Automating CRM Cleanup Without Deleting Live Deals by Mistake

"Stale" needs a precise definition before you can automate around it: no activity logged, no open tasks, and no stage change within a specific window, applied consistently by record type. A stale contact and a stale deal should use different windows, since a deal can legitimately sit quiet for weeks during a procurement process in a way a contact with no engagement usually can't.

Write the rule down as an actual query you could run today, not a description. If you can't write the query, you don't have a rule yet, you have an intention.

Exact-match deduplication on email or domain is safe. Fuzzy matching on company name is where most bad merges happen, since "Acme" and "Acme Industries" might be the same account or might be two genuinely different customers who happen to share a common word. Require a second matching signal, like a shared domain or a shared primary contact, before a fuzzy company-name match auto-merges anything.

Route anything that matches on a single weak signal to a manual review queue instead of an automatic merge. The volume there is usually small enough that a person reviewing it for ten minutes a week catches the true duplicates without risking a bad automatic merge.

Before any automated deletion or archival rule goes live, run it in report-only mode for at least one full cycle and read the actual list of records it would touch. It's common to find the rule would archive a genuinely engaged account whose activity happened to be logged in a field the rule didn't check, or that it would merge two records that share a domain because they're a parent company and a subsidiary.

Once the rule's report-only output looks right for two consecutive runs, turn on archival rather than deletion first. Archived records can be restored if the rule was wrong somewhere you didn't catch; deleted records generally can't.

Confirm your stale-record windows are defined per record type, not one blanket window for everything. Confirm deduplication requires at least two matching signals before an automatic merge, with single-signal matches going to manual review. Confirm the rule runs in report-only mode first and someone has actually read the output, not just glanced at a row count. Confirm archival, not deletion, is the default action, and confirm there's a way to restore an archived record within a defined window if it turns out to have been active after all.

Confirm each of these before you turn any automation on:

• Define stale with a written rule tied to your sales cycle length, and make the window longer than a typical deal cycle.
• Auto-merge duplicates only on exact matches such as identical email addresses, and log which record was kept as primary.
• Run every rule in report-only mode first, so edge cases like a quiet long-term account show up before anything changes.
• Exclude any record tied to an open deal or recent activity from archiving or merging.

A one-time cleanup degrades again within a few months without an ongoing process, since new duplicates get created the same way the old ones did: reps creating a new contact instead of searching for an existing one. Enforce deduplication rules at creation time, not just in periodic cleanup passes, so the same duplicates don't have to be caught twice.

Both Pipedrive and Close support duplicate warnings at record creation, which stops a large share of new duplicates before they ever get saved, rather than relying entirely on a scheduled cleanup job to catch them after the fact. Schedule a lighter monthly review of the report-only output going forward, even after the initial purge is done, so drift gets caught in weeks rather than accumulating for another year.