Why AI Projects Fail and How to Keep Yours on Track — Iavor Bojinov

Iavor Bojinov studies how AI products get built and put into use, and his starting point is a number. Most AI projects fail, with some estimates putting the rate as high as 80 percent, close to double the failure rate of corporate IT projects a decade earlier. The article asks why that happens and what a leader can do about it at each stage.

The difference he identifies is that AI projects are not deterministic. An IT project has a known end state, and running it twice gives the same answer. An AI project carries every difficulty of a technology programme plus real uncertainty about whether the thing will work at all, which makes the familiar way of managing such projects a poor fit.

His answer is a sequence of five stages: selection, development, evaluation, adoption and management. Selection is deciding which project to run in the first place. Development and evaluation are where most of the attention goes, and where many organisations stop. Adoption and management are where projects that technically work still fail, because nobody uses them or nobody keeps them running. The framing is useful because it puts most of the risk outside the model, in the decisions surrounding it.