There is almost certainly more than one reason, but it’s most likely that it’s some combination of the seven items below.

1.   The pilot didn't cover the whole problem

The bit that AI covered wasn’t the hard part of the problem that had to be solved, or the pilot ignored all the other changes that needed to be made for a robust solution.

2.   Everyone wasn’t on the same page

The goals weren’t clear, there wasn’t clear communication between the business people and the technical people, or not all of the people necessary for success were brought into the process.

3.   The data wasn’t good enough

Your data wasn’t collected for AI, so it’s probably incomplete, outdated or tainted in some way that makes it difficult to use or unreliable for your AI plan.

4.   AI wasn’t the best answer to the problem you were trying to solve

Enthusiasm for the technology rather than a focus on the problem led the project away from the actual job to be done and the better routes to doing it.

5.   Expectations were out of line with AI capabilities

Generative AI is a brittle solution, and the line dividing where it works well and where it does not work is a ‘jagged frontier,’ where likely areas of failure aren’t always obvious.

6.   The planned users weren’t on board and obstructed implementation

Fear of a threat to their jobs, a reduction in their autonomy or even just a disbelief in AI’s capability may have caused employees to resist, consciously or unconsciously, actively or passively.

7.   People weren’t equipped to take advantage of the technology

Neither the employees nor the managers had the skills or training to take full advantage of the technology.

There are other ways your implementation may have failed, relating to the difficulty in adapting workflows or IT infrastructure but the seven above are the most common.

It is also worth bearing in mind that that there may be problems with your AI implementation that are not obvious right now but which may become apparent later. Organisations have got into trouble from ‘vanity metrics’ (declaring the success of their AI implementation on very narrow grounds), creating ‘technical debt’ (the success of the AI implementation is built on extra work elsewhere or downstream), lack of built-in governance or compliance leading to future problems or employees using ‘shadow AI’ (unapproved AI tools) because official programs haven’t delivered what they want.

There are ways of mitigating all of these problems. Please get in touch HERE for our crib sheets on the individual problems or CONTACT US for a discussion of how we might be able to help with your specific needs.

We’ll also be creating a series of industry-specific notes so please PUT YOURSELF ON OUR WAITING LIST for these papers.