How do you test whether an AI use case is worth it before investing in a full build?

You test an AI use case by measuring what it costs today, then building a small prototype and checking its results against that measurement. We call this method Measure first, then build.

The measurement comes before the build, and the comparison happens on real cases from the process you already run.

What should an AI proof of concept prove?

A proof of concept for AI should show whether AI changes the cost of one specific process in your business.

A proof of concept answers a narrower question: for this one process, with these real cases, does the AI prototype do the work faster, slower or about the same as today.

Which use case should the prototype take on first?

The choice happens in a workshop. The workshop runs four to six hours. In that time, you and your team pick the one process the prototype will take on, and you sign off on the requirements for it before any build work starts.

The people who actually run the process take part in this workshop. They are the ones who know where the steps are and what counts as a case. That is also where the current cost of the process gets measured, with them in the room.

Choosing the use case and measuring its cost happen in the same session: you cannot judge whether a prototype is worth building until you know what the process costs before AI touches it.

How do you measure what the process costs before AI touches it?

The baseline is measured in minutes per case and cases per month. Both are taken at your own stated rate, so the baseline reflects your process as it runs.

These figures are written into the contract before the build starts. The standard the prototype will be measured against is fixed in advance, agreed by both sides, and in place before a line of the prototype is built.

A prototype can only answer whether AI changes the cost of the process if there is a clear, agreed starting point to compare it with.

How do you test an AI prototype on real cases?

On working days eight and nine of the build, the prototype runs against real cases from the process you chose at the workshop.

Each case gets measured against the baseline that was recorded in the contract: the same minutes per case, the same cases per month, now compared to what the prototype actually achieves.

What happens to your data during an AI prototype?

Before anything is loaded into the prototype, where that data will be processed is agreed in writing.

If your data cannot leave your own systems, that goes into the requirements signed at the workshop, before the build starts.

When is stopping after the prototype the right call?

After the AI prototype has run on real cases, the decision sits with the client. The presentation that closes the AI Prototype states the recommendation plainly, including when stopping is the right answer.

This is what the baseline is for. The measured minutes per case and cases per month give a fixed point to compare the prototype's results against, so the recommendation can be stated plainly.

A clear stop is a useful outcome. It tells you, with a measured comparison behind it, that the use case does not carry a full build, and it lets you move your attention to the next candidate process.