World Model Readiness Demos are cheap, production is not

Module · Demos are cheap, production is not

The Post-Pilot Scaling Check

Most AI pilots succeed and then quietly never ship. The demo works, the room applauds, and the thing dies somewhere between the slide deck and a system that runs on Monday. This module checks the five places pilots go to die: the success bar, the production owner, the integration debt, the economics at real volume, and whether your company can ever say stop.

Question 1 of 5 · Success bar was set

Did this pilot start with a number that decides go or no-go?

A pilot without a pre-agreed threshold cannot fail, which means it cannot really succeed either. 'It went well' is not a criterion; 'cut handling time 30 percent across 500 real tickets' is. Set the bar before you see the results, or the results will set it for you.

Question 2 of 5 · Production owner named

Is there a named owner who will run this in production, not just the pilot?

Pilots are owned by innovation teams; production is owned by whoever gets paged when it breaks at 2am. If that handoff has no name and no capacity attached, the pilot has nowhere to land, however well it performs.

Question 3 of 5 · Integration debt scoped

Do you know what it takes to wire this into the systems it must live in?

A demo runs on clean sample data through one friendly API. Production means legacy systems, permission models, edge cases and data pipelines that fight back. That gap is the integration debt, and it is usually where the months and the money go.

Question 4 of 5 · Unit economics hold

Do the per-transaction costs still work when volume is a hundred times the pilot?

Pilot economics flatter you: low volume, subsidised vendor pricing, and a human quietly fixing the model's mistakes off the books. At scale the token bill, the review labour and the error handling all grow together. Model it before you sign, not after.

Question 5 of 5 · Kill discipline exists

Can this pilot be killed, and has your company ever actually killed one?

Pilots that cannot die accumulate as zombie projects, consuming budget and attention nobody will admit is lost. A company that has never stopped a pilot will not stop this one either; it will scale it just to avoid the awkward conversation.

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 many AI pilots has your company run in the last two years?

None yet
One or two
Three to five
Six or more
We have lost count

How many of your AI pilots have reached durable production?

None so far
One
A few
Most of them
Too early to tell

Has your company ever deliberately stopped an AI pilot that was not working?

Never
Once
Occasionally
Routinely
No pilots yet

Your context

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