Data-hygiene robots: keeping the CRM honest automatically

CRM hygiene should be enforced by automations, not by quarterly clean-up afternoons. A small set of scheduled checks — stale deals, missing fields, duplicates, dead contacts, impossible stage jumps — runs daily, flags or fixes what it finds, and never gets bored. Manual hygiene decays in weeks; robotic hygiene compounds.

Why does manual clean-up always fail?

Because it fights a flow with a one-off effort. Records decay continuously — people change jobs, deals go quiet, duplicates creep in through form fills and imports — while clean-ups happen at best quarterly, usually when someone senior notices the reports look wrong. The result is a sawtooth: brief accuracy after each purge, steady rot in between, and reporting you can only trust for about a fortnight per quarter.

The stakes are higher than tidiness. Every number on the management dashboard — the live instrument panel set out in The MD Dashboard Blueprint — is computed from these records. A pipeline stuffed with dead deals overstates the future; duplicated contacts undercount conversion; missing close dates make the forecast unweighable. Dirty data does not merely look untidy, it steers decisions wrongly.

What should the robots actually check?

Five checks cover most of the decay in a 5–50-staff service firm:

1. Stale deals. Any deal idle beyond its stage's normal age — no activity logged, no scheduled next step — gets flagged to its owner, and after a further grace period moved to lost with a reason code. This single check usually removes the largest lie in the pipeline.
2. Missing critical fields. Deals without a value, a close date or an owner; contacts without a company. The robot lists them daily; nothing enters a report until it is complete.
3. Duplicates. Same email, same company domain, or fuzzy-matched names. Flag for merge rather than auto-merging — machines are good at finding duplicates and mediocre at choosing the survivor.
4. Decayed contacts. Bounced emails, leavers, dead domains. Mark, do not delete; the history still has value.
5. Impossible transitions. A deal that jumps from Enquiry to Verbal Agreement in an hour, or moves backwards twice, is either a data error or a process being bypassed. Either way, someone should see it.

These checks only work against stage definitions that mean something in the first place — verifiable events with time-outs, as set out in pipeline stages that mean something. Robots cannot enforce rules that were never written.

How do you build the hygiene system?

The mechanism is a daily loop, and each rule follows the same pattern:

1. Define the condition. When a deal has had no logged activity for 14 days and no future task, then it is stale. Precise, checkable, arguable about once — at design time, not in every meeting.
2. Choose the response tier. When the fix is unambiguous (a lowercase email, a missing country inferred from the phone code), then the robot corrects it silently. When judgement is needed (a merge, a close-as-lost), then it flags to a named human with a deadline.
3. Schedule it. Every rule runs daily, before the working day, so people arrive to a short exceptions list rather than discovering problems mid-meeting.
4. Report the exceptions, not the successes. One digest — five stale deals, two duplicates, three incomplete records — sent to the person who owns pipeline quality. An empty digest is the goal state.
5. Escalate on repetition. When the same record is flagged three times, then it goes to the MD's list. Chronic exceptions are process problems wearing a data costume.

When every record is checked every day, then no individual error survives long enough to distort a report — which is the entire point.

Does automated nagging annoy the team?

Less than the alternative. A robot that flags your stale deal privately at 8am is considerably kinder than a manager raising it in the Monday meeting. Done well, hygiene automation reduces the surveillance temperature: the rules are explicit, applied identically to everyone, and mostly resolved before anyone senior looks. Reps also gain something concrete — a clean record of who needs chasing. Most firms stop at two follow-up touches while deals typically need five or more, a gap examined in the follow-up cliff, and a clean CRM is the raw material for closing it. Flagged-stale deals are, in practice, a to-do list of recoverable revenue.

What does this cost to run?

Very little, which is rather the point. Most mid-market CRMs can express these rules natively through workflows; where they cannot, a lightweight automation layer alongside the CRM covers the gap. Build the five checks once, review the rules quarterly, and the system holds its own accuracy — as opposed to renting accuracy two weeks at a time from clean-up afternoons. Where sales reports still mislead after the data is clean, the causes are structural rather than hygienic, and that is the subject of where sales reports lie.

---

Next step: the Growth System Audit — £450, seven days, credited against any build — maps where your growth system leaks, including whether your CRM data can be trusted, and what to build first.