
Lead Management
Prevent CRM duplicates: verify identity and preserve every event
Prevent CRM duplicates safely by preserving each event, applying explainable identity rules and checking the master, fields and automations around a merge.
Short answer
Preserve every new form submission, call or import row as a separate event before treating it as a person. Normalise email and phone values without discarding the raw input, then look for candidate contacts through written rules. Automatically link only when one strong candidate exists and no protected information conflicts; shared or ambiguous identifiers belong in a visible review state. Merge records only after selecting the master contact and checking fields, history, open work, suppressions and active automations both before and after the change.
Key takeaways
- Preserve every form, call or import as an event before treating it as a person or CRM contact.
- Normalise email and phone transparently while retaining raw value, origin, verification and scope; never automate a name-only match.
- Allow exactly four outcomes: confirmed match, possible match, new identity or conflict stop with an owned review.
- Choose the master and field policy before an irreversible merge, then read history, suppressions, open work and active automations back.
A duplicate record is not a duplicate event
Consider one Monday morning. At 08:17, a prospect submits a quote form using her work email. Seven minutes later, she calls from a mobile number. That afternoon, a colleague imports a conference list containing the same person under a general company address. These are three events, perhaps two contact addresses and probably one person. A system that collapses those layers too early either creates three contact records or hides two useful interactions inside a silent update.
The distinction affects real work. The later call may carry a different question, greater urgency or a request for a callback. A contact match can connect that event to a known person; it should not erase the event. Nor does a matching name prove a shared identity. Two employees at one company can have the same name, a household can share a telephone number, and an address such as `info@` can represent several people.
The safe operating rule is simple: events may legitimately repeat, but people must not be created twice or merged without evidence.
Qualitative discussions in CRM and automation communities surface the same boundary cases: different phone formats fail to match, a returning contact re-enters as a cold lead, original source information is overwritten, or a merge restarts tasks and messages. Those accounts are useful for designing tests. They do not tell us how common the problem is and they do not validate one universal matching method.
Keep the person, event, business case and contact separate
A CRM contact is a record in a system. It is not the person themselves and it is not every interaction associated with them. A reliable intake process distinguishes at least four objects.
Object | What it represents | Unique evidence | What a match must preserve |
|---|---|---|---|
Person | a hypothesised real-world identity | a confirmed combination of appropriate identifiers | uncertainty and alternative identities |
Event | a particular form, conversation, message or import row | event ID, source and timestamp | raw values, content, permission and delivery context |
Business case | a particular enquiry, quote, appointment or service matter | its own case ID and status | owner, stage, next action and outcome |
CRM contact | the shared customer record in the system | immutable contact ID | profile data, history, suppressions and relationships |
An event can link to an existing contact and still open a new business case. Conversely, two confirmed contact records may represent one person and become candidates for a controlled merge. This separation prevents two common failures: a new need disappearing inside an old record, and one shared identifier connecting two different people.
The incoming event ID is not an identity key. It proves only that the same technical delivery should not be processed twice. The contact ID is not external identity evidence either; it identifies the record that is intended to survive after the decision.

A match-key register makes the rules explainable
“Match by email” is not a complete specification. An email value can contain casing differences, spaces or a typo; a telephone number may arrive with a country code, national prefix or separators. Normalisation should make equivalent representations comparable without inventing an identity. A match-key register therefore records how each identifier is handled and where it is valid.
Field | Example content | Purpose |
|---|---|---|
key type | email, phone, external customer ID, verified channel ID | prevent identifiers from being treated as interchangeable |
raw value | exact value received | preserve intake evidence and support later diagnosis |
canonical value | technically standardised representation | compare equivalent formats |
verification state | unknown, technically checked, confirmed by person | distinguish valid format from verified identity |
scope | person, company, location or shared household | expose identifiers that may be shared |
origin | form, call, import, employee or existing system | retain trust and purpose context |
first/last seen | timestamps | separate current from obsolete identifiers |
suppression or conflict | do not use, ambiguous or contradictory | stop dangerous automatic matches |
Normalisation should be narrow and reversible. For a phone number, whitespace and common separators can be removed and a plausible international representation formed for comparison. The raw value remains attached to the event. Broad provider-specific email transformations are risky: dots, plus addressing and local-part rules do not mean the same thing across every provider. A rule that collapses them “just in case” can join two genuine mailboxes.
An external customer or member ID can be a strong identifier, but only with its issuing system and tenant. The value `12345` from system A is not automatically the same identity as `12345` from system B. The issuer therefore belongs to the compound key.
Use four outcomes, not an uncalibrated confidence score
A label such as “87% match” looks precise even when no calibration set or error cost exists. A small team can operate a discrete decision model more reliably. Each incoming event ends in exactly one of four states.
State | Minimum evidence | Permitted action | Prohibited action |
|---|---|---|---|
confirmed match | exactly one candidate, strong rule passed, no protected conflict | link event to existing contact; open a separate business case when required | silently overwrite fields or source history |
possible match | one or more candidates, but identifier is shared, unconfirmed or incomplete | create one review card with comparison and missing decision | automatic merge |
new identity | no candidate under valid rules and sufficient minimum data | create a contact with source and event attached | claim that no match proves absolute newness |
conflict stop | strong identifiers disagree or protected states collide | preserve the event and escalate to a named owner | guess, overwrite or retry indefinitely |
A confirmed match is not a universal claim about the person. It means the current rule selects one contact for this particular purpose. A later process may require stronger identity evidence—for example, changing a contract is not equivalent to answering a general enquiry.
Review is not a failed automation. It is the intended outcome when the cost of a false merge exceeds the cost of a short human decision. To prevent a stagnant queue, each card needs an owner, reason, due date and the one question that remains unanswered.
Match rules must reflect the risk of each identifier
There is no universal “best” single field. The appropriate rule depends on the source, business model and consequence of error.
Candidate rule | Typical use | Automatic link? | Important limitation |
|---|---|---|---|
verified external ID plus issuer | an existing customer portal or governed source system | often possible if no conflict exists | ID may be imported under the wrong tenant or source |
canonical email plus canonical phone | a new lead carrying two independent identifiers | strong pilot candidate | both may be stale or shared by a household |
verified personal email only | a return through the same confirmed mailbox | depends on purpose and conflicts | role addresses, forwarding and reassigned mailboxes |
verified personal phone only | a return through a known number | depends on purpose and conflicts | family, business and reassigned numbers |
name plus company plus another identifier | B2B candidate search | usually review | job changes, spelling variation and namesakes |
name only | manual search | no | insufficient discrimination |
role inbox or shared number | service and B2B intake | no | identifier may belong to a group, not a person |
social profile reference | messaging under the connected business profile | only within that channel route | profile and display name are not general person proof |
Salesforce's documentation on matching and duplicate rules is a useful reference from another system. Its account example combines either account name and billing street, or website and phone. It does not describe ONE LOOP or prescribe a universal rule. It illustrates a sound design boundary: decide the object and identifier combination first, then separately decide whether a candidate is blocked or allowed with a warning.
A ONE LOOP pilot should begin with a short explicit rule set. A narrow compound rule may link automatically. Individual, shared or contradictory identifiers create review. Only audited production cases justify expanding or tightening that rule.
From a new event to a verified customer identity
From a new event to a verified customer identity
Secure the event
Preserve a unique event ID, raw values, source, timestamp and the specific request unchanged.
Normalise keys
Create canonical email or phone values while retaining raw input, verification, origin and scope.
Review candidates
Evaluate strong or compound rules, shared identifiers and protected contradictions.
Evidence contact state
Link the event or create or merge under control; read history and automation back from CRM.
Ambiguous identifiers, protected-field conflict, uncertain master or active work at risk?
Create one review card with candidates, conflict, owner and due date. Do not blindly link, overwrite or merge.
Recommended flow: preserve the event, normalise identifiers, inspect candidates under a rule and confirm exactly one contact state.
1. Secure the event unchanged
Store the event ID, timestamp, channel, raw values and relevant content before searching for or changing a contact. Process the same technical event once. A later event from the same person remains a distinct interaction and may open its own business case.
2. Normalise keys under a controlled rule
Create canonical comparison values using the documented method while preserving raw input. Record origin, verification state, scope and known conflicts. Passing a format check does not prove identity.
3. Build candidates and decide
Search only with rules approved for this event type. Store the rule version and candidate set. One strong candidate without protected conflicts becomes a confirmed match; no candidate can become a new identity; ambiguity or contradiction enters review.
4. Read the destination state back
Verify in CRM that the event and any new business case exist exactly once under the intended contact. After an update or merge, inspect master fields, source history, suppressions, notes, open opportunities, tasks and active automations. The run ends only after this readback passes.
Define the master record and field policy before merging
Matching links an event to a contact. Merging removes an already existing second contact record. That is a larger intervention. The master must be chosen before the operation; “oldest” or “most recently edited” is not a sufficient business rule on its own.
Data area | Robust default | Human decision required when |
|---|---|---|
primary email and phone | retain the confirmed current value and preserve alternatives separately | two confirmed values differ or an identifier is shared |
name and company | inspect origin and recency; do not shorten automatically | company changed, legal name differs or several roles exist |
consents and suppressions | protect the most restrictive evidenced state; preserve origin and time | purpose or evidence is unclear |
lead source | preserve first origin and each later event separately | teams use competing definitions of first touch or campaign |
contact owner | do not silently replace the established relationship owner | several active relationships or a deliberate handover exists |
open opportunities | do not combine merely because contacts merge | similar titles may represent different purchases |
notes and conversation history | keep origin and timestamps | sensitive statements conflict |
tasks and appointments | check duplicates and preserve owners and deadlines | parallel open work or calendar overlap exists |
active automations | inspect execution, waiting state and re-entry before the merge | a message, goal or wait remains tied to the disappearing contact |
A completed contact merge in ONE LOOP should not be treated as an easily reversible housekeeping step. For material cases, place an exportable decision record in the review card before merging: both contact IDs, chosen master, field decisions, open objects, operator and timestamp. It is not a full system backup, but it makes the business decision auditable.
GDPR Article 5 provides a narrow and useful frame: personal data should be purpose-limited, minimised and accurate. It does not dictate one technical matching rule, and this guide is not individual legal advice. It does support operational restraint: collect only identifiers required for the purpose, do not hide contradictions, and make erroneous links detectable so they can be corrected.
What the external figure shows—and what it does not
For its State of Sales, Seventh Edition, Salesforce anonymously surveyed 4,050 sales professionals across 22 countries from August to September 2025. All respondents were third-party panelists. Duplicate data ranks second among the reported data issues for teams with AI agents. Separately, 46% of sales professionals with agents say data-quality issues hurt their sales.
Sales professionals report an impact from data-quality issues
Share of sales professionals with AI agents · Surveyed August–September 2025 · report 2026
- Sales professionals with AI agents46 %
View data table
| Category | Share of sales professionals with AI agents |
|---|---|
| Sales professionals with AI agents | 46 % |
Duplicate data ranks second among data issues for teams with AI agents in the same report; 46% is still not a duplicate-record rate.
Source: Salesforce · State of Sales, Seventh Edition · Surveyed August–September 2025 · report 2026. The 46% refers to sales professionals with AI agents; the chart does not disclose the size of that subgroup. Full study: 4,050 sales professionals across 22 countries.
Global vendor survey and self-report; no Germany-only or SME-only subset for this finding. Not a duplicate rate, false-merge rate, causal effect or ONE LOOP performance figure. Scale 0–100%.
The 46% is not a duplicate-contact rate. It does not report how many records were duplicated or how many merges were wrong. The chart does not disclose the size of the subgroup described as sales professionals with agents. This is a global, self-reported vendor survey—not an observed causal effect, a Germany-only or SME-only subset, or a ONE LOOP performance figure.
The useful operating question is therefore not whether one study declares duplicates important. It is whether the actual process preserves unique events, links confirmed identities correctly, stops ambiguity and protects open work after a change.
Recover safely after a partial failure
A workflow may stop after candidate search, time out while updating a record, or lose its connection just after a merge. Blindly starting again can create another contact, a second business case or duplicate automation. Recovery begins by reading the current state.
Has this event ID already been processed?
Which contact is currently linked to the event?
Which rule version and decision state were stored?
Does a review card already exist, and does it have an owner?
Was a master selected or a merge already completed?
Are source, suppressions, open opportunities, tasks and appointments intact?
Which automations are active on the surviving or removed contact?
Which single operation is actually missing?
If the event is correctly linked, do not create another contact. If a review card exists, do not open a second one. If the merge has finished, do not repeat it; perform the destination readback. A technical failure must never convert a valid conflict stop into an automatic match.
Twelve cases to test before the pilot
The cases below are fictional tests, not measured ONE LOOP results.
Test case | Expected evidence |
|---|---|
the same form event is delivered twice | one event, one decision state and no second contact |
the same personal email arrives with different casing | canonical candidate found; raw values from both events remain |
the same phone arrives once nationally and once internationally | documented normalisation finds the candidate without deleting raw input |
one role inbox represents two contacts | possible match or person-specific new contact; no automatic merge |
two partners share a phone number | conflict stop when names or other confirmed identifiers disagree |
two employees at one company share a name but not email | two people or a review card; never a name-only merge |
an existing contact returns with a second project | one contact, two distinct business cases and events |
imported external ID collides with another source system | issuer scope prevents the false match |
marketing-suppressed contact sends a service enquiry | current request remains serviceable; suppression is not overwritten |
both contact records have open opportunities | master decision reviews each opportunity separately |
both contacts run in the same automation | expected re-entry and execution state documented before and verified after merge |
merge request is retried after a timeout | no second merge; readback on the surviving contact |
A test passes only when event, contact, business case, protected fields and active work all show the expected state. A green workflow run, a matching email or the disappearance of the second contact row is not enough.
Measure without false precision
Define numerator, denominator and review interval before the pilot. Keep absolute case counts visible, particularly with a small sample.
Metric | Numerator | Denominator |
|---|---|---|
candidate rate | events with at least one contact candidate | all unique business events processed |
audited auto-match precision | automatic matches confirmed by a person in the audit sample | all automatic matches inspected in that sample |
review confirmation rate | decided review cards that confirmed the same person | all review cards decided |
escaped duplicates | repeated person records confirmed in an audit sample | all contact groups inspected in that sample |
false-merge incidents | confirmed merges of different people | all completed merges, plus absolute incident count |
preserved-event rate | unique events still discoverable after match or merge | all events belonging to the audited contacts |
protected-field conflicts | conflicts stopped before automatic overwrite | all matches carrying at least one protected contradiction |
post-merge automation defects | merges followed by an unexpected stop, restart or duplicate action | all merges involving an active or waiting automation |
review-queue age | time since creation for each open review card | report the distribution and absolute open inventory |
Auto-match precision is only as credible as its audit sample. It cannot be calculated from the automatically accepted cases alone. A false merge is a safety incident and deserves individual investigation even when counts are low. A falling duplicate-record rate is not a success if the system has merely begun joining new people to existing contacts incorrectly.
Start with a bounded ONE LOOP pilot
Choose one intake route—such as a quote form—and a small rule set. First secure each event with its own ID. Then define which email and phone representations become canonical, which identifiers may be shared and which conflicts always create a review card. Enable automatic linking only for a narrow, contradiction-free combination.
ONE LOOP can use account-level contact deduplication preferences based on email or phone, present possible duplicate contacts for controlled review, and merge within a tightly scoped workflow using email, phone or both. Before launch, test source-specific behaviour, the master contact, field retention, opportunities, appointments, tasks, suppressions and active automations with the twelve cases. Never treat one identifier as universal proof of a person.
For new web enquiries, the website form to CRM guide covers reliable event intake. Moving social DMs into the pipeline explains how to keep the channel identity and business case separate. ONE LOOP can then review a bounded deduplication pilot with you.
Conclusion
A clean CRM does not come from merging as aggressively as possible. It comes from preserving each event, normalising identity keys transparently and giving ambiguity a visible state. Automate only a narrow, audited match. Decide the master and field-conflict policy before a merge, then read the contact, history, open work, suppressions and automations back from the destination. That prevents both duplicate customer records and the more dangerous error: combining two different people into one profile.
FAQ