Is your AI paying for itself?
Hi Everyone,
If your board asked tomorrow whether the AI tools you are paying for are worth the money, could you answer with one number?
Most finance teams can't.
When the FinOps Foundation surveyed 1,192 finance professionals this year, 40% said they were unable to quantify the value produced by their organization's AI spending.
The problem isn't necessarily that the AI is failing. It's that most companies are measuring the wrong thing.
Here are four questions that will help you defend your AI investment to the board—without talking about tokens.
Why tokens don't answer the question
AI tools may charge by token, credit, seat, conversation, or API call.
Those measures tell your finance team what the AI costs. They don't tell them what it produces.
Two teams can incur exactly the same token bill while producing very different amounts of useful, completed work.
Some of the largest AI software vendors have already recognized this problem and changed the way they charge.
In April, HubSpot reduced the price of its Breeze Customer Agent from $1.00 per conversation to $0.50 and now charges only when the AI successfully resolves the conversation.
Intercom, Zendesk, and Sierra use a similar principle: customers pay for completed outcomes rather than simply for activity or usage.
These companies changed their pricing because customers wanted a clearer connection between cost and value.
You should apply the same principle to the AI tools your business already pays for.
Four questions your finance team should be able to answer
1. What completed outcome does the AI produce?
A strong answer identifies something specific, finished, and countable.
For example:
- A resolved customer-support conversation
- A validated supplier invoice
- A qualified sales meeting booked
- A completed contract ready for signature
Be cautious of answers such as “increased productivity,” “faster response times,” “greater efficiency,” or “an improved customer experience.”
Those may be benefits, but they aren't completed outcomes.
Ask instead:
What specific piece of work does the AI finish?
If nobody can identify one, stop there. You aren't yet in a position to measure its value.
2. What did that outcome cost before AI?
Calculate the fully loaded cost of completing the same work before the AI was introduced.
That should include the employee's salary, benefits, software, management time, and any other costs associated with delivering the outcome.
Then divide the total by the number of outcomes produced.
For example, if a support team costs $500,000 a year and resolves 50,000 conversations, the previous cost was $10 per resolved conversation.
If the company never measured this, use the best reasonable estimate available. An imperfect baseline is still more useful than having no baseline at all.
3. What does the same outcome cost now—all in?
Include every cost required to produce the result:
- The AI software subscription
- API or usage charges
- The time employees spend reviewing or correcting the AI's work
- Human escalations
- Additional software or infrastructure
- Ongoing management and maintenance
Then divide the total cost by the number of completed outcomes during the same period.
That gives you your true cost per outcome.
It will almost always be higher than the vendor's advertised price, because the headline price rarely includes the surrounding human and technical work.
4. Was the saving worth the investment required to achieve it?
Cost per outcome tells you what the workflow costs to operate.
It doesn't tell you how much you spent to create it.
Add the one-time costs of implementation, including engineering time, integration work, process redesign, testing, training, and employee retraining.
Then compare that setup cost with the annual savings generated by the new workflow.
For example, if implementation costs $300,000 and the workflow saves $600,000 a year, the payback period is six months.
A payback period of a few months is usually an easy case to defend.
A longer payback period may still be worthwhile, but it needs a broader justification than cost reduction alone. The case might depend on improved quality, additional revenue, faster growth, reduced risk, or work that a human team couldn't perform at the required scale.
The board doesn't need to know how many tokens you consumed. It needs to know what the AI completed, what that outcome used to cost, what it costs now, and how long it will take to recover the investment.
If your finance team can answer those four questions, you can explain the value of your AI spending in language the board already understands.
Go deeper
👉 MarTech: HubSpot moves to outcome-based pricing for some Breeze AI agents — the actual HubSpot pricing announcement from April. Useful if you're negotiating with an AI vendor and want a real benchmark to point at.
👉 SaaStr: HubSpot Switching AI Pricing From Per Use to Per Resolution — Jason Lemkin's read on why HubSpot, Intercom, Zendesk and Sierra are all charging by outcome. The clearest single explanation of where AI pricing is heading.
👉 FinOps Foundation: State of FinOps 2026 — the survey behind the 98% number, with 1,192 finance practitioners explaining what they still can't measure about their AI spend.
👉 Cogent Infotech: From Cost Per Token to Cost Per Workflow — a longer walk-through of cost per outcome, worth reading if you're going to bring this to your CFO in the next quarter.
Coming up tomorrow
In tomorrow's issue, you'll get a short Q-end audit that makes each owner defend anything they want to keep going into Q4.
P.S. One of the things we track is the AI research cost per issue. This one was $0.15. What outcomes will you track?