Boards want numbers. After years of AI investment, CIOs and CTOs across the UK are being asked to prove the spend, and most are reaching for the same tool: licence counts, seat numbers, and usage dashboards. It is the wrong tool for the job. Research suggests up to 70% of UK businesses are either already using AI or they’re planning to. After the “AI bubble’ hit headlines last year, for some, the mood has swung sharply. 2026 is supposedly the year AI ROI gets real. But the fixation with measuring returns at the tool level, licence by licence, seat by seat, isn’t financial discipline. In some cases, it signals that the original investment had no clear problem to solve. You can't calculate the return on an answer when there was no agreed question in the first place.
In most instances, AI is brought in to innovate, solve a problem, enhance productivity, or stay ahead of the competition. Some employ AI to meet employee demand for best-in-class tools. Whatever the reasoning, AI quickly shot to prominence, and technology-focused leadership came under pressure to implement quickly to credibly claim 'AI Capabilities’. Then came pressure to show immediate ROI, and that's where the framing breaks down. The better question is "has this investment created genuine value, beyond just a figure on a page?"
If a business is unable to articulate why AI is needed beyond a vague notion of "staying ahead", the demand for immediate returns is doing the work that strategic intent should have done initially.
Ask the wrong question, and you’ll reveal the wrong strategy
When a leader pushes hard for tool-level ROI, they often reveal something they didn't intend to. Ask what the AI is doing inside the business, such as the decisions it changes or which workflows it reshapes, and the answer often thins out quickly. This isn’t a measurement failure, it’s a failure of definition.
AI implementation isn't the same as rolling out a new CRM. It reorders how people make decisions, and often exposes what was already broken in those decisions. The groundwork needed is an honest assessment of which problems are AI-shaped and which aren’t, skipping that and asking a spreadsheet to retrofit a justification leaves teams confused and leadership on the defensive.
If a business cannot define what ‘return’ means beyond financial determination, they will not see genuine returns at pace. Handing staff access to such a powerful tool simply because it
might increase profits doesn't just fail to deliver results - it compounds the friction and skepticism that make the next AI investment harder to justify.
Who will be the AI ‘winners’?
In short: the ones that ask the right questions. The businesses that win with AI in 2026 won't be the ones who have built the most sophisticated ROI dashboards. They'll be the ones that started with a carefully defined problem and, from here, worked toward a solution, rather than starting with a tool and working backwards to a reason.
The difference is rooted in how the conversation begins. The statement, "We need to deploy generative AI across customer service" is tool-first; ROI becomes unmeasurable because the goal from the start is the deployment itself. "Our agents spend 40% of every call searching three systems for policy information, and we want to cut this by 50%" is a problem-first frame; and here, the ROI is built into the brief. That makes it measurable. And when those agents did reduce search time by half, the ROI didn’t need calculating. It was already there from the start.
This is not to say that task-level ROI is irrelevant. In fact, it is often the right place to start. If an AI system is able to reduce the time it takes to summarise a case, draft a response, or check a contract, that gain should be measured. These task-level metrics help teams understand where the adoption is successful, whether the technology is genuinely useful, and which use cases need more investment. But they are still only a starting point. A faster task does not automatically result in better business outcomes if the workflow around it stays the same.
The real question to look at is whether the task-level gains accumulate into something much more meaningful: such as fewer handoffs, better customer experience, quicker decision-making, lower operational risk, or more capacity for higher-value work.
Layering ROI frameworks onto that first kind of decision rarely creates accountability. They only create the appearance of it, often after the fact, and once leadership senses the investment isn't landing, they reach for measurement as a shield. While defining the problem first doesn't guarantee a return, it is the only clear way to know what one would look like. The diagnostic is simple. Before any AI investment, a business should be able to complete three sentences: the problem we are solving is X, we will know it will be solved when Y, and the people most affected are Z. If a leadership team cannot complete those sentences without defaulting to technology, the investment case isn’t ready.
This is easier when an organisation controls the procurement decision. When a tool arrives via a vendor bundle or board mandate, the discipline shifts: not whether to define the problem first, but how quickly a team can build that definition around what’s already been committed to.
ROI compounds the human work
Even the most well-defined problem won't produce returns if the people using it most often aren't properly equipped. This is the part that gets unfunded and overlooked. The