03 AI

AI approval check

Should we approve this AI feature or workflow, and what does it do to our margin or our costs?

For CEOs, CFOs and operations leaders. Takes about four minutes. Opens with an example.

What you're approving and your numbers

1What are you approving?

A feature customers useMargin at full volume An internal workflowCash return and payback

Industry

Currency

2The feature

Type of feature

Picking a type fills typical token counts and use. Money fields keep your numbers.

Feature name, for the report

Paid seats or users

3Price and margin

Revenue per seat, per month

Gross margin today

How you'll charge for it

Included in the plan Paid add-on Price increase

Add-on price per seat, per month

Price increase

Lowest gross margin you'll accept

4Use at full volume

Share of seats that use it

Uses per active seat, per month, in the pilot

Full-volume use compared with the pilot

Heavy users and new use cases usually push use up after launch.

5Cost of one use

How you know the model cost

Estimate from tokens I know the cost per use

Input tokens per use

Output tokens per use

Price per million input tokens

Price per million output tokens

Example prices. Use your provider's current price list.

Model cost per use

Other cost per use: vendor fees, retrieval, monitoring

Outputs your team must review or fix

Cost of one review

6The eight questions

Mark how well the team can answer each one today.

The answer

Feature approval request

Approved

Question	Answer	Mark

Take the result with you

A PDF report with your inputs, results, charts and method. Or the same as plain text for an email.

How this tool works

What it does

It checks an AI proposal two ways before you approve it. For a feature customers use, it works out what one use costs at full volume and what that does to gross margin. For an internal workflow, it works out the hours freed, how much of that becomes cash, and how fast the setup cost pays back.

Both modes then check readiness: eight questions that predict whether the AI earns its place. The result is one of three stamps, with each condition written out.

What you enter

Feature mode. The type of feature, price and margin, how you'll charge, use at full volume, and cost per use from tokens or a known price.
Workflow mode. One of 44 common workflows, cases a month, minutes per case today and with AI, extra checks, the cost of an hour, and what happens to the freed time.
Running costs. Vendor fees, monthly running cost, setup cost, and the cost of errors that get through.
The eight questions. Mark each one clear, partly answered, or not answered yet.

What you get

Feature mode: gross margin at full volume against today and your floor, the price that holds the floor, the highest cost per use you can carry, and a chart of margin against use.
Workflow mode: net cash a month, payback, first-year cash, hours freed, and a chart of net cash against the share of freed time that becomes cash.
The stamp and conditions in red pen, plus a PDF report for the approval meeting.

The maths, assumptions and sources

Feature mode. Cost per use = input tokens × input price ÷ 1,000,000 + output tokens × output price ÷ 1,000,000 + other cost per use + review share × cost of one review. AI cost per seat = cost per use × pilot uses × full-volume multiple × share of seats using it. New gross margin = 1 − (today's cost of serving a seat + AI cost per seat) ÷ new revenue per seat. With a paid add-on, the add-on revenue counts on the seats that use it.

Workflow mode. AI cases = cases × share using AI. Minutes saved per AI case = minutes today − minutes with AI − share needing a check × minutes per check. Hours freed = AI cases × minutes saved ÷ 60. Value of freed time = hours × loaded hourly cost. Cash from freed time = value × share that becomes cash. If the AI adds work, the full extra cost counts. Net cash a month = cash from freed time − AI cases × cost per use − running cost − AI cases × error rate × cost per error. Payback = setup cost ÷ net cash a month. First-year cash = 12 × net cash − setup cost. Full-time equivalents assume 160 working hours a month.

Why the margin floor matters: AI products run on thinner margins than classic software. ICONIQ projects average AI product gross margin at 52% in 2026, up from 41% in 2024. Software companies have long benchmarked at 60 to 80% or more. Workflow presets are typical starting points, not benchmarks. Time 50 real cases before you trust any saving.

ICONIQ Capital, State of AI: Bi-Annual Snapshot (January 2026), as summarised by SaaStr.
Casado, M. and Bornstein, M. (2020). The New Business of AI. Andreessen Horowitz.
FinOps Foundation, Unit Economics. NIST, AI Risk Management Framework.
Token prices change often. The example prices on this page are placeholders. Check your provider's current price list.