Your HubSpot already knows where the money is. Which big deals have gone quiet. Which kinds of customer you keep losing, and why. The trouble is that getting those answers means building reports, and nobody has time. Claude can now connect straight to HubSpot, ask the question in plain English, and get back an answer calculated from your real data. I’ll skip the setup tour. Here are three plays.
Play 1: Sweep for big deals that have gone quiet
The money: high-value deals that nobody has touched in weeks. They’re usually not dead; they’re just nobody’s priority.
How I’d run it:
- Connect Claude to HubSpot (admin setup, once).
- Ask: “List open deals over $20k with no logged activity in the last 21 days. Show the owner, the stage, the last thing that happened, and the main contact.”
- Because the connector runs the question as a real query, you get a calculated list, not a guess. Check two or three against HubSpot to make sure.
- Ask Claude to draft a specific follow-up for each, based on the last activity. The owner edits and sends.
The maths, as an example: eight quiet deals at $30k is $240k. Wake up two, win one, and that’s $30k back from a question that took ten seconds to type.
Play 2: Find out why you really lose
The money: the same kind of deal being lost again and again, for a reason nobody has put in front of the team.
How I’d run it:
- Ask: “For deals closed-lost in the last 12 months, break down the lost reason by industry, deal size band and owner. Which combination loses most often?”
- If the lost-reason field is mostly empty, that’s your first finding. Fix the field and come back in three months.
- Pick the single worst pattern, for example “we lose food manufacturers on price above $50k,” and change one thing: a cheaper option, a different pitch, or stop chasing that segment.
The maths, as an example: if you stop spending quoting time on a segment you lose 9 times out of 10, those hours go into deals you actually win.
Play 3: Clean the forecast before the forecast meeting
The money: decisions made on a pipeline number that includes deals that died months ago.
How I’d run it: ask Claude to list every open deal with a close date already in the past, then draft new close dates or a “closed-lost” recommendation for each, based on the activity history. Because the connector can write back to HubSpot, it can apply the changes, but only after the owner has approved the list. Approve in batches, never “just do it.”
How I’d actually do it
Start with Play 1. It’s the quickest money and it shows you whether Claude’s answers match what’s really in HubSpot. Play 2 is the most valuable long term. Play 3 is the one to be careful with.
The caveat is write access. This connector can change your CRM, not just read it. Decide who’s allowed to use that, keep a human approving every batch of changes, and start read-only until you trust the answers. And if you don’t have a paid Claude plan, the first two plays still work with a HubSpot export and a spreadsheet. Claude makes them faster, not possible.
Where this fits
This is the HubSpot side of the same idea as Claude inside Salesforce. For prospecting rather than pipeline, see HubSpot’s Prospecting Agent. And if Play 2 shows that half your deals have no owner or lost reason, that’s the work of CRM Pipeline Rescue.
What shipped, for reference: HubSpot’s connector for Claude is the first CRM connector in Anthropic’s Connectors Directory. It’s built on MCP, reads and writes HubSpot records, and since mid-2026 runs analytical questions as computed queries. Available to all HubSpot tiers with a paid Claude plan (Pro, Max, Team or Enterprise). Anthropic says it doesn’t train on connector data unless customers opt in.
Source: HubSpot.
