Define input and output
Connect token volume and cost to an outcome with economic meaning: revenue, conversion, retention, resolution, cycle time, risk, or quality.
AI production economics
How much business value do your AI tokens create?
Tokens are the input. Revenue, conversion, retention, resolution, and quality are the outputs. I help executive teams estimate the response curve between them—so AI ROI becomes measurable rather than assumed.
Get a defensible answer your CFO, board, and operating leaders can use to decide what to scale, fix, or stop.
Independent analysis. Decision-ready evidence. No vendor scorekeeping.
Executive measurement brief
Decision-ready
When token use increases, how much does the business outcome move?
Token → outcome curve
Estimated, not assumed
Marginal yield
Output per additional 1M tokens
Recommendation
Scale / optimize / stop
01Measure tokens
02Define the output
03Estimate the response curve
04Allocate the next token
The measurement problem
Token counts become useful when they are connected to outcomes.
A usage dashboard tells you how many tokens were consumed. An economic model estimates what those tokens produced, whether marginal returns are rising or flattening, and how much of the observed outcome would have happened anyway.
Estimate the response curve
Model how outcomes change with token consumption, including nonlinearities, diminishing returns, workload mix, and lagged effects.
Separate cause from correlation
Use experiments, panel models, matching, interrupted time series, or synthetic controls to estimate the incremental effect.
Allocate the next token
Identify where another million tokens is likely to create the most value—and where usage has reached diminishing returns.
Annual estimate
ROI results
Status
Recommendation
Token-to-revenue estimate
The lower-bound estimate uses a simple OLS prediction interval. It is a conservative planning guardrail, not causal proof.
Model routing scenarios
Lower-spend ROI scenarios
Cost discipline can materially change AI ROI. Use this as a rough model-routing, caching, prompt-pruning, or vendor-optimization scenario - not as a guarantee.
| Scenario | Adjusted AI spend | Net return | ROI |
|---|
Export
Copy a summary or download a CSV. Entered dollar values are not sent to the server.
Want a more rigorous read on AI ROI?
Move from a directional calculation to a causal measurement design and an executive capital-allocation decision.
Start the conversationConfidential conversation
Bring the question. I’ll help design the answer.
Tell me where your AI program stands and what decision is blocked. I’ll review the question and follow up directly.
- Measurement strategy grounded in your available data
- Methods selected for the decision—not for show
- Clear uncertainty, limitations, and next steps
Your inquiry is stored privately and used only to respond to you.
FAQ
What executives usually ask next
What is AI ROI?
AI ROI estimates the incremental business output generated by AI relative to its total cost. Token volume is the measurable production input; revenue or another valued outcome is the output.
Why model revenue as a function of tokens?
It reveals the economic response curve: how output changes as token consumption grows, where marginal returns begin to flatten, and whether the next block of usage is likely to pay back.
What should I count as AI spend?
Include model/API costs, software subscriptions, infrastructure, implementation costs, and ongoing labor where relevant.
Can we measure AI ROI without a randomized experiment?
Often, yes. The right quasi-experimental or time-series design depends on rollout timing, comparison groups, outcome frequency, and how adoption was assigned.
What if our data is incomplete?
The first step is a measurement audit: identify what can be estimated now, which assumptions matter most, and what instrumentation would make the next decision more credible.