What makes an AI project worth funding?

Boards approve AI budgets under pressure: competitors are “doing AI,” vendors promise transformation, and no one wants to be the laggard. That pressure funds a lot of projects that were never going to pay off. A worthwhile AI initiative survives a short list of unglamorous questions. If it can't answer them, it isn't ready for funding.

1. What decision does this improve?

Not “what model does it build”: what decision does it change, and who makes that decision today? If the answer is vague (“better insights,” “a single view of the customer”), the project is a technology purchase, not a business case. Fundable projects name a specific, recurring decision and how it will be made differently.

If you cannot name the decision that changes, you are funding a dashboard, not an outcome.

2. What is the decision worth?

Every fundable project has a value estimate: decision frequency times the improvement times the value per decision. The number can be a range. It just has to be honest and defensible. “We make this call 200,000 times a year and a 3% improvement is worth roughly $X” is a business case. “AI will make us more efficient” is not.

3. Do we have the data, with outcomes?

AI learns from history. The critical question is not whether you have data, but whether you have labeled outcomes: did the customer churn, was the claim fraudulent, did the offer convert? Without outcomes, there is nothing to learn from. Data readiness is where most timelines quietly slip, so it belongs in the funding conversation, not after it.

4. Can we act on the output?

A recommendation no one executes creates zero value. Before funding, confirm there is a process, a team, and a system path to act on what the model produces. The best-run initiatives design the action first and the model second.

5. How will we know it worked?

Fundable projects define success up front, in business terms, and ideally with a way to measure lift: a holdout group, a before/after, a controlled rollout. If the plan has no way to prove causal impact, you will never be able to say whether the money was well spent.

6. What is the smallest version that proves value?

Beware the eighteen-month platform build. The strongest proposals start with a contained pilot on one decision, prove lift, and earn the right to scale. Fund the pilot, set the success metric, and make the next tranche conditional on results. That discipline is the difference between an AI program and an AI expense.