“Enterprise” on an AI pricing page is a signal, not a specification. On one product it means contractual no-training terms, single sign-on, admin controls and an audit log; on another it means a higher usage limit and a sales call. Before a business moves its staff from personal accounts onto a governed tier, it should know which of the five things the word ought to cover are actually in the agreement.
1. Data terms
The commitment that prompts, files and outputs are not used to train the provider’s models, plus retention, residency and subprocessor terms. This is the one that makes governed access defensible, and it is covered in detail in no-training licence terms: what to check. If a tier does not carry it contractually, it is not enterprise whatever the page says.
2. Identity and access
Single sign-on through your identity provider, so joiners and leavers are handled where every other system handles them; group-based provisioning; and the ability to restrict who may use which features and integrations. Without this the sanctioned-tool list cannot be enforced, only requested.
3. Admin controls
A console where the business can set policies, manage connectors and integrations, see usage by user and group, and export or delete data. The usage reporting is what makes a phased rollout measurable; the policy controls are what make the data classification real.
4. Audit and compliance
An audit log of who did what, retention the business controls, and the certifications and assessments your own compliance function will ask for. For regulated sectors, and for anyone selling into the EU, this is where the EU AI Act’s documentation duties are met.
5. Support and commitments
A named route to the provider when something breaks, uptime commitments, and a roadmap conversation rather than a changelog. Less glamorous than the others and the one most often missing at the lower “business” tiers.
Reading a tier
| Question | Why it matters |
|---|---|
| Is the no-training commitment in the contract for this tier? | The whole governance case rests on it |
| Does it integrate with our identity provider? | Enforceable access, clean leavers |
| Can we see usage by group? | Phased rollout, measurement, licence value |
| Can we control retention and integrations? | Data classification made real |
| Is there an audit log we own? | Compliance, incident investigation |
| What is the support route? | What happens when it breaks |
Three of the major providers now offer tiers that answer all six well; the differences between them are real but secondary to whether you are on such a tier at all. The expensive mistake is not choosing the wrong product; it is leaving staff on personal accounts because the enterprise conversation never happened.
We are platform-agnostic and take no margin on licences, so the reading is clean. The selection sprint that compares tiers on your own tasks and terms is described under AI tools and platform rollout.