them. Shared inboxes and changing email addresses also need review rather than a blanket automatic merge rule.

Does Zapier prevent every duplicate CRM record?

No. Zapier distinguishes trigger deduplication from the destination app’s action behaviour. A second workflow or a receiving app that allows duplicates can still create another record. Test repeated events and retries across the full route. Verify the final CRM records rather than relying only on the trigger’s success status.

Can we undo a HubSpot merge?

HubSpot states that users cannot unmerge records. Its supported recovery options do not recreate every prior state automatically. Review the effects on properties, activities and associations before merging. Save the evidence you need and test a small set first; do not describe a CSV export as a guaranteed undo button.

Should we replace the CRM to fix duplicate data?

First find which routes create duplicates and how they identify existing records. Replacing the CRM can carry the same faulty imports and overlapping workflows into a new system. Compare a scoped repair with replacement only after you understand the current failure, required features and cost of moving useful history.

Free AI Automation Snapshot

Find the work worth automating first

Tell us how your team works today. We map the first workflows worth automating and reply within 24 hours.

Your data is safe. We never share details without permission.

Free growth audit

Find out what's costing you leads

We check your site, ads and tracking and send the fixes that matter most within 24 hours. Your first 15 days are free on a new plan.