You Probably Have an AI Cost Problem

On October 29 I attended the second session of Flexera’s three-part webinar series, The AI Cost Crisis. Its theme was AI spend that doesn’t go through procurement, the AI that arrives inside software a company already owns, through individual purchases and departmental experiments, and through autonomous agents. Afterward I went back through the first session, where Forrester principal analyst Tracy Woo laid out the numbers. After more than 30 years of negotiating software agreements, I recognized most of the problems she described as contract problems.

How Big the Problem Is

As presented in the series, 82% of the enterprise AI cost decision makers Forrester surveyed have at least one generative AI application in production, and 76% have at least one agentic AI application. Their AI spend grew about 40% over the past year, and they expect it to grow five to nine times next year.

Ownership hasn’t kept up. When Forrester asked who owns AI strategy, no single role, not the CIO, not the CFO, not anyone else, came in above a quarter of organizations. In the same survey, 30% of respondents said embedded AI charges, the AI fees inside software they already pay for, leave them without visibility. And the meters don’t match. Tokens, requests, credits, throughput and storage all measure different things, and a single vendor can bill in several of them.

Then there was the story that stayed with me. Woo described a company where one employee asked an AI tool to search all of the organization’s documents. It searched 15,000 of them, and that single request cost about $30,000.

Flexera’s advice was sound and mostly operational. Establish a baseline, tag and attribute spend, assign owners and set budgets. I agree with all of it. But every one of those steps measures spend after it happens.

The Price Was Set When Someone Signed

The rate per unit, what happens when usage runs over, whether the vendor can raise the price mid-term, and who pays when an automated process runs away were all agreed to by people, months or years earlier, in a master agreement, an order form or a renewal. Often they were agreed before AI was part of the product, by someone who has since moved to another role.

A dashboard can show you a $30,000 request. It can’t renegotiate the terms that allowed it. A usage cap, a spending alert the vendor is obligated to send, or a clause saying who pays for runaway consumption would have.

Seven Places AI Cost Hides in Your Software Agreements

These are my own views, not Forrester’s or Flexera’s. They’re the terms I’d look at before signing, and again before renewal, with any vendor that charges for AI.

1. What a unit actually is. If the price is per token, per request or per “AI credit,” the agreement needs to define it. If you can’t translate the unit into a number you can forecast, you can’t budget for it.
2. Overage and minimum commitments. Confirm what happens when you exceed your allowance, whether unused credits expire, and whether you pay for a commitment you don’t use.
3. Price changes. Look for a notice period and a ceiling on increases, and check whether the price protection you negotiated extends to AI features added later. Consumption pricing makes it easy for rates to drift.
4. Controls on runaway usage. Ask for spending caps, alerts and the right to throttle usage. Settle who pays when an automated process runs wild, and whether it matters that the cause was the vendor’s software rather than your configuration.
5. Visibility and audit rights. You should be able to see usage in enough detail to check the bill and attribute it to specific teams.
6. Terms that live on the vendor’s website. Many software agreements incorporate the vendor’s online terms, acceptable use policy or AI addendum by link, and let the vendor update those pages. A new AI fee or a new right to your data can arrive that way without anyone at your company signing anything.
7. What the vendor may do with your data. Embedded AI often comes with a right to use customer data to improve the vendor’s models, and opting out can sit in a higher tier or cost extra. That’s a price too, even when it never shows up on an invoice.

AI Charges That Arrive Without a Purchase Order

Embedded AI is the part I’d put in front of a CFO first. Most software agreements signed a few years ago didn’t contemplate AI. Now product updates and upgrades include it, and the charges, often as a variable cost, get passed through. They rarely arrive as a new purchase with a purchase order. They arrive through a renewal, an amendment, a new order form attached to an agreement signed two years ago, or an update to terms the agreement points to.

No single document shows the whole picture. The master agreement says one thing, the order form says another, an amendment changes both, and the vendor’s online terms may have changed since. Finding the AI fee means reading them together.

