Research
The real cost of AI isn't the budget. It's the value gap.
Most AI money isn't wasted on the wrong tools. It leaks in the space between running a pilot and realizing a return. Here is where to look, and how to size your own gap.
AI spend is easy to approve because it is visible. There is a licence, an invoice, a line item, a vendor to call. The waste is the opposite. It never appears on a bill. It sits in the distance between what a company paid for and what it can actually show for it, and nobody sends a statement for that.
That distance has a name. It is the value gap, and for most companies it is now larger than any tool cost on the ledger.
The gap is bigger than the tool bill
The research keeps arriving at the same place from different directions. MIT's The GenAI Divide: State of AI in Business 2025 reviewed hundreds of deployments and found that, despite $30 to $40 billion of enterprise investment, around 95% of generative AI pilots produced no measurable impact on the profit and loss statement. Only about 5% reached production with real value.
McKinsey's State of AI survey finds most organisations still stuck in experimentation, with only about a third beginning to scale. And the UK government's AI adoption research found something quietly damning: many businesses reported better workforce productivity from AI while seeing no change in revenue.
Put plainly: adoption is widespread, and return is rare. The money going into tools is real and mostly modest. The money going nowhere is larger, harder to see, and rarely measured. Which raises the useful question. If your AI budget is not really the problem, where is the leak?
Where the money actually leaks
The waste is not one big hole. It is five small ones, and most companies have all five open at once.
01
Pilot purgatory
A pilot proves the technology can do something, then stalls before production. It was never wired into a real process, so it dies quietly once the novelty fades. The cost is not just the pilot. It is the quarter of attention it absorbed and the decision it postponed.
02
Efficiency booked as value
A tool saves people time, so it gets counted as a win. But time saved is not return realized. Unless it becomes throughput, better decisions, retained customers, or margin, it stays an anecdote. This leak hides best, because it feels like progress every week.
03
No owner
The initiative sits with an innovation team or with IT by default, not with the person accountable for the workflow's numbers. So the loop never closes. Nobody is on the hook to turn usage into a result, and nobody notices when it does not.
04
No definition of success
The project launched without a threshold that would count as a win. When you cannot say in advance what proof looks like, a demo starts to feel like a result, and a company scales things it never confirmed were working.
05
Built beside the work, not inside it
People have to leave what they are doing to use the tool, so usage decays and the value with it. MIT's data is blunt: pilots that survive integrate into an actual workflow and improve as they run. The ones that fail sit beside the work and stay static.
None of these is a technology failure. Each is a decision that was skipped. We wrote about the shape of those decisions in Most companies don't have an AI strategy. They have activity. The leaks above are what it costs to skip them.
How to size your own gap
You do not need a consultant to estimate this. You need an honest afternoon. Run four numbers.
Now look at the overlap of the last three: initiatives that are owned, measured, and live. That small fraction is the part of your spend doing real work. The rest is your value gap. For most companies the first time they run this, the honest answer is uncomfortable, and it has almost nothing to do with which tools they chose.
This is why the fix is rarely "spend less" or "spend more." Both miss the mechanism. The waste is not in the budget. It is in the decisions wrapped around the budget.
Closing the gap
Closing it is unglamorous and specific. Pick one initiative that matters. Give it an owner who already carries the number it should move. Define, before you go further, what result would justify scaling and what would justify stopping. Then wire it into the actual workflow rather than beside it.
Do that, and the same spend starts converting. Skip it, and you can double the budget and widen the gap.
This is also where AI stops being a technology question and becomes an organisational one, which is the harder and more important problem. We take that up in AI ROI is an organisational problem. The model was never the bottleneck.
The budget was never the expensive part. The gap is.
Size the gap before you approve the next licence.
Our Discovery work does exactly this: names the workflow, the owner, and the threshold, and tells you plainly whether there is a case to build.