Claude Code cleans Freckle HubSpot duplicates
Toizer reports more than 2,500 merged HubSpot contacts and companies after Claude Code built a deduplication workflow, corrected false positives and added staged merge limits.
Original prompt
Shared by the source author. See source notes for context.
Set this up for my CRM and walk me through anything I need to configure.
Reach people exploring AI agents
See placements →Evidence at a glance
Sourced account- Source author
- Andy Toizer
- Original account
- Read the source ↗
- Published
- Source link checked
- Reported result
- Toizer reports more than 2,500 merged HubSpot contacts and companies after Claude Code built a deduplication workflow, corrected false positives and added staged merge limits.
- Human involvement
- Toizer approves the plan, inspects dry-run pairs, stops an unbounded fetch and verifies staged live merges. OpenClaw supplies a separate review, not a correctness guarantee.
A linked account supports attribution. It does not mean we independently reproduced its result.
Report an issue with this case
Open correction emailOpens your email app with the case link included. Nothing is sent automatically.
How the agent did it
- Define duplicate criteria and the fields that should survive a merge.
- Run read-only candidate checks on ten and fifty contacts.
- Correct placeholder-name false positives and investigate unclear company domains.
Andy Toizer builds a HubSpot deduplication workflow with Claude Code to clean Freckle’s live contact and company records. The initial system scores surviving records by completeness and engagement, queues uncertain matches and prepares a daily Slack digest.
Dry runs on ten and then fifty contacts expose false positives caused by placeholder company names. Another problem is excessive ambiguity: 944 records remain for review. Toizer asks the system to investigate company domains through redirects, canonical addresses and web evidence instead of relying only on similar strings.
A missing batch limit then starts fetching the full contact database when he expects a smaller run. He stops it before merges occur and adds a maximum-merge option. Subsequent execution proceeds ten merges at a time, with verification in HubSpot before expansion. An OpenClaw assistant reviews the session as an additional check.
The source reports more than 2,500 contact and company merges and a continuing four-hour scan. These figures are the operator’s account, not a guarantee that every merge was correct. The useful detail is examining candidate pairs before mutation, explicitly deciding which record’s information survives and bounding live batches. The recurring software contains an AI reasoning layer for unclear cases; Claude Code is also the agent that constructs and inspects that operational system.
Prompt
Original prompt
Source: Blog · Andy Toizer
Frequently asked questions
What did the dry run reveal?
Placeholder company names could produce false matches, so the matching logic needed correction before merging.
Why add a maximum-merge option?
A command unexpectedly began fetching the full live contact database; the operator stopped it before merges and moved to bounded batches.
How were unclear domains investigated?
The workflow followed redirects and checked canonical URLs and web evidence rather than only comparing domain strings.
More details
Remaining steps
- Add a maximum-merge limit before live execution.
- Merge ten at a time, verify in HubSpot and expand only after inspection.
About this account
Claude Code builds and runs the deduplication workflow, adds domain investigation for ambiguous cases and prepares recurring scans with a human-review digest.
Firsthand account by Andy Toizer; original publication date verified against Primary Substack byline Apr07,2026. Outcomes are the author’s reports, not independently reproduced. Original shared setup instruction for the published deduplication repository; the full original build prompt is described but not reproduced.
Credit: Andy Toizer
Source published
Added to Agents In Use
Source checked
Tools and topics
Claude Code · HubSpot MCP · Slack MCP · Web enrichment · OpenClaw review
CRM surviving-record rules · Bounded live merges · Company-domain investigation
Editorial completeness: 100/100. Measures documented fields, not agent performance or verified outcomes.