Seven questions to ask before funding an AI idea

Every business now has a list of AI ideas. Most are plausible and some are good. These seven questions sort them faster than a scoring workshop.

1. What decision or task does this replace, and how is it done now?

If nobody can describe the current process in specifics — who does it, how often, how long it takes, what goes wrong — the idea is a capability looking for a use. Write down the current state first. Occasionally that is enough to reveal a simple fix that does not need AI at all.

2. Is there a clear notion of a right answer?

Some work has one: this invoice codes to this account, this policy says this. Some does not: the best wording for a proposal, the right strategic direction.

Both can be worth doing, but only the first can be measured, improved and trusted without a person checking every output. Start there.

3. What happens when it is wrong?

Ask before asking about accuracy. If a wrong answer is caught immediately and costs a minute, a high volume of small errors is tolerable. If it is expensive and hard to detect — a payment released, a customer commitment made, a safety instruction — the bar is different and the supervision has to be designed in rather than added later.

4. Can the data and the process underneath carry it?

This is where most strong ideas become later items. The content is out of date, the permissions are wrong, the process varies by site, or the definitions are disputed.

Saying so at evaluation is much cheaper than discovering it after a failed pilot. It also usually reveals that the preparatory work has value of its own.

5. What is the volume?

Value comes from frequency. A task done four times a year, however painful, will not repay the effort of automating it and will not accumulate enough examples to improve. Look for the work that happens daily.

6. Who owns it after go-live?

Not the project — the running service. Somebody has to handle the complaints, keep the source content current, watch the cost, and decide when something needs changing.

An idea with no answer to this is a pilot, whatever the funding request calls it.

7. What does it cost to run, not to build?

Usage-based billing means the running cost scales with success. Add the content maintenance, the evaluation, the person who owns it, and the support. For many ideas the run cost exceeds the build cost within the first year, and a business case that omits it will be wrong in the quarter after go-live.

How to use these

Run every idea through all seven and most lists shorten dramatically — which is the point. Two funded, owned ideas will deliver more than twenty on a roadmap, and the ones that fall out are not lost. Several will come back later, once the thing they needed underneath them has been fixed.