Learning Objective: By completing this Nano Learning course, participants will be able to:
- Identify vendor agreements with consumption or token-based pricing that fall below the dollar approval threshold, and route contracts to Finance and IT based on pricing structure rather than contract value.
- Evaluate why usage monitoring and model education control AI spend more effectively than hard per-user caps, which relocate consumption to unapproved tools rather than reducing it.
Category: Technical > Finance
Level: Basic
Credits: 0.2
Prerequisites: None
Advance Prep: None
Delivery Method: Nano Learning
Lesson summary
1Password’s CFO wrote publicly about an early AI bill that came in roughly five times higher than expected. This lesson uses that case, plus real client experience, to explain why AI costs are so hard to budget and what finance teams can do about it.
It starts with how AI pricing changes as adoption grows. Teams often begin on individual licenses at about $20 to $200 a month, move to per-seat plans, then shift to enterprise agreements priced on token usage. Seat-based software is predictable: the contract value is a ceiling and finance can forecast it easily. Under usage-based pricing the contract value becomes a floor, and the adoption curve drives the bill. In one client example, moving from a team plan to an enterprise plan put projected spend at about three times the original budget, with no increase in users.
The lesson explains why the usual controls struggle:
- Approval gates measure the wrong thing. They key off contract value, while consumption grows underneath once the tools are embedded in daily work.
- Usage keeps climbing. AI lets people do more of the same work, such as a recruiter going from four interviews a day to seven, so consumption grows even with headcount flat.
- Hard limits are hard to hold. Most tools meter by the month, not the day, and cutting people off mid-month triggers strong pushback.
- Defaults cost money. Users often default to the most expensive model without realizing it, and default settings can differ across desktop, phone, Excel and Word.
It closes with what has worked in practice: targeted training sessions by user group (heavy users, non-users, expensive-model users) that both teach better habits and surface why people use the tools the way they do. That education measurably reduced spend.
Key takeaways
- Accept that some AI spend is R&D, and that any AI forecast is likely to be significantly wrong.
- Expect a spend spike with each new group of users, then a decline as habits improve, then growth again as adoption deepens.
- Usage-based pricing turns the contract value into a floor, not a ceiling.
- Education on model choice and usage is one of the most effective cost controls.
Read the full article: The Invoice Nobody Approved →
Course Assessment
🎉 Congratulations!
You have successfully passed this course.
Your certificate has been generated and is ready for download.
