By Jonas AnderssonCo-founderJun 20, 2026

5 min readJun 20, 2026

How-to

How to stop duplicate CRM records at intake

Duplicate CRM records feel like a database problem because the pain appears inside the database. In practice, the cause is usually earlier: unclear intake rules, weak matching, and too many places where new records can be created without checking what already exists.

The fastest fix is not another cleanup sprint. It is a small intake contract that decides when a record is new, when it should match an existing account, and what happens when the system is not sure.

Start upstream

A duplicate record is usually created before anyone sees it.

Most teams clean duplicates after they have already damaged reporting, ownership, and follow-up, every duplicate can spawn extra tasks, split communication history, and make pipeline numbers less trustworthy. Industry data-quality research routinely puts duplicate or fragmented records at 20-30% of a typical B2B CRM database.

The better move is to treat intake as a gate. Before a lead, company, account, or contact becomes official, the system should check the few fields that reliably identify it and decide whether the new submission is a match, a possible match, or new.

Where new CRM records are created: forms, imports, integrations, manual entry, and enrichment tools.

Which fields are reliable enough to match on: email, domain, company number, account ID, or normalized company name.

Who owns ambiguous matches instead of letting every user make a different judgement.

What happens after a match: merge, update, reject, or route for human review.

In this articleTap to open the section map

Overview
Start upstream
Minimum rule set
The matching, one level down
Implementation path
Before the gate
In motion

Minimum rule set

Three rules catch most duplicate trouble without making intake heavy.

You do not need a perfect master data program to make progress. You need enough structure that the same account cannot quietly enter the workflow in four different shapes.

Normalize before matching

Lowercase emails, clean domains, remove legal suffix noise, and trim whitespace before comparing records. Most apparent duplicates are the same account written three different ways, normalization catches them without any fuzzy logic at all.

Separate exact and fuzzy matches

Exact matches can be handled automatically. Fuzzy matches should be routed to a review queue with the reason shown plainly, "same domain, different contact name" is actionable; a numeric confidence score is not.

Protect the owner field

Ownership should not be overwritten just because a duplicate or import arrives with a different assigned person. Silent ownership changes are how accounts fall through, the original owner stops seeing it and the new one does not know they have it.

The matching, one level down

Good fuzzy matching combines measures and looks for a reason to say no.

"Separate exact and fuzzy" is the rule; this is what fuzzy has to actually do. A single similarity score, taken alone, both misses real duplicates and invents fake ones. A few habits fix most of it.

Combine complementary measures. Edit distance alone rejects a reordered name: "John Smith" and "Smith, John" look far apart character by character, so pair it with a token-set measure that ignores word order. Each measure covers the other's blind spot.

Look for a reason to say no, not only yes. Any shared value can "prove" sameness if you only look for agreement. Before auto-merging, check for a disqualifier: two records with the same company name but different company numbers are two companies, not one.

Guard the placeholder. A shared info@ address or a blank company is not a match signal; block on it and you merge every contact at the company into one. Skip the values that thousands of rows share.

Leave the ambiguous ones unproven. A partial overlap should route to the review lane, not silently unlock a merge. Proven pairs merge; unproven pairs wait for a human.

Implementation path

A calm duplicate-control build usually follows four steps.

The point is not to freeze the CRM. The point is to make record creation predictable enough that sales, service, and reporting stop fighting the same data problem every week.

01
Inventory every intake source
List every place a record can be created or imported. Do not skip manual entry, because it is often where the quiet exceptions live.
02
Choose the matching keys
Pick the smallest set of fields that can reliably identify the record. If a field is often missing or messy, it should not carry the whole match.
03
Create a review lane
Ambiguous records should not silently create duplicates. Put them in a small queue with the likely match and the reason for uncertainty.
04
Measure recurrence
Track which source creates duplicates and which rule catches them. That shows whether the fix is working or one source still needs attention.

