The AI Adoption Problem Isn't About AI

Most AI initiatives fail for a reason that has nothing to do with technology.

The people who approve AI investments — executives, boards, operating partners — are almost never the people who would have to change their behavior for AI to actually work. They approve the budget. They set the ambition. They do not update the CRM. They do not change how they run meetings. They do not document the processes that would need to exist for automation to touch them.

The people who would have to change are several layers down. They have different incentives. Nobody asked them whether the foundation is ready. And by the time the pilot rolls out, they've watched enough previous technology initiatives fail that their default posture is to absorb the new tool without changing anything — which is exactly what happens.

This is a capital allocation problem wearing a technology costume.

A leadership team I worked with had made AI a stated priority for two consecutive planning cycles. They had approved three pilots. What they had not approved was a CRM cleanup project sitting in the queue for eighteen months, a documentation initiative that would have made their core processes legible to automation, or any change to the email-and-meeting operating model that their actual work ran on. Each of those felt slower and less exciting than a pilot. The result was that every pilot produced results that couldn't be replicated at scale, because the conditions that made the pilot work didn't exist in the broader organization.

The tell is in the approval patterns. Companies serious about AI transformation eventually approve the boring investments — data governance, process documentation, communication habits — at the same level of commitment as the visible ones. Until that happens, the strategy is aspiration with a budget line.

The labeling problem makes this worse. When everything gets called AI — rule-based anomaly detection, recommendation engines, large language models, a slightly smarter dropdown — organizations lose the ability to sequence their investments correctly. A rules-based system that flags invoice exceptions has been sitting in most ERPs for a decade. It requires clean data and someone to configure the thresholds. It doesn't require a transformation initiative. An LLM that synthesizes unstructured customer feedback requires clean data, documented processes, and a workforce willing to act on outputs rather than ignore them. These are fundamentally different interventions. Calling them both AI in the planning conversation means they get evaluated on the same terms — and the visible, exciting one usually wins the budget, while the one that would actually work goes unconfigured.

The technology works. The organizational conditions for the technology to work are the hard part. And the people who fund the technology are rarely the ones responsible for building those conditions.