Why an AI use case with no business owner never gets past the pilot

The sequence has become familiar. A use case is identified, a prototype is built in a few weeks, the demonstration impresses everyone. Then nothing. Six months later nobody is using it, and the company concludes that AI was not for them.

It is almost never a technology problem. The projects that make it all have one thing in common, and the projects that die share the same absence: somebody whose own results depend on the tool.

The mechanism

A pilot run by the IT department or by an outside supplier has no business owner. Nobody loses anything if it stops. Nobody fights to get it into daily habits, nobody arbitrates when a process has to change for it to work, nobody defends the budget to scale it. The tool stays a curiosity.

A sales director whose conversion rate depends on a proposal-building tool, by contrast, becomes its strongest advocate. They flag its flaws, they train their teams, they push for the changes they need.

How to qualify a use case

• State the problem without using the word “AI”. If that proves impossible, there is no business problem, only an appetite for doing AI.
• Name a business owner. Not a sponsor in name only: someone whose numbers will move if the tool works.
• Check the data before you start. A use case that rests on unreliable data should be postponed, not launched.
• Put a figure on the expected gain, however rough. An order of magnitude is enough to decide.
• Run one at a time, until it has produced a measured result.

One last observation, and a counter-intuitive one: the uses that work best in mid-sized companies rarely cover what somebody was already doing. They cover what nobody was doing, for lack of time: mining years of documents, tracking a market continuously, handling the first level of customer enquiries.