Claude Code for HubSpot is a way to run a full RevOps cleanup from a terminal: Claude audits your CRM, merges duplicates, learns your ICP from closed-won and closed-lost deals, enriches accounts, tiers every company, builds the HubSpot workflow, and writes a one-line account summary reps can act on. This page is the playbook we use, with an interactive walkthrough you can click through.
Try the interactive walkthroughHave us run it on your CRMThe play in one line: audit, clean, learn the ICP from your own deals, enrich, tier, then route. In that order, with a backup, small test batches and shadow properties at every step. AI does the clicking; operators own the rules and the review.
How to read it: on the left is Claude Code, the AI agent you type instructions into. On the right is HubSpot, where every change shows up. The card in the corner of HubSpot says exactly what Claude did through the HubSpot API on each step. We ran the play on a fictional copy of JayOh's own book: 25 company records, 3 duplicate groups, 13 closed and open deals. Every number on every screen is calculated from the same sample data, so the audit, the tiers and the readout always agree.
Use Next / Back or your arrow keys. Try the interactive bits: merge duplicates (step 5), flip before/after enrichment (step 7) and filter by tier (step 10). Illustrative mockup with fictional data.
Each stage has one job, one example prompt and one guardrail. Skip the order and the later stages break: you can't tier accounts whose industry is blank, and you can't route accounts that are duplicated.
Claude Code runs in a terminal and talks to HubSpot through its API. Before it reads a single record, give it the ICP written as testable rules (geography, employee range, business model, CRM, funding), the buyer and the pain. Then ask for a read-only audit of the fields that decide fit, tier and territory.
Every later count depends on a clean book, so duplicates go first. Tell Claude exactly which record survives: the one with activity; if neither has activity, the one with the higher property fill rate, then the more recently created one. Edge cases go to a human.
This is the strategic step. Claude compares the firmographics of every won deal against every lost deal and then checks where reps actually spend their time. In most books we audit, a large share of activity sits on accounts that look like the losses. That misallocation is the business case for everything that follows.
Build a small static segment, enrich it, check it by hand, then run the same script on everything. Classify industry into a fixed picklist so dashboards get ten clean values instead of a hundred creative ones. If the CRM or tech stack can't be confirmed, mark it "not confirmed" rather than guessing.
Claude writes the tier rules into a shared sheet and back-tests them against closed deals. Then one prompt creates the HubSpot workflow that tiers every company, re-tiers when inputs change and enrolls new companies. Check the branches in order: Disqualified first, so a big logo with a bad fit never lands in Tier 1.
Finish with the rep view: a one-line account summary on every company that starts with the tier and says who they are, plus owner assignment and a first-touch task for Tier 1 and Tier 2. Reps open the record and know who to call and why, without doing manual research.
The difference between a clean CRM and a mess you can't undo is process, not the model.
| Tested scripts (what we use) | Live agent changes | |
|---|---|---|
| Who decides | Rules you wrote and approved | The model, record by record |
| Repeatable | Same input, same output, every run | Can vary between runs |
| Reviewable | Dry run on a sample before full run | Changes land as they happen |
| Rollback | Backup + shadow properties | Depends on what was logged |
It's using Claude Code, Anthropic's coding agent that runs in a terminal, to work on a HubSpot portal through the HubSpot API. Instead of clicking through settings, you describe the RevOps job in plain English and Claude writes and runs the scripts: audits, merges, enrichment, workflows and account summaries.
It is if you follow five rules: back up first, test on small batches, use tested scripts instead of live AI decisions, write to shadow properties first, and run first jobs off-hours. A person still owns the rules and reviews every write.
No. It removes the clicking and the manual research. The judgment stays with operators: writing the ICP, choosing the keep-rule for duplicates, validating the win/loss pattern with sales, and reviewing every workflow before it goes live.
Tiers give reps a clear priority and route cleanly: Tier 1 gets an owner and a task this week, disqualified accounts get nothing. Points-based scores drift, get argued about and are harder to back-test against closed deals. We cover scoring separately in our lead scoring model guide.
Keep the record with activity. If neither has activity, keep the one with the higher property fill rate, then the more recently created one. Anything that doesn't fit the rule goes to manual review.
Yes, and it's where the play pays off most, because tech stack and CRM fit are usually the emptiest fields. Fix the HubSpot–Salesforce sync mappings first so enriched values don't get overwritten on the next sync.
JayOh is a HubSpot Platinum Partner and Claude partner with 75+ CRM migrations and 100+ documented GTM plays. In a free RevOps sprint, senior operators run the audit and the first fix on your portal: read-only until you trust us, and documented either way.
We engineer growth systems. JayOh builds the RevOps architecture, GTM execution and data infrastructure that turn messy revenue operations into predictable pipeline for B2B SaaS and PE-backed teams.
