Six things, none of which are the model.
Data quality
An agent grounded on duplicate accounts and half-populated fields will answer confidently and wrongly. Every time.
Dataverse architecture
Relationships, ownership and table design decide what an agent can actually reason across — and what it silently can't see.
Security model
Copilot respects the security model you have, not the one you meant to build. Weak record-level security becomes a disclosure problem the moment it is queryable in natural language.
Governance
Environment strategy and DLP policy decide whether citizen-built agents become an asset or the next thing to unpick.
Integration
If the answer lives in a system Dataverse can't reach, no amount of prompting will find it.
Process design
Automating a process nobody has fixed just produces the wrong outcome faster.
Find out whether your platform is ready before you buy licences.
AI and Copilot readiness is assessed as part of the Architecture Health Check — data quality, Dataverse design, security model, governance and the grounding sources an agent would actually read from. You get a straight answer on what needs fixing first, and what is genuinely ready now.