Research
AI ROI is an organisational problem. The model was never the bottleneck.
The same tool produces real return in one company and nothing in the next. If the tool is identical, the difference is the organisation around it.
Two companies buy the same AI tool. Same model, same licence, same vendor onboarding. A year later, one can point to a shorter cycle time and a lower cost per case. The other has a wall of usage statistics and no movement on the profit and loss statement.
The tool did not decide that. The organisation around it did. This is the part of AI that gets discussed least, because it is the least fun to sell. The model is ready. The open question is whether the company is.
The evidence points at the organisation, not the technology
When AI initiatives fail to return, the instinct is to blame the model, the data, or the pace of the market. The research keeps pointing somewhere else.
Gartner found that 88% of HR leaders say their organisations have not realized significant business value from AI tools, even as adoption and employee enthusiasm run high. MIT's The GenAI Divide reached the same conclusion from the technical side: the pilots that fail do so not because of model quality but because of how they are wired into the business. And McKinsey's State of AI finds only a small minority of firms capturing meaningful earnings impact, while most remain stuck in piloting.
Three separate bodies of research, one shared finding. The bottleneck is not the technology. Modern models are more than capable. The value is getting trapped somewhere after the tool works and before the business benefits. That somewhere is the organisation.
Where the organisation decides the outcome
If the model is not the variable, what is? Three things, and none of them ships with the software.
01
Whether AI sits in the workflow or beside it
A tool that lives in a separate tab, one someone has to remember to open, produces a demo and little else. A tool wired into the actual flow, where its output triggers the next step, changes the economics. Where the tool sits is an organisational decision, not a technical one.
02
Whose number is supposed to move
When an initiative is owned by an innovation team or by IT, success gets measured in models deployed and accuracy achieved. Those are not business results. Value appears when the outcome sits with the person already accountable for the cost per ticket, the days to close, the conversion rate.
03
What happens to the time AI frees up
Gartner found only 7% of organisations give any guidance on how to use the time AI saves. The other 93% are paying for a benefit and then letting it evaporate. Time saved is not value until someone decides where the saved time goes.
We took up the ownership question in the piece on why most companies have activity rather than a strategy. Seen from the ROI side it is the same point: with no owner of the number, no one's job is to turn usage into money.
Why a better tool cannot fix this
Here is the uncomfortable part for anyone hoping to buy their way out. If the return depends on where AI sits, who owns the outcome, and how freed time gets spent, then a more capable model does not help. It only makes a tool that was sitting beside the work sit beside the work faster. The gap is not in the technology, so more technology does not close it.
Evolve work, not the workforce.
Gartner's phrasing for the fix is blunt and worth borrowing. The change that produces ROI is a change to how work flows, who owns which outcome, and how freed capacity gets used. That is organisational design. It is slower than a procurement cycle, harder to hand to a vendor, and it is the actual job.
It is also why the value gap we described earlier is so stubborn. Companies keep treating a structural problem as a shopping problem, and the two never meet.
What this means if you are deciding how to build
The practical read is almost reassuring. You do not need the newest model to get a return. You need to do three unglamorous things well: put AI inside one real workflow, give the outcome to the person who owns the relevant number, and decide in advance where the freed time will go. Get those right with a competent tool and the return follows. Get them wrong with the best tool on the market and it will not.
This also explains a finding worth sitting with. Initiatives built with an external partner tend to succeed more often than internal builds, and not because outsiders carry better models. It is because an outside team is forced to name the workflow, the owner, and the outcome out loud before anyone writes code. The discipline is the deliverable. It is the part we hold the line on in every engagement, and it is described in how we work.
AI ROI is not a technology problem. The technology is the part that already works. The return depends on the organisation the technology lands in: its workflows, its accountabilities, its willingness to redesign rather than bolt on. That is harder than buying a licence. It is also the only part a competitor cannot copy by purchasing the same thing you did.
Bad news for the shortcut, and good news for anyone willing to do the real work.
The model is ready. The question is whether the company is.
We work on both halves: the system, and the accountability around it that decides whether the system ever returns anything.