Duplicates enter the CRM from every direction — form submissions, CSV imports, integration syncs, and manual data entry. The agent identifies duplicates using fuzzy name matching, email domain clustering, phone number normalization, and company name variants, catching merges that exact-match logic misses entirely. When two records resolve to the same entity, the agent surfaces a merge recommendation with a confidence score and a preview of which fields from each record would survive the merge. High-confidence duplicates are merged automatically; ambiguous cases land in a review queue with context. The result is a CRM where every person and company has exactly one canonical record.

Contact data decays at roughly 30% per year — people change jobs, companies are acquired, email addresses bounce, phone numbers are reassigned. The agent runs a continuous reverification cycle across all contacts in the CRM, checking email deliverability, validating phone numbers against carrier databases, cross-referencing LinkedIn profiles for current employer status, and flagging records where the contact's listed title no longer matches their current role. Records that fail verification are flagged with specific failure reasons — not silently left in the database to corrupt pipeline metrics and campaign deliverability rates.

A CRM is only as useful as the fields that are actually filled in. The agent identifies records missing critical data — no phone number, no company size, no industry vertical, no LinkedIn URL — and runs enrichment passes using a waterfall across 150+ providers to fill the gaps. Enrichment is prioritized by record importance: contacts attached to active deals and high-tier accounts get enriched first. Every enriched field is stamped with the source and date, so downstream users know what was added programmatically versus what came directly from the contact. The agent doesn't overwrite human-entered data — it fills what's missing and flags conflicts for review.

Inconsistent data entry undermines every downstream workflow — segmentation breaks, routing rules misfire, and reports become untrustworthy when the same value is entered fifteen different ways. The agent enforces field-level governance rules across the CRM: normalizing country codes, standardizing phone formats, correcting industry taxonomy mismatches, and flagging records where custom field values don't match the approved picklist. Rules are configurable by field and severity level — critical violations are auto-corrected, cosmetic inconsistencies are flagged in a weekly hygiene digest. Teams get a health score per object type so data quality is a visible, trackable metric rather than an invisible problem.