A Salesforce Pipeline Prompt That Audits Hygiene in One Pass

Most pipeline reviews die the same way. A rep opens the forecast, eyeballs forty open opportunities, nods at the ones that look healthy, and moves on. The deals that quietly rot (no logged next step, a close date that's slipped twice, a single champion who went dark) survive another week because nobody had the time to read every row. A salesforce pipeline prompt fixes that by reading every row for you and flagging only the ones that need a human.

This isn't about connecting Salesforce to an LLM. It's about one reusable prompt that takes a CRM export and returns a hygiene audit with a fixed shape every time. Paste the open opportunities, get back a flagged list. Same output contract on Monday as on Friday.

The pages that rank for pipeline prompts mostly miss this. ClickUp's pipeline-analysis page is a broken redirect with no copyable prompt at all (clickup.com). The listicle sites give you "key questions to ask in a pipeline review," or a guide to wiring up a Salesforce-to-ChatGPT connector, but not a single prompt with a locked output contract scoped to the four failure modes that actually break a forecast.

What this pipeline hygiene prompt does

A pipeline hygiene prompt is an instruction that scans a CRM export and flags every open deal that violates a hygiene rule, then tells you the one thing to fix. Here's what one pass covers.

• Stale opportunities sitting past the normal age for their stage
• Missing next steps: open deals with no logged action and no date
• Close-date slippage: deals whose close date has already moved at least once
• Single-threaded deals with exactly one contact and no second relationship
• Stage-amount mismatches: six-figure deals still parked in an early stage
• Forecast-category drift: "commit" deals that fail two or more hygiene rules
• Owner load: a quick count of how many flagged deals sit with each rep

That's the audit a sales manager does by hand before every forecast call. The prompt does it in one paste.

Anatomy of the prompt

The structure is plain: you set a few variables, paste the export, and the prompt returns a table plus a short rollup. Keep the output contract at the very end so it survives a long pasted CSV. Models weight the most recent tokens, so a contract buried above the data gets ignored on big inputs.

Variables
  {{crm_export}}        → the pasted CSV of open opportunities
  {{stage_age_limits}}  → max days per stage (e.g. Discovery:21, Proposal:14)
  {{current_date}}      → today, so age and slippage math is anchored

Prompt
  Role: a sales operations analyst running a weekly pipeline hygiene audit.
  Task: score every open opportunity against the hygiene rules below.
  Rules: stale stage, missing next step, slipped close date,
         single-threaded, stage/amount mismatch.

Output contract (the last lines of the prompt)
  | Opp | Owner | Amount | Stage | Flags | Top fix |
  Then a 4-line rollup: total at risk ($), worst stage,
  count of single-threaded deals, top 3 deals to work today.

The variable that earns its keep is {{stage_age_limits}}. Without it, "stale" is a guess. With it, the prompt has a deterministic rule per stage, and the same export run twice returns the same flags. That's the difference between a prompt that audits and a prompt that improvises.

Step-by-step usage

1. Pull the export

In Salesforce, build a report on open opportunities with these columns: Opportunity Name, Owner, Amount, Stage, Close Date, Last Activity Date, Next Step, and Contact count (or a roles-per-opp formula). Export to CSV. You don't need an integration for a weekly review; a paste-in beats a connector you have to babysit.

2. Set the variables

Fill {{stage_age_limits}} with your real stage durations, not generic ones. A team selling a 90-day cycle needs different thresholds than one closing in two weeks. Set {{current_date}} so close-date slippage is measured against today.

3. Run the prompt

Paste the CSV into {{crm_export}} and run. On Claude, the ## Output format heading at the end holds the table shape reliably. On GPT-4o, restate the column list on the final line or it sometimes drops a column on a wide table. Gemini handles the long CSV context well but tends to over-explain, so add "no commentary outside the table and rollup" to keep it tight.

4. Work the rollup, not the table

The table is the evidence. The four-line rollup is the action. "Top 3 deals to work today" is what you take into the forecast call. The rest is backup when a manager asks why a deal got flagged.

5. Re-run weekly

Save the filled prompt. Next week, swap in a fresh export and run it again. Because the output contract is locked, you can diff this week's flags against last week's and see which deals got cleaned up and which got worse.

Prompt-craft patterns that make it reliable

Deterministic rules over vibes. Don't ask the model to "assess deal health." Give it arithmetic it can check: flag stale if days_in_stage > stage_age_limit. A rule the model can verify against a field produces the same answer every run. A vibe doesn't.

Flag a deal STALE only if (current_date − stage_entry_date) > the limit
for its stage in {{stage_age_limits}}. Show the day count in the Flags cell.

One fix per flagged deal. A pipeline audit that returns five problems per deal is noise. Force a single top fix, the one action that most moves the deal, so the rep knows where to start.

For each flagged deal, return exactly one "Top fix": the single
next action that most moves it (e.g. "book a 2nd contact").

Refuse to invent fields. If the export lacks a column, the prompt should say so, not guess. That refusal boundary is what keeps the audit trustworthy. A hallucinated "next step" is worse than a blank one.

Worth being honest about the limits here. The prompt only knows what's in the export. If reps don't log activity, the "missing next step" flag will light up half the pipeline, which is itself useful signal. And a model can misread an ambiguous date format, so keep dates ISO (YYYY-MM-DD) in the export and the slippage math stays clean.

Variables you'll set

Getting started

1. Build the open-opportunities report in Salesforce and export to CSV.
2. Set {{stage_age_limits}} to your actual stage durations.
3. Paste the export, run the prompt, read the rollup first.
4. Take the "top 3 to work today" into your next forecast call.
5. Save the filled prompt and re-run it weekly against a fresh export.
6. For a maintained version with the rules and contract already wired, start with the Salesforce Pipeline Agent Pack.

If you also send the recap after the pipeline call, the Sales Meeting Follow-up Agent Pack turns those notes into the email and the next-step list in the same shape.

A clean pipeline isn't a personality trait, it's a checklist someone actually runs. Run it with a prompt and the checklist never gets skipped on a busy week. If your next step after the audit is qualifying the deals that survived, the MEDDPICC qualification prompt scores them against the eight elements that predict a close. And when a flagged deal needs a recap to re-engage a quiet buyer, the sales follow-up email prompt turns your notes into the message.