AI pricing is more varied and more volatile than classic software pricing. Knowing the unit you pay for is the start of spend control.
The main models
- Seats (per user per month): assistants inside productivity suites and finance tools. Predictable, but you pay for every seat whether it is used or not.
- Consumption (per token, request, action or conversation): model APIs and agents. Cost follows usage and can spike.
- Platform (annual fee or commitment): AI platforms and vertical finance tools, often with usage tiers or credits.
- Outcome-based (per resolved case, per invoice processed): emerging; aligns cost with value but needs clear definitions.
Real price points
- Microsoft priced Microsoft 365 Copilot at $30 per user per month for enterprise customers at general availability (November 1, 2023).
- Salesforce priced Agentforce at $2 per conversation, then in May 2025 introduced Flex Credits, where each action consumes 20 credits ($0.10 per action), sold in packs of 100,000 credits for $500.
Prices like these change, and contract prices differ from list prices, but they show the shift from paying per person to paying per unit of work.
What finance should ask for every AI line
- What is the pricing unit, and what drives volume?
- What are the rates, tiers and overage terms?
- Is there a spend cap or alert in the vendor's console?
- How much notice before price changes, and can we exit?
- Who internally owns the budget and the usage?
Budgeting
Seats: budget seats × price, then manage utilization. Consumption: budget a volume forecast × rate, with a buffer and a cap. Platforms: budget the commitment and track usage against it so you neither overbuy nor run out.