Almost every leader I meet can tell me what their AI rollout cost to buy. Far fewer can tell me what it is costing them to leave half-finished. Those are different questions, and the second one is the expensive one. A stall rarely announces itself. It hides inside budgets that were always going to be spent and reports that were always going to read green.
Why the number stays hidden
The trouble with a stalled rollout is that nothing obviously breaks. The licences are paid for. The project is “on track.” The steering committee still meets. None of that shows up as a loss, because none of it is a line anyone would flag in a board pack.
What is actually happening sits underneath the reporting. The capacity you were promised never arrives. The decisions still move at the old pace. The work-arounds multiply quietly. The cost of all of that is real, but it is spread across a dozen places that each look individually fine.
So the first job is not to fix anything. It is to make the invisible cost visible, in hard money terms, so you can decide whether it is worth chasing.
Where the money actually leaks
In my experience the leak is rarely one big hole. It is several small ones, and they tend to fall into the same few buckets.
There is spend with nothing behind it: seats bought and barely used, tools paid for and quietly bypassed. There is capacity that never materialised: the efficiency gain that justified the investment, still sitting in the business case rather than the P&L.
There is the cost of running two ways of working at once, because nobody quite trusts the new one enough to switch off the old. And there is the slowest, hardest cost to see: the opportunity that goes to a competitor who moved while you were still getting ready.
None of these is dramatic on its own. Added up, they are usually the difference between a rollout that pays for itself and one that becomes a permanent line of drag.
How to start counting it
You do not need a forensic audit to get a useful first number. You need to look at the spend against the capacity it was supposed to buy, and be honest about the gap.
Start with what you committed. Then ask what you can actually point to as a result, in money or in freed time. The distance between those two figures is the working estimate of what the stall is costing you, and it is almost always larger than the team assumes.
The point of the exercise is not precision down to the last digit. It is to turn a vague sense that “AI isn’t quite landing” into a clear diagnosis you can take to the board to demonstrate you understand the underlying issues and exactly how to fix them.
The first move
Once the cost is a number rather than a feeling, the conversation changes. You can decide what is worth fixing first, instead of treating the whole thing as one problem.
If you want a quick read on where your own rollout is leaking, that is what the Friction Scan is for. Four questions, your biggest problem named on the spot, and a straight answer either way.
