A flat per-seat fee is a bet that your whole company will use AI uniformly. No company does.
Per-seat does buy something real: a predictable line item. Procurement can budget it, finance can amortize it, nobody gets a surprise invoice. The honest framing is that you're paying a premium for that flatness, and the premium is the unused capacity of every light-usage seat. ## What you're paying for (and it's real) The Enterprise tier is not padding. Per OpenAI's published feature list: SCIM provisioning, Enterprise Key Management, role-based access controls, a compliance API, IP allowlisting, data residency across ten regions, SOC 2 Type 2 plus the ISO 27001 family, and 24/7 priority support with SLAs. OpenAI states it doesn't train on business data by default. If your requirement is a managed assistant with mature enterprise controls, and your security posture allows prompts and chat history to live in a vendor's cloud, it's a strong product. That second clause is the real decision point, and it has nothing to do with price: for a regulated organization, where the data lives ends the conversation before the per-seat rate starts it. ## What to ask in the sales call Since the price is negotiated, the call is where the real terms surface. Five questions worth arriving with: 1. **What's the renewal uplift?** Year-one pricing often reflects a land-and-expand discount. Ask for a written cap on the year-two increase; benchmarks are useless if the renewal resets them. 2. **What does the seat commitment actually commit us to?** A lower per-seat rate priced against committed growth means you owe those seats whether adoption shows up or not. Model the contract at your realistic adoption curve, not the optimistic one. 3. **What do credits cost when users hit rate limits?** The flexible-pricing overage is part of your effective rate. Get the credit pricing in the quote, not discovered in month three. 4. **Is the API in scope?** It isn't — so ask how developer access is expected to work and whose budget it lands on. 5. **Which of the compliance features are in this quote?** Data residency region, Enterprise Key Management, the compliance API: confirm what the quoted tier includes versus what's an add-on conversation. None of this is adversarial. It's the same diligence you'd apply to any six-figure annual commitment; the difference with AI seats is that usage data to sanity-check the commitment usually doesn't exist yet inside the buying organization. ## When per-seat stops making sense The alternative model is consumption: models are called through your own provider accounts and billed by the tokens used. A seat that sends ten prompts a month costs what ten prompts cost. The trade is that consumption needs governance to be safe — someone has to answer "who spent this" and stop a runaway script before the invoice does. That governance layer is exactly what SecondStack ships: [no per-seat licensing](/#pricing), LLM usage billed by upstream providers to accounts you control, and per-user, per-team, and per-key budgets with hard cutoffs so the spend curve can't get away from you. The same virtual keys govern chat, IDEs, and agents through one [LLM gateway](/blog/what-is-an-llm-gateway/) — and the whole stack is self-hosted, so the data-residency question disappears rather than getting negotiated. In our view, per-seat pricing fits tools everyone uses in roughly the same amount. Enterprise AI isn't one of them. Price the tail before you sign for it. If you're holding a per-seat quote right now, run both numbers: the quote, and your actual expected usage priced as consumption. We'll help with the second one — write to [hello@secondstack.ai](mailto:hello@secondstack.ai).