Inventory every intake source
List every place a record can be created or imported. Do not skip manual entry, because it is often where the quiet exceptions live.
Choose the matching keys
Pick the smallest set of fields that can reliably identify the record. If a field is often missing or messy, it should not carry the whole match.
Create a review lane
Ambiguous records should not silently create duplicates. Put them in a small queue with the likely match and the reason for uncertainty.
Measure recurrence
Track which source creates duplicates and which rule catches them. That shows whether the fix is working or one source still needs attention.

Before the gate

An intake gate stops new duplicates. It does not tell you how much mess already got in.

The gate is the forward fix. Sizing the backlog already sitting in the CRM is a separate, faster pass worth doing first.

Size the duplicates already inside before building the gate that blocks new ones.

Run a CRM data audit in a day ->

In motion

What the intake gate surfaces as it runs.

intake-gate · live

0 changes shown

The useful metric is not how many duplicates you cleaned. The useful metric is how many duplicates never entered the workflow. Once the intake gate works, cleanup becomes maintenance instead of a repeating project.

A useful starting point

Want the intake map checked?

Bring one intake source, the fields it sends, and a few recent duplicate examples. That is enough to see whether the fix is rules, routing, or a small build.

Send CRM intake briefOpen the Data Auditor

Bring the page, report, or workflow as it is now.

We reply with the clearest next step, or an honest no.

Key takeaways

Duplicates form at intake when multiple sources add records without a matching process.
Three rules catch most problems: normalize before matching, separate exact and fuzzy matches, protect the owner field.
Fix at intake, deduplication campaigns are cleanup, not cure.

If you only remember one thing

A duplicate that never enters is better than a duplicate that gets merged later.

Data & Visibility CRM Data quality Intake

Continue the thinking

Part of this guide

Data Visibility for Commercial Teams: The Complete GuidePrincipleA derived path to the complete guide for this systems problem.15 min

Builds on

How to Audit Your CRM Data in a DayMethodUses the audit findings to focus duplicate prevention on the fields and handoffs causing damage.7 min

Normalise phone numbers to E.164MethodExplores a neighboring systems problem from a different operational angle.8 min

CRM vs Spreadsheet: When to Move OnPrinciple5 min

Sources

References behind this piece

Data Quality Trends: The State of Data Quality in 2021
Salesforce
Referenced for data on duplicate record rates and the downstream cost of fragmented contact and account data in CRM systems.
2021 Global Data Management Research
Experian / 2021-06-01
Used for data on how duplicate records affect data quality scores, reporting accuracy, and operational cost in B2B commercial databases.
Data Quality Assessment: The Cost of Bad Data
IBM Institute for Business Value / 2016-01-01
Background data on how data duplication contributes to the broader $3.1 trillion annual cost of poor data quality.

Written by

Jonas Andersson

Co-founder, Mannvit Labs

Co-founder of Mannvit Labs, leading systems, data, automation, and technical delivery. He spent four years before this running e-commerce for a Swedish retailer, from SEO to campaigns.

He builds the internal tools and data workflows these guides come out of. See the builds.

Commercial data & reporting
CRM data quality
Process optimization

Connect on LinkedIn More from Jonas

How to stop duplicate CRM records at intake

The fastest fix is not another cleanup sprint. It is a small intake contract that decides when a record is new, when it should match an existing account, and what happens when the system is not sure.

A duplicate record is usually created before anyone sees it.

Want the intake map checked?

Bring one intake source, the fields it sends, and a few recent duplicate examples. That is enough to see whether the fix is rules, routing, or a small build.

Bring the page, report, or workflow as it is now.

We reply with the clearest next step, or an honest no.

Jonas Andersson

Co-founder, Mannvit Labs

Co-founder of Mannvit Labs, leading systems, data, automation, and technical delivery. He spent four years before this running e-commerce for a Swedish retailer, from SEO to campaigns.

He builds the internal tools and data workflows these guides come out of. See the builds.

Commercial data & reporting

CRM data quality

Process optimization

How to stop duplicate CRM records at intake

A duplicate record is usually created before anyone sees it.