Emergency Unmerge Case Studies: Real HubSpot Merge Recoveries, Real Numbers
Client details are anonymized. Every number below comes directly from the engagement's machine-verified completion records.
Two real engagements, all numbers machine-verified: an executive search firm where a bulk dedup run collapsed 610 contacts and 1,885 contacts were reconstructed across two waves, and a global technology consultancy where 31 of 31 wrongly merged contacts were recovered with zero failures. Recovery covers HubSpot contacts, companies, and deals.
Last updated: July 2, 2026
An executive search firm: 1,885 contacts reconstructed after a bulk dedup run collapsed 610 records
In June 2026, a bulk deduplication run at an executive search firm merged 610 contact records into surviving records. The match rules were too aggressive, so unrelated people were combined: separate individuals ended up sharing one contact record, with their email addresses, company links, and properties blended together. For a firm whose business is people data, that is the worst kind of corruption, because every blended record is a live risk of contacting the wrong person with the right details.
Recovery ran in two waves. When the engagement closed, the final old-to-new ID map contained 1,885 reconstructed contacts, each released cleanly from its survivor. The second wave alone covered 179 survivors the first pass had missed: 520 people were recreated with zero duplicates and zero collisions, 452 email addresses were moved and individually verified with none left behind on survivors, 514 company associations were restored with domain matching, and 346 properties were restored only where they could be attributed to exactly one person.
| Metric | Machine-verified result |
|---|---|
| Contact records collapsed by the June 2026 dedup run | 610 |
| Recovery waves | Two |
| Reconstructed contacts in the final old-to-new ID map | 1,885 |
| Survivors the first pass missed, handled in wave two | 179 |
| Wave two: people recreated | 520, with zero duplicates and zero collisions |
| Wave two: email addresses moved and individually verified | 452, none left behind on survivors |
| Wave two: company associations restored with domain matching | 514 |
| Wave two: properties restored (single-person attribution only) | 346 |
| Full-portal completeness check | 649 merged records accounted for, zero gap |
What made it hard?
The dedup run had mixed unrelated people, so every reconstruction had to decide which person each email address, company link, and property value belonged to. An adversarial review pass caught look-alike duplicate names that a naive rebuild would have re-collapsed. The same review discarded an enrichment step that was pulling job titles out of quoted email text, because a quoted signature block is evidence about the sender of the quoted message, not the record owner. The operating rule for the whole engagement was that a blank is better than a wrong value: the 346 restored properties were written only where they could be attributed to exactly one person, and everything ambiguous was left blank rather than guessed.
How was it verified?
Honesty beat first: the initial pass measured itself against the dedup job's own input list instead of a live-portal census. By that yardstick it looked complete, but the yardstick was wrong, and it hid 179 merged survivors. The fix was a full-portal completeness check that counted every merged record actually in the portal, not just the ones the job knew about.
That check closed with zero gap: 649 merged records accounted for in total, 431 from wave one, 179 from wave two, and 39 older legitimate merges that were deliberately left alone because they predated the incident and were correct. On top of the census, wave two's 452 moved email addresses were verified individually, and the 520 recreated people were checked for duplicates and collisions, with zero of either.
A global technology consultancy: 31 of 31 merged contacts recovered with zero failures
In March 2026, a global technology consultancy found that 31 contacts had been wrongly merged. The engagement ran as an expedited guided recovery sprint, and it closed at 31 of 31: every merged contact was recreated, with zero failures.
The rebuild set 1,265 properties, sourced from HubSpot's own merge-history export, which served as the source of truth for pre-merge values. It restored 216 associations and enriched 162 email activities with full metadata. Before anything was written to the portal, all 225 ambiguous association attributions were reviewed and approved by the client. And to keep the fix fixed, each survivor's stray email address was cleared so HubSpot would not automatically re-merge the records.
| Metric | Machine-verified result |
|---|---|
| Contacts wrongly merged (March 2026) | 31 |
| Contacts recreated | 31 of 31, zero failures |
| Properties set from HubSpot's merge-history export | 1,265 |
| Associations restored | 216 |
| Email activities enriched with full metadata | 162 |
| Ambiguous association attributions reviewed and approved by the client before writing | 225 |
| Survivor stray email addresses cleared to prevent automatic re-merge | All survivors |
What made it hard?
The sprint was expedited, so there was no room for a rebuild-and-see-what-breaks approach. The dangerous part was attribution: 225 association attributions were ambiguous, meaning the evidence alone could not prove which reconstructed contact each one belonged to. Instead of guessing, every one of the 225 went to the client for review and approval before a single write happened. The other trap was HubSpot itself: after a merge, the survivor keeps the lost record's email address as a stray secondary email, and leaving it in place invites HubSpot to re-merge the freshly separated records. Each survivor's stray email address was cleared as part of the recovery.
How was it verified?
Property values were not reconstructed by inference where a better source existed: HubSpot's own merge-history export was used as the source of truth, and all 1,265 restored properties trace back to it. The client approval gate covered every one of the 225 ambiguous association attributions before writing. The completion record shows 31 of 31 contacts recreated, 216 associations restored, and 162 email activities enriched with full metadata, with zero failures across the run.
Could your portal be recovered?
Probably, if the damage came from merges. Start with the Free Bulk Scan: upload a CSV or pick a HubSpot list of up to 10,000 contacts or companies, and it flags every merged record, free and read-only. For engagements over roughly 100 records, contact us to scope a guided recovery.
The Free Bulk Scan handles up to 10,000 records per job, works on contacts or companies, costs nothing, and never modifies your CRM. It tells you exactly which records were merged and how complex each recovery would be, so you know the real scope before deciding anything. Both engagements above started exactly this way: with a census of the damage, not a sales call.
If your scan turns up more than roughly 100 affected records, the recovery should run as a guided engagement like the two above, where rebuild order, attribution review, and completeness checks are handled for you. Reach out via the support page for a free consultation.
Related reading: how to audit merges in HubSpot and what to do when a duplicate cleanup goes wrong.
Case study questions, answered
Can hundreds of merged HubSpot records be recovered?
Yes. In the executive search engagement above, a bulk dedup run collapsed 610 contact records, and the final old-to-new ID map contained 1,885 reconstructed contacts, each released cleanly from its survivor. A full-portal completeness check closed with zero gap, with all 649 merged records in the portal accounted for.
Is bulk merge recovery all-or-nothing?
No. Recovery runs in waves and only writes what the evidence supports. In the executive search engagement, 346 properties were restored only where they could be attributed to exactly one person, 39 older legitimate merges were deliberately left alone, and the operating rule was that a blank is better than a wrong value.
How are these recovery numbers verified?
Every figure comes from the engagement's machine-verified completion records, not marketing estimates. Study one closed with a full-portal completeness check that accounted for all 649 merged records with zero gap, and in study two all 225 ambiguous association attributions were reviewed and approved by the client before anything was written.
Will recovered records duplicate or automatically re-merge?
Neither happened in these engagements. The executive search firm's second wave recreated 520 people with zero duplicates and zero collisions, and 452 email addresses were individually verified after moving. In the consultancy engagement, each survivor's stray email address was cleared so HubSpot would not automatically re-merge the recovered records.