Chapter 11
Play 6: Run the Economics of AI
From seats to actions, from spend to proven value
Play 6: Run the Economics of AI
Goal: you know what AI costs, who it serves, and what it returns — and you can prove it.
Why this is a finance discipline now
When software was priced per seat, cost tracked headcount and was settled at renewal. When AI is priced per token, per action or per conversation, cost tracks usage, moves every day and lands across many teams at once. Without a discipline for it, AI spend becomes the new cloud bill: large, growing and owned by nobody.
How to run it
- Meter the spend. Capture tokens, model calls and agent runs at the rate in force when they happened. Input, output and cached tokens are priced differently; volume is not cost.
- Assign it. Allocate every dollar of AI spend to the team, product and outcome it served, in the same cost model as cloud, software and people.
- Budget it. Give teams AI budgets with alerts as they approach them, and route expensive work to the right model for the job.
- Measure unit economics. Cost per task, per resolved request, per readout — whatever unit the work is measured in.
- Prove the value. Trace AI spend through the team and the work to the outcome it was meant to move and the metric that measures it. Count value only when the link and the evidence are both in place: planned, then proven, then banked.
A simple value chain
| Step | Question |
|---|---|
| Spend | What did the AI cost? |
| Team | Who used it? |
| Work | What was it working on? |
| Outcome | What was that work meant to move? |
| Metric | Did it move — and is the evidence good enough to count? |
Outputs
- AI spend metered and allocated alongside the rest of technology spend.
- Budgets and alerts by team.
- A value readout that separates proven from claimed.
Signs it's working
- Finance can forecast AI spend within a reasonable range.
- The AI program's return is reported with evidence, not anecdotes.
Pitfalls
- Counting activity as value. Tasks completed isn't the same as outcomes moved.
- A separate AI cost tool. AI spend reported apart from the rest of technology spend can't be compared or traded off.
- Ignoring the people cost. AI changes how labor is spent. Measure both together.