The most useful test I know for a new idea is also the most brutal one:
If we stopped working on it tomorrow, who would care?
Not the founder. Not the team. Not the investor. Not the internal sponsor who likes the technology.
The customer.
This sounds simple, but it cuts through a lot of noise. Many projects survive because they are interesting internally. They create meetings, prototypes, dashboards, demos, and momentum. They give people a reason to explore a technology and a story to tell afterward.
But nobody outside the building is waiting for them.
That is where many ideas become zombies. They are not dead enough to kill. They are not alive enough to matter. They keep consuming attention because the team has already invested time, or because the technology is exciting, or because stopping would force a difficult conversation.
The market does not care about that.
Urgency Has to Live Outside the Building
A real idea creates pressure somewhere else.
Somebody is losing time. Somebody is carrying risk. Somebody is paying for a workaround. Somebody is explaining the same exception every week. Somebody is avoiding a decision because the system does not give them enough trust.
That is the place to start.
Not with the feature.
With the absence.
What happens if this thing does not exist? Who keeps suffering? Who keeps paying? Who keeps doing manual work? Who keeps accepting a risk they do not want to accept?
If the answer is vague, the idea is probably vague too.
This is especially true with AI. AI makes it too easy to build something that looks impressive in a room. A demo can summarize documents, answer questions, classify requests, write reports, or generate a plan. The room nods because the capability is visible.
But capability is not demand.
The better question is not “Can AI do this?”
The better question is: “Who has a painful enough problem that this becomes necessary?”
Name the Buyer, Not Just the Budget
A budget is not a buyer.
A budget can be allocated because the company wants to experiment, because a board asked about AI, because a department received transformation funding, or because nobody wants to look late.
A buyer is more specific.
A buyer has a problem with a cost attached to it. The cost might be money, delay, risk, customer churn, audit exposure, employee frustration, or operational fragility. But it has to be real enough that doing nothing is no longer neutral.
In AI pilots, this is often where the weakness appears.
The team can describe the technology. They can describe the model. They can describe the workflow. They can describe the possible benefit.
But they cannot name who loses if the pilot stops.
That is not a small gap. That is the gap.
If nobody outside the project is hurt by its absence, the project is probably not a priority. It may still be a useful experiment. It may teach the team something. It may create internal learning.
But it is not yet a business idea.
The Cost of Doing Nothing
The second part of the test is numbers.
Not a fantasy ROI spreadsheet. Not a consultant’s promise. Something closer to operational evidence.
How many hours are lost each week? How many exceptions require manual handling? How many errors reach the customer? How much revenue waits on a decision? How often does the same person become the bottleneck? What happens during an audit, a peak season, a rollout, or a customer escalation?
A good idea can usually point to one of these with some precision.
A weak idea hides behind narratives: better productivity, smarter workflows, improved experience, more innovation. Those words may be true, but they are not enough to carry a project through the hard part.
The hard part is not the demo.
The hard part is the moment after the demo, when the work has to change and somebody has to own the result.
That is when a weak idea starts asking for patience. A strong idea already has urgency behind it.
Why This Matters More in AI
AI lowers the cost of building plausible things.
That is good. It means more people can test more ideas quickly. It also means companies can create more noise than ever before.
Every department can have a chatbot. Every process can have a copilot. Every report can have an assistant. Every internal document can be turned into a question-answering interface.
Some of these will matter.
Many will not.
The difference is not intelligence. The model may be very good. The interface may be polished. The architecture may be solid.
The difference is whether the work underneath has a real wound.
If a field team is losing hours because scanner exceptions never reconcile properly, that is a wound. If a security team cannot prove where sensitive data went during an AI workflow, that is a wound. If forecasting depends on two people who hold the business logic in their heads, that is a wound. If a board wants AI adoption but nobody can say who owns the risk, that is a wound.
Those are ideas worth pursuing because absence already has a cost.
The Rule
The rule is simple:
Do not ask whether the idea is interesting.
Ask who is worse off without it.
If the answer is clear, keep going. If the answer is weak, stop before the project becomes another polite internal hobby.
Brutal, yes.
But in a world where AI can produce endless prototypes, being brutal about demand may become one of the few disciplines that still protects the work.