Sales · CRM hygiene
CRM cleanup as a standing job instead of a quarterly panic
Nobody sets out to let the CRM rot. It rots because cleaning it is a four-hour job with no owner and no deadline.
What the worker does
CRM hygiene in Polaris is a recurring task assigned to an AI worker with the HubSpot connection. The worker audits records against the rules in its skill file, then delivers a correction list as a comment: missing owners, stages that contradict activity, close dates in the past, duplicates. A person approves each change, because a field on a deal is a forecast input.
- Connection
- HubSpot
- Cadence
- Weekly recurring task
- Approval
- Every correction reviewed by a person
Dirty data is a forecast problem before it is a data problem
A deal with no owner does not get worked. A close date three weeks in the past keeps a number in the quarter that will not land. Two records for the same company mean two reps calling the same buyer in the same week.
None of this is dramatic on any single record, which is why it accumulates. It surfaces at the end of a quarter, as a surprise, in a meeting where nobody can explain it.
What the audit looks for
| Check | Why it matters | What the worker delivers |
|---|---|---|
| No owner | Unowned deals are unworked deals | The record, the amount, a proposed owner |
| Stage contradicts activity | A negotiation with no meeting is not a negotiation | The stage, the last activity date, the mismatch |
| Close date in the past | Keeps dead pipeline in the number | The record and how many days overdue |
| Probable duplicate | Two reps, one buyer | Both records and what matches |
| Empty required fields | Reporting silently degrades | The field, the record, and what it should probably be |
The weekly cycle
- 1
The rules live in the skill file
What each stage requires, which fields are mandatory, what counts as a duplicate. Written once, edited as a document when the rules change.
- 2
The worker audits
A cloud machine wakes for the task, reads the records through the connection, and works down the checklist, ticking each acceptance criterion.
- 3
The correction list arrives
One comment, grouped by check, with the record and the proposed fix side by side rather than a count of problems.
- 4
A person approves
The rep or the operations owner applies what is right and rejects what is not. Approving a list takes minutes; finding the list took the hours.
- 5
The trend gets visible
Same task every week, same structure, so a check that keeps producing the same failures tells you the process is broken, not the data.
Questions people ask
+Can it fix records automatically if we want that?
The Polaris delivery model is that work arrives as a comment for a person to close. Corrections are proposals, and a human applies them. That is a deliberate constraint, not a missing feature.
+How long does the first audit take?
The first pass is the largest, because every historical problem surfaces at once. Sessions are clamped between five minutes and eight hours, and the work log shows what each session covered and what it cost.
+Does this need our CRM to be tidy already?
No. A messy CRM produces a longer correction list, which is the useful output. What it does need is written rules, because an audit without a definition of correct produces opinions.
+Who should own the approval?
Whoever answers for the forecast. In small teams that is the founder or the sales lead; in larger ones, sales operations. Polaris assigns the approval as a task with a date like any other.
Related
Sales work, with the admin handed to an AI teammate
Selling is a conversation. Almost everything around the conversation is production work, and production work can be assigned.
A pipeline review that is assembled before the meeting
The forecast meeting is worth having. Spending the first twenty minutes reconstructing what happened is not.
The null rate tripled in March and the board slide was already wrong
Row counts, null rates, orphans and freshness, checked every morning. Silence means clean.
Hire an AI ops coordinator
It writes down how your company does things, so the answer stops living in whichever colleague happens to be on holiday.
Connect HubSpot to Polaris
Private-app tokens are scoped where you create them, which makes HubSpot one of the easier connections to grant narrowly.
Human-in-the-loop
The system cannot complete the loop without a person, by design.
Work log
The record that makes an unwatched run reviewable afterwards.