Sales & CRM

How to clean up stale deals in your pipeline with AI

Sales & CRM3 AI tools6 steps6 friction points

Want an agent to handle this on your live data? See the Starch version →

A stale deal is any opportunity that's stopped moving — no recent activity, no response from the prospect, no next step on the calendar. Most pipelines accumulate them quietly. A rep adds a deal, things go quiet, and three months later you're looking at a CRM full of entries that are technically 'open' but practically dead. Cleaning them up means deciding which ones to revive, which to close out, and which to escalate before they expire entirely.

The reason this workflow feels like AI territory is that the signal is mostly in text — email threads, call notes, deal descriptions, last-contact timestamps. There's no formula for 'dead vs. dormant.' You need something that can read context, spot patterns like 'no reply in 45 days after two follow-ups,' and draft a re-engagement message that doesn't sound like a form letter. That's exactly the kind of pattern-matching and drafting work that LLMs are genuinely good at.

ChatGPT, Claude, and Gemini can all help here in meaningful ways. Paste in a deal list with last-activity dates and they'll flag the ones worth prioritizing. Feed them an email thread and they'll summarize the stall and suggest a next move. Give them a deal's history and they'll draft a follow-up. The real value is speed — what would take an hour of staring at a spreadsheet takes five minutes with a good prompt.

Skip the manual work → See the Starch version

Sales & CRM3 AI tools6 steps6 friction points

AI walkthrough

How to do it with AI today

A practical walkthrough using ChatGPT, Claude, and other off-the-shelf LLMs — what they're good at, what you'll have to do by hand.

Tools that work for this

ChatGPTClaudeGemini

Step-by-step

1 Export your pipeline from your CRM — HubSpot, Salesforce, Pipedrive, a spreadsheet, whatever you use — as a CSV or paste the relevant columns (deal name, stage, last activity date, deal value, owner) directly into Claude or ChatGPT.

2 Ask the LLM to identify deals with no activity in the past 30, 60, and 90 days, sorted by deal value. Tell it to flag anything over a dollar threshold as high-priority for manual review.

3 For each flagged deal, paste in the most recent email thread or notes. Ask the LLM to summarize what stalled the deal in one sentence and suggest whether to revive, archive, or escalate.

4 Use the LLM to draft re-engagement emails for the deals worth reviving. Give it context about the deal stage and the last communication so the draft doesn't sound generic.

5 Ask the LLM to generate a cleanup decision log — a table with deal name, recommended action, reason, and a suggested follow-up message — that you can paste back into your tracker or share with your team.

6 Manually execute the outputs: send the emails, update deal stages in your CRM, archive the dead ones. The LLM gave you the thinking; the clicks are still yours.

Prompts you can copy

Here is my pipeline as a CSV. Identify all deals with no activity in the last 45 days, sort by deal value descending, and flag any over $10,000 as high priority.

Here is the email thread for deal [X]. Summarize why this deal stalled in one sentence and recommend: revive, archive, or escalate. Explain your reasoning briefly.

Draft a re-engagement email for a prospect who went quiet after a demo two months ago. Tone: direct, not desperate. Keep it under 100 words. Context: [paste deal notes].

Given this list of 20 stale deals with last-activity dates and stages, produce a decision table with columns: Deal Name, Days Since Last Activity, Recommended Action, Reason, Draft Subject Line.

I want to set a rule for my team: any deal with no activity for 60 days gets auto-flagged for review. Write a short internal process doc explaining the criteria, review steps, and what counts as 'activity.'

Reality check

Where this gets hard

The walkthrough above works — until your numbers change, the LLM hallucinates, or you have to re-paste everything next month.

No live connection to your CRM — every session starts with a manual export, and by the time you paste it in, the data is already slightly out of date.

Context limits bite fast on real pipelines — if you have 150+ deals with notes, the LLM truncates or you have to split the work across multiple sessions and reconcile manually.

Outputs don't persist anywhere. The decision table you prompted last Tuesday lives in a chat window; there's no record in your CRM, no audit trail, no way to track what actions you actually took.

Draft emails generated by the LLM still require you to copy them one by one into your email client and send them manually — there's no connection between the LLM's output and your Gmail or Outlook.

Nothing reruns automatically. Your pipeline gets stale again next week, and you're back to exporting, pasting, and re-prompting from scratch — the same hour of work, every cycle.

Prompt consistency degrades over time. The categorization logic you carefully engineered in one session won't produce identical results next month unless you save and re-paste the full prompt every time.

Tired of the friction?

Starch runs the whole workflow on live data — no copy-paste, no hallucinated numbers, no re-prompting next month.

See the Starch version →

Starch alternative

The same workflow on Starch

Starch is an agentic operating system — an agent builds the persistent app that runs this workflow continuously against your live pipeline data, so stale deal cleanup isn't a manual session you schedule, it's a surface that stays current.

Connect HubSpot or Gmail once — Starch syncs your deals, contacts, and email threads on a schedule, so when you ask 'who haven't I spoken to in 45 days,' the answer reflects today's data, not last Tuesday's export.

The Sales Agent CRM starter app gives you a working pipeline view out of the box. Customize it in plain English: 'add a column for days since last activity, highlight anything over 30 days in red, sort by deal value.'

Starch builds re-engagement automations, not just drafts. Describe what you want: 'every Monday, find deals with no activity in 60 days over $5,000, draft a follow-up email for each, and queue them in my Gmail for review.' The agent builds that automation and runs it on schedule.

The CRM app lets you ask natural-language questions directly — 'which deals in the proposal stage have gone quiet this month?' — and get a filtered answer from your live data, not a one-off prompt you have to remember to re-run.

Actions taken in Starch update the record. When you archive a deal or log a follow-up, it persists in the app — so next week's cleanup review starts where last week left off, not from a blank chat window.

Start from the CRM or Sales Agent CRM template and fork it to match your exact pipeline. Or describe your workflow from scratch: 'build me a deal tracker that flags anything stale, drafts revival emails, and logs every action I take.' The agent builds it.

Get closed-beta access →

Toolkit