Section 20 · Sales Rep Work Elimination / Zero-Entry CRM

20.7Duplicate detection + smart merge

Built in the POC

What DMA asked for · Decision Book p.37

Find duplicate companies and people, and suggest the merge.

Why it matters to DMA: Duplicates split a customer's history and send two reps after one buyer.

In our words, from DMA's Product Decision Book v3, page 37.

A rule compares company names and web domains, and people's names, e-mails and phones. Each likely pair gets a score and what matched.
The book sets this at L1 (suggest). The pair is shown with its evidence; the merge itself is done by a person in Nutshell, which keeps Nutshell's own merge history.
A likely duplicate's score is capped until it is resolved, so the same buyer isn't worked twice.

It plays on its own while it's on screen; hover or use the controls to pause or step through.

Step 1 of 4

The records

Every company and person in the mirror.

The same six stages on every page. Nothing reaches Nutshell except through the write gate.

1Nutshell
Source
Deals, contacts, tasks, activities, notes and users.
2Sync
Copies Nutshell changes into the bridge database
Keeps the mirror current, so the rules see today's state.
3Bridge database
A copy of the CRM data, plus what the AI works out
One open flag per record and problem, with its details and the rule version.
4AI
Claude models, only through DMA's own gateway
Not needed for most flags: they are rules. The audit adds what only the notes say.
5Write gate
The only way back into Nutshell: checked, approved when needed, sent once
Flags don't write to Nutshell; fixes happen there, by people (or through the write gate where a field rule allows it).
6Nutshell
The system of record
Reps fix the record; the flag closes on the next run.

Drawn from the POC's database catalogue: structure only, no data.

PK primary keyFK reference the database enforces (solid line)ID reference kept by id (dashed line)

Table	What it holds	Columns
ai.duplicate_candidates	Pairs of records that look like the same company or person, with the score and what matched.	10
ai.hygiene_flags	One open flag per record and problem (stuck deal, missing next step, former owner, bad email, duplicate), with severity, details and the rule version.	10

Synthetic demo data: every name, business and number is made up.

Likely duplicates in the POC today (live counts, synthetic data).

Kind	Pairs	Top matches on
Companies	24	Name, web domain and a shared contact
People	7	Name, e-mail or phone

localhost:13301/accounts

1 / 2A likely duplicate: why its score is capped.

Dashboard · Accounts, a likely duplicate, why the score is capped. Captured from the running POC on September 25, 2026; synthetic data.

Opens the exact screen in the running POC, in a new tab.

Open	Where	What to try
Pipeline hygiene: possible duplicates	DMA AI dashboard	Every likely pair, with what matched.

Built in the POC

Built in the POC. This capability runs in the POC today, on synthetic data, end to end.