Cold Cases : Reviving Dormant AI Pilots

Each police force employs dedicated units to deal with cold cases. They do not work on active cases but re-examine cases that remain unsolved and in which available evidence does not provide any clear answers. Moreover, the passing of time might have blurred the facts rather than brought more clarity.

I am starting to view dormant AI pilots as cold cases in the same manner.

In recent years, a plethora of AI pilots, proofs of concept, and other experiments have been carried out by various organisations. Some made it into production and became an integrated part of daily activities, while others just vanished. They weren't failures; they were lost in the organisational flow and stopped receiving priority attention until their memories faded completely.

Of interest here is the ability of many organisations to provide information on the number of pilots they've begun, yet relatively few can explain why particular pilots were shelved. In almost all cases, the responses seem all too similar. The technology wasn't there. There was user resistance to the solution. The data wasn't good enough. It became too costly. Priorities shifted. These explanations are likely true in many instances. But in other instances, they are mere assumptions accepted as truths.

What makes these scenarios even more intriguing is the similarity between the reasons AI pilots end up as cold cases and those of some investigations that linger unresolved for many years.

In police work, an investigation becomes cold due to the disappearance of witnesses, reassignment of resources, shifting priorities, or insufficient evidence to yield a conclusive result.

AI organisation. Lots go along similar lines. An executive sponsor leaves the organisation. Technical results are encouraging, but users never fully engage with the solution. After the pilot phase, data quality issues arise. Integration issues are growing and going beyond what was expected. Governance concerns appear. Budgets tighten. Strategic priorities shift. The technology itself is seldom the real issue. Organisational conditions in the surrounding environment evolve more rapidly than a pilot can adapt.

Documentation becomes outdated, team members leave, and gradually, evidence is replaced with guesswork. The second eventually gathers dust in the organisational equivalent of an evidence room, where it may hide a real failure or a lost opportunity.

The question is whether that dust hides a real failure or a lost opportunity.

Enter the cold case perspective.

Rather than starting with conclusions, it is more insightful to open the document and try to figure out what really happened. This implies an understanding of the initial situation, an analysis of the audience, an assessment of what remains today, and an examination of dependencies, limitations, and data related to the project. It also means comparing projected benefits with reality, identifying the moment when momentum was lost, and assessing the current environment and its ability to yield a different result.

The idea is not to argue whether the pilot was a success or failure. In most cases, such a question is too superficial. An operationally successful pilot can be adopted only on paper. A pilot that had commercial justification can come too early for the organisation to absorb it properly. Alternatively, projects that seemed to have all the potential back then can turn out to have serious vulnerabilities upon thorough assessment.

What makes the conversation relevant today is the change of circumstances. Technology progressed, governance methods improved, and costs were reduced.

• How many of these AI projects have been shelved, put on ice, or quietly dropped without any kind of formal debrief on what really happened?

• How many failures were really missteps in timing, adoption, governance, ownership, or organisational readiness?

• And the final, potentially most fascinating of the lot: how many of these would turn out differently today if reopened?

Cold cases are reviewed by police forces when circumstances change, new information surfaces, or questions still linger. It is worth applying the same approach to your portfolio of AI projects that have stopped, at least in some cases. You might discover that your organisation was right to close the case the first time around, but other times, you might find something that was overlooked previously.

Do you have any cold cases lying around your AI portfolio? Maybe now's the time to take a look at them.

Let's talk about it.