How to build pipeline hygiene automation

What matters here

• One staleness threshold across all deals is wrong for most of them. Scale it by stage and by deal size.
• Nudge the owner directly. A dashboard of dirty deals is a report about a problem, not an intervention.
• Escalate by value, not by count. Twenty stale small deals matter less than one stale deal that carries the quarter.
• Measure whether the field changed after the nudge. If it did not, the nudge is noise and should be redesigned or removed.
• Never auto-edit a commercial field. Closing a deal or moving a date on the rep's behalf destroys trust in the forecast.

Who this is for

You run revenue operations. The forecast review spends its first twenty minutes on data quality, and the weekly hygiene report has been emailed for a year without the underlying numbers changing.

How it works in practice

Everything that sits between a deal going stale and somebody doing something about it.

1. 1

   Staleness is defined per stage and per value band

   Fourteen days in discovery on a transactional deal, thirty in negotiation on an enterprise one. One number fits neither.

2. 2

   Detection runs on the fields that actually predict risk

   Next step missing, close date in the past, close date moved more than twice, no activity, amount unchanged since creation.

3. 3

   The owner is nudged where they work

   One message, their deals, the specific field, and a link straight to it. Not a digest of the whole team.

4. 4

   Escalation is weighted by value

   The manager hears about the deal carrying the quarter, not about twenty small ones with a blank next step.

5. 5

   Nothing commercial is edited automatically

   The app flags and asks. A close date moved by an integration is a forecast nobody trusts.

6. 6

   The change rate is measured

   Did the field get fixed after the nudge. That number decides whether the rule stays.

Paste these into Claude Code or Codex with the Tray Headless plugin installed. Each stage runs on its own. The systems named in them are the worked example rather than a requirement, and every prompt says so.

Once per project, run /tray-workflows:set-workspace to pick the workspace these build in. Point it at a sandbox first.

1. 1

   Set up, then find the fields that predict risk

   Hygiene rules built on the wrong fields generate noise forever.

   Headless skills build-workflow

   Copy

   Use build-workflow. The systems in play are Salesforce, or whatever we
   run in those seats.
   
   I need Opportunity with its stage model, close date, amount, next step,
   last activity date and the history object that records field changes.
   
   Specifically: how do I query how many times a close date has been
   changed. That is a better slip signal than any single stale date, and it
   lives in history rather than on the record.

2. 2

   Define staleness by stage and by value

   One blanket threshold is wrong for nearly every deal it touches.

   Headless skills build-workflow

   Copy

   Use build-workflow. Build a detection table I can edit, not rules in code:
   
     stage          value band       stale after
     Discovery      under 25k        10 days no activity
     Discovery      25k and above    21 days no activity
     Negotiation    under 25k        7 days no activity
     Negotiation    25k and above    14 days no activity
   
   Plus these regardless of band:
   
     Next step empty or older than the last activity
     Close date in the past and the stage is not closed
     Close date changed more than twice
     Amount unchanged since creation and past discovery
   
   Every flag carries which rule fired, so a rep can be told exactly what to
   fix instead of that their deal is dirty.

3. 3

   Nudge the owner, not the room

   The difference between an intervention and a report.

   Copy

   Send each rep one message, weekly, in Slack. Their deals only, ordered by
   value, with the specific problem and a direct link to each record.
   
   Keep it short. Five deals maximum in the message with a link to the rest.
   A message listing thirty things gets the same treatment as a dashboard.
   
   Never send a rep a message about a colleague deal, and never post a
   league table of who has the dirtiest pipeline. That produces gaming
   rather than accurate data, and inaccurate data is the thing you were
   trying to fix.

   The five deal cap matters more than it looks. A nudge that feels achievable gets actioned; an exhaustive list gets archived.

4. 4

   Escalate on value, and never edit a commercial field

   Two rules that decide whether the sales team trusts this.

   Headless skills tray-gotchas

   Copy

   Use tray-gotchas, then build escalation.
   
   Escalate to the manager only when: the deal is above a value threshold,
   it is in a late stage, and it has been flagged for two consecutive weeks.
   Rank by value at risk, not by count of problems.
   
   Two things the automation must never do:
   
     Never change a close date, a stage, an amount or a forecast category.
     It flags and it asks. A forecast that an integration has been editing
     is a forecast nobody defends in a board meeting.
   
     Never close a deal as lost on age. A stale deal is a question, not a
     verdict.
   
   The only field it may write is a hygiene flag and the timestamp of the
   last nudge.

5. 5

   Measure whether anything actually changed

   The step that separates this from the report it replaces.

   Copy

   For every nudge, record whether the specific flagged field changed within
   seven days.
   
   Then report:
   
     Change rate per rule. A rule below 20% is noise, and should be
     redesigned or deleted instead of kept out of politeness.
     Flags per rep and per team, trending.
     Value at risk from flagged deals, which is the number for the
     forecast review.
     Repeat offenders by rule, which usually indicates a process gap rather
     than an individual one.
   
   If the overall change rate does not improve within a quarter, this is a
   report with extra steps and should be turned off.

6. 6

   Prove it works, then hand the thresholds to RevOps

   Because thresholds change with the sales motion.

   Copy

   Run the per-step schema checks and the whole-workflow audit before this
   touches production.
   
   Then open the same workflow in Tray Build so revenue operations can tune
   the staleness table and the escalation thresholds in the visual canvas.
   Those change whenever the sales motion changes, which is more often than
   anybody expects.

What it connects to

Hygiene reads the pipeline and reaches the people who own it.

Named systems are the ones most teams run, not the only ones that work. Each is an authentication in your Tray workspace, referenced by name, so the workflow never holds a credential. Where we have a connector page, the name links to it.

Questions people ask

Further reading

Background on the same subject, for the case rather than the build.

• Automate your way to CRM data hygiene

Related guides