World Model Readiness
Engraved commercial instrument

For your company

Money & Vendors

Module · What one run actually costs

The AI Unit Economics Check

AI does not cost what the licence says. It costs per call, per token, per document, and the bill grows with use in a way that a flat subscription trained nobody to expect. The danger is a workflow that looks like a productivity win while quietly costing more than the work it replaced. This module checks whether you can see the cost of a single run, tie it to the value it creates, get warned before it runs away, and switch off the automations that lose money.

Question 1 of 5 · Cost per use is visible

Do you know what a single run of your main AI workflow costs?

Not the monthly bill: the cost of one run. One drafted email, one processed invoice, one answered ticket. If you only see the total, you cannot tell a cheap workflow used often from an expensive one used rarely, and only one of those is a problem.

Question 2 of 5 · Costs land on owners

Can you attribute AI spend to the team or product that caused it?

When AI cost sits in one central bucket, no owner feels it and nobody optimises it. Attribution, by team, by product, by workflow, is what turns an invisible shared expense into a number someone is accountable for and motivated to bring down.

Question 3 of 5 · Value is measured too

For your AI workflows, do you measure the value they create, not just the cost?

Cost without value is half the equation and the misleading half. A workflow that costs more than last year can still be the best money you spend if it produces more. The number that matters is cost against the value or the work it replaces, per use.

Question 4 of 5 · Runaways trip an alarm

If an AI workflow suddenly costs ten times more tomorrow, would you find out fast?

AI costs spike quietly: a loop, a larger model, a jump in volume, an integration calling in circles. Without an alert tied to spend, the first sign is the invoice at month end, by which point the damage is done and repeated thirty times over.

Question 5 of 5 · Losers get switched off

Do your AI automations have a cost line at which you switch them off?

Not every automation earns its keep, and the honest answer is to kill the ones that do not. A kill criterion set in advance, a cost or a cost-to-value ratio, turns switching off from an admission of failure into a decision you already agreed to make.

For the statistics · one click each

Three questions for the public picture

These do not affect your score. They feed the anonymised, aggregated statistics; groups under 8 respondents are never shown.

How well can you see what your AI actually costs?

One total bill only
By vendor
By team or product
Down to the run
We do not track it

Do you know whether your AI spend pays for itself?

No idea
We believe so
For some workflows
Measured and positive
Too early to say

Has an AI bill ever surprised you?

No surprises
Once
More than once
It is a running problem
We cannot tell

Your context

Used to calibrate the report. Company size and sector remain in the anonymized dataset; your email does not.

What the five levels look like

Every dimension in this assessment is scored 1 to 5. This is what the levels mean, dimension by dimension. The graded report diagnoses where your own answers land and what to do about it.

Cost per use is visible

  1. 1No idea
  2. 2Total bill only
  3. 3Rough estimate
  4. 4Measured per workflow
  5. 5Measured per run

At the low end: If you cannot cost a single run, you cannot tell which workflows are worth it. Start by dividing this month's AI bill across the workflows that generated it; even a rough split is a beginning. What good looks like: Cost measured per run is the foundation everything else here stands on. Keep the measurement close to real time; a cost you learn about a month late is a cost you cannot manage.

Costs land on owners

  1. 1One central bill
  2. 2Split by guesswork
  3. 3Tagged partially
  4. 4Allocated by team
  5. 5Allocated to workflow

At the low end: A single central AI bill is a cost nobody owns and therefore nobody controls. Tag spend by team or product so the number reaches the person who can actually change it. What good looks like: Spend allocated down to the workflow puts the cost in front of the person who can cut it. Keep the allocation visible monthly; attribution that nobody reads stops changing behaviour.

Value is measured too

  1. 1Cost only
  2. 2Value assumed
  3. 3Value estimated once
  4. 4Value tracked
  5. 5Value against cost, ongoing

At the low end: Watching cost alone tells you what AI takes, never what it gives back. Pair each workflow's spend with a simple measure of what it produces or saves, so you are judging a ratio, not a bill. What good looks like: Value tracked against cost per use is how you know which workflows to feed and which to starve. Revisit the value measure as the work changes; yesterday's saving can quietly become today's overhead.

Runaways trip an alarm

  1. 1No alerts
  2. 2Notice at invoice
  3. 3Monthly review
  4. 4Threshold alerts
  5. 5Real-time anomaly alerts

At the low end: With no spend alert, a runaway workflow bills you all month before anyone notices. Set a simple threshold alert on daily AI spend this week; it is the cheapest insurance here. What good looks like: Real-time anomaly alerts mean a runaway fails cheap and loud instead of quiet and expensive. Tune the thresholds as volume grows, or the alarm you stop trusting is the one that never fires.

Losers get switched off

  1. 1Never killed
  2. 2Kill by argument
  3. 3Kill after it hurts
  4. 4Kill criteria set
  5. 5Reviewed against criteria

At the low end: An automation that can never be switched off is a cost with no ceiling and no exit. Agree a cost or a cost-to-value line in advance, so ending it is a rule, not a fight. What good looks like: Automations reviewed against a preset kill line means your AI portfolio prunes itself. Keep the review on a schedule; a kill criterion nobody checks is the same as not having one.