Most AI projects fail the same way: a demo looks magical, everyone gets excited, and six months later nothing has actually changed about how the work gets done. The gap between an impressive demo and a system people rely on is where most of the budget quietly disappears.
This is a short, practical checklist for deciding whether an AI project is worth building — and for spotting the ones that are theater, not value. It is the order I use to evaluate an AI project before it has a budget: what decision it changes, who owns the result, what good enough means, and what the return has to clear. The order matters, because a project that fails the first question cannot be rescued by a good answer to the fourth. It comes out of the same approach behind the Innovation or Theater decision framework and the AI implementation assessment workbook.
Before you build
- Name the decision or task it changes. If you can't point to a specific decision, output, or step that will be faster, cheaper, or more reliable, you don't have a project — you have a demo.
- Find the human who owns the result. Tools nobody owns quietly stop being used. Someone has to be accountable for the outcome, not just the rollout.
- Write down what "good enough" looks like first. Define the bar before you see the output, or every result will feel impressive enough to ship.
- Weigh the cost of being wrong. Low-stakes, reversible tasks are where AI earns its keep first. Save the high-stakes, hard-to-undo work for after you trust it.
While you build
- Keep a human in the loop where it matters. The goal is reviewed workflows, not unattended magic — decide which steps a person checks and which can run on their own.
- Measure against the old way. If you can't compare the AI path to what you did before, you can't tell whether it's actually better or just newer.
- Ship the smallest useful version. A narrow tool one person uses every day beats a broad platform nobody opens.
Signs it's theater, not value
- The win is "it's so cool" rather than a saved hour or a clear number.
- It only works in the demo — on clean inputs, with the person who built it driving.
- Nobody can say what happens when it's wrong.
- It adds a step instead of removing one.
If a project clears the checklist, it's usually worth building. If it trips on the theater signs, the honest move is to stop — or to scope it down to the one piece that genuinely earns its cost.
This is the kind of call I help with directly. If you're weighing an AI project and want a second opinion before you spend, here is how to work with me — or start from the practical AI implementation hub.