The AI Idea Stoplight

1

What problem does this solve?

Write it as a sentence about the work, with no AI in it. "Our support team retypes information from PDFs into the CRM" is a problem. "We should use AI for support" is not. If you can't write the sentence, you don't have an idea yet, you have a feeling about a technology.

Red if you can't write the sentence without the word AI in it.
2

What is that problem costing you right now?

In hours per week, errors per month, deals lost, or salary. A rough number is fine, a made-up one isn't. This is the question that kills most ideas, and it should. If the problem costs you four hours a month, no AI solution is worth the project around it.

Red if the honest number is small, or if nobody can give you one at all.
3

What data does it need, and do you actually have it?

Where does the data live, who owns it, is it clean enough for a person to use it today, and are you allowed to use it that way under GDPR? "We have a lot of data" is not an answer. Most AI projects that fail late fail here, and it was knowable early.

Red if the data sits somewhere else, belongs to someone else, or is off limits under GDPR.
4

Who is going to use this, and will they?

Name the actual people. Then ask whether it fits into how they already work or requires them to change their habits. A tool that saves twenty minutes but lives outside the system someone already has open all day usually gets abandoned in week three.

Red if you can't name the people, or if it lives outside the tools they already have open.
5

What happens when it gets it wrong?

Not if. When. Does someone catch it, or does it go straight to a customer, a regulator, or a financial system? An AI that drafts an internal summary and an AI that approves a payment carry completely different risk, and they deserve completely different levels of caution. If a wrong answer is expensive and nobody checks it, stop here.

Red if a wrong answer is expensive and nobody sees it before it leaves the building.
6

Buy, build, pilot, or stop?

If a tool already does this well, buy it. If it's specific to how you work and the value is real, pilot it small before you build. If it only survived the first five questions on optimism, stop, and spend the money on the boring process fix that would have solved it anyway.

Red if the only thing keeping the idea alive is that someone senior likes it.

How to read your light

There's no score. There are three lights, and you get the worst one you earned.

●

Green, nothing red

Real problem, real cost, data you have, users who want it, and a manageable failure mode. Pilot it, small and time-boxed, with a number you agreed in advance that decides whether it continues.

●

Yellow, what you can fix

Usually the failure mode, sometimes the users. The idea might be right and the timing wrong. Either add a human check and shrink the blast radius, or wait until you can. Don't put it live and hope.

●

Red, what you can't fix

Almost always cost or data. The most common result, and it's a good one. You just saved a quarter. Write down what would have to change for the idea to come back, then move on.

The AI Idea Stoplight

How to read your light

Green, nothing red

Yellow, what you can fix

Red, what you can't fix

This is one hour of a full day