Data hygiene fails when it is treated as a project. The version that works is five specific checks on a fixed cadence, each with a named owner and a report, so drift is caught in days rather than discovered in a quarterly meeting. What follows is the cadence, not a philosophy.
| Check | Cadence | What it catches |
|---|---|---|
| Duplicates | Weekly | Records re-created by feeds and forms, before anyone builds on them |
| Required-field completeness | Weekly | Records created around the form, missing the fields reports depend on |
| Ownership and assignment | Monthly | Records assigned to departed staff, which vanish from every view |
| Cross-system reconciliation | Weekly | The CRM disagreeing with billing, the custodian or the DMS |
| Stale and dead records | Quarterly | Contacts who left, matters long closed, households no longer served |
Define clean before you measure it
"Clean" is not a general property. It is a per-field statement: this field is required on this record type, these are its allowed values, and this system is authoritative for it when two disagree. Write that down for the twenty fields your reports actually use. Everything else can wait, and pretending otherwise is why cleanup projects stall.
The reconciliation check nobody runs
The other four checks look inside the CRM. This one compares it to the outside world: does the CRM's fee schedule match billing, do positions match the custodian, does the matter status match the docket, does the document count match the DMS.
It is the check that decides whether people trust the system, and it is skipped because it is laborious. It is also the most automatable, because the comparison rules are explicit and the answer is a list rather than a judgement.
Why annual cleanups do not hold
A cleanup fixes the stock and ignores the flow. Every day, feeds create records, staff type into free-text fields and forms accept whatever the client wrote. If nothing catches that within the week, the org returns to its previous state on a predictable schedule, and the firm concludes the CRM was a bad purchase.
The compliance angle
For registered advisers, the client record supports obligations under the Advisers Act books-and-records rule (17 CFR 275.204-2). For law firms, matter data touches retention rules and duties over the client file. That raises the stakes on two of the five checks in particular: completeness, and anything that discards data during a merge. Keep a record of what each pass changed.
Where Caddi fits
Caddi is the owner for the mechanical part. It runs each check on its cadence, applies your definitions rather than generic rules, fixes what is unambiguous, and hands back a list of what needs a human decision. The reconciliation check in particular is the kind of work that only happens if something runs it without being asked.
Related: Duplicate management, CRM data entry automation, and RIA operations challenges.
The recurring half
See the hygiene cadence run itself
Caddi runs the checks, fixes the unambiguous cases, and reports the ones that need a decision.
Frequently asked questions
What is CRM data hygiene?
The recurring process of keeping CRM records accurate, complete, deduplicated, correctly owned and consistent with the other systems of record. It is distinct from a data cleanup, which fixes the current state and does nothing about the rate of new errors.
How often should you clean CRM data?
Weekly for duplicates, completeness and cross-system reconciliation; monthly for ownership; quarterly for stale records. The cadence matters more than the depth, because errors compound as soon as reports and workflows start depending on them.
What is the most overlooked CRM data quality check?
Reconciliation against the other systems of record: billing, the custodian, the docket, the document management system. It is the check that determines whether staff trust the CRM, and it is usually skipped because doing it by hand is slow.
Who should own CRM data quality?
A named person, with the checks documented and reported. In practice most firms leave it to whoever notices, which is the same as leaving it undone. The mechanical passes can be automated; the definition of what clean means cannot.