Data foundation

Lifecycle stage is the strongest signal AI uses for routing and scoring. Stale or manually overwritten values make recommendations reflect old process debt.

First step: Export contacts grouped by lifecycle stage, agree one definition per stage with sales and marketing, then bulk-fix blanks.

Agents hand off and escalate to the record owner. Blank or deactivated owners mean leads and tickets silently go nowhere.

First step: Build a list of contacts where owner is unknown or deactivated and reassign with a rotation rule.

ICP targeting and routing rely on industry. Free-text variants ("SaaS", "Software", "software co") split one segment into many.

First step: Convert Industry to a dropdown, map existing variants to the standard values, enrich blanks.

Duplicates split activity history, so AI summaries miss conversations and agents email the same person twice.

First step: Run the duplicate management tool, merge by email, and require email on all forms and imports.

Contacts tied to duplicate companies break the account view; AI cannot see the buying committee.

First step: Merge companies by primary domain and turn on automatic company association by domain.

When a field means different things to different teams, the model learns a shortcut, not a business rule.

First step: Inventory custom properties, archive unused/near-duplicate ones, and write a description for every property AI depends on.

Activities attached to unassociated contacts never appear at the account level, so AI account summaries are incomplete.

First step: Enable auto-association by email domain and backfill associations for existing contacts.

Deals missing contacts hide who is actually buying; AI deal summaries and next-step recommendations go generic.

First step: Require at least one associated contact at deal creation and review open deals with none.

It surfaces formatting issues, property anomalies and duplicates you would not see manually.

First step: Open Data Management > Data Quality, run the Data Agent and work its recommendations top-down.