Ownership Is a Record of Decisions

Woo’s central question was who owns AI spend. I’d answer it from the other direction. Every AI charge traces back to a person who agreed to it. Someone signed the order form, accepted the renewal, let the updated terms stand, or decided not to push back on a usage clause because the deal had to close that quarter.

Most companies keep the signature and lose the reason. When the CFO asks why the company pays per credit on a tool that used to be a flat subscription, the answer is in someone’s inbox, or it left with the person who negotiated it. Ownership of AI spend starts with that record, what was agreed, by whom, when, why, and when it can be changed.

What to Do Before the Next Renewal

Start with the agreements you already have. Pull the contracts for your software vendors, with their order forms, amendments and the terms they incorporate, and ask four questions.

1. Where does AI pricing appear?
2. Who agreed to it, and why?
3. What rights does the vendor have to your data?
4. When does it renew, and what’s the notice deadline?

The fourth question is where the leverage is. Once an agreement auto-renews, the AI terms in it are locked for another term. Before the notice deadline, everything in them is negotiable.

Then review new AI terms before they’re signed, not after the first invoice. And if your company sells software with AI in it, expect your customers’ procurement teams to start asking these questions of your paper. It’s better to know your answers before they ask.

How Revver Contract Intelligence Handles Each of These

We built Revver Contract Intelligence for exactly this kind of reading. Every problem above maps to something specific in it.

• Units, overages and commitments. RCI evaluates every software agreement against standards for Metered Consumption and Overage Control and for Minimum Commitments, Volume Flexibility and True-Down. It checks whether the chargeable unit is defined, how failed, retried and agent-generated requests are counted, when the allowance resets, and what your maximum committed spend actually is.
• Price changes and renewals. The Price Escalation and Adjustment standard tests for a fixed period, an annual ceiling and a named index. The renewal standard checks that you get price information in time to decide before the agreement renews.
• Runaway automation. The standard for Authorized Automated Actions checks whether the agreement says what an automated process may do on your behalf, and that permission to use a tool isn’t permission to commit you to new charges or accept new terms.
• Terms on the vendor’s website and click-through charges. Fixed Referenced Terms and Controlled Future Changes checks that linked terms are pinned to an accepted version and that fees can’t change mid-term without your agreement. Portal Assent and Additional-Form Terms checks that an employee’s click in a vendor portal can’t create a new fee.
• Order forms and amendments that change the deal. RCI reads each order form and amendment with the master agreement it’s linked to, as one family, and flags where a later document overrides a protection the master gave you.
• Your data. The standards for AI in the Purchased Product and for the Prohibition on Training with Protected Buyer Information check whether the vendor has disclosed the AI in the product and whether it may train on your data.
• Findings you can act on. Every finding quotes the contract language behind it, is ranked Must, Should or Nice to address, and comes with the language to send back already drafted. A Review Receipt lists every standard checked, including the ones the agreement met.
• The record of who agreed. Every decision your team makes on a finding is recorded with who made it, when, in which round and why, alongside the emails and meeting notes you attach to the contract. At the next renewal, the reason is still there.
• The notice deadline. Renewal dates and notice windows are captured when an agreement is loaded and kept on a renewal calendar, so the deadline shows up while the terms are still negotiable.

For the software agreements you’ve already signed, Revver and our Advisors run a fixed-fee Portfolio Audit that reads the whole book against these standards and hands back a prioritized list of what to renegotiate, and by when. When it’s time to go back to those vendors, our Advisors can negotiate on your behalf or alongside your team. You get the findings and the help to act on them, so you don’t have to take on the vendors alone.

We hold our own pricing to the same standard. RCI is an annual license to your organization, sized to your contract book. Documents, evaluations and negotiation rounds aren’t metered, there are no AI credits to buy, and a multi-year agreement holds the price for the term. The details are on our pricing page.

The statistics and the $30,000 example are as presented by Tracy Woo, Principal Analyst at Forrester, in “Who’s actually spending your AI budget?”, the first session of Flexera’s The AI Cost Crisis series, hosted by Tracie Stamm of Flexera.

Revver Contract Intelligence is an AI-native platform that evaluates a company’s contracts from its own position, turning the judgment behind every decision into institutional knowledge that compounds over time.