Most AI projects fail before a line of code is written

There is a comfortable story about why AI projects fail. The story says the technology was not ready, or the data was not clean enough, or the vendor over-promised. It is comfortable because every one of those explanations puts the failure somewhere outside the room where the decision was made.

In our experience the failure happens earlier and closer to home. It happens in the first meeting, when a project is defined as “we should do something with AI in customer service” rather than “our service team spends eleven hours a week finding information that already exists in our own systems, and we want that down to two.”

The difference those two sentences make

The first framing cannot fail, because it cannot be measured. It also cannot succeed. Six months later there is a pilot, some enthusiasm, a demo that goes well, and no way to answer the only question the CFO is going to ask: what did we get for it.

The second framing is uncomfortable, because it commits. It names a process, a cost, and a target. It tells you what data you need and who has to change how they work. It tells you, quite early and quite cheaply, if the idea is bad.

What a well-defined AI case actually contains

Before we build anything, we want four things written down:

• The process, named specifically. Not “customer service” — the exact sequence of steps a named team performs, and where in that sequence the time goes.
• The current cost. Hours, error rates, delay, lost revenue. A number someone in the business recognises as true.
• The target, and who owns it. What the number should become, and which person’s objectives it sits under. If nobody owns it, it will not happen.
• The change in behaviour. What people will do differently on the Monday after go-live. This is the one most often skipped, and the one that most often kills the project.

None of that requires knowing which model you will use. All of it can be done in a couple of weeks.

Why this gets skipped

Because it is slower at the start, and because AI is unusually good at producing an impressive demo before anyone has thought about the business case. A convincing prototype is easy now. That is exactly why the discipline matters more than it used to: the demo no longer tells you whether the project is real.

The cheapest thing you can do

Take the AI initiative currently furthest along in your organisation and try to write those four things down in a single page. If you can, you are in good shape. If you cannot, you have just saved yourself a great deal of money, and you know precisely what to fix first.

That page is roughly what we produce in an AI Opportunity Sprint — usually across a portfolio of candidate cases rather than one, and with the scoring that lets a leadership team choose between them. But you do not need us to start. You need someone in the room willing to ask what the number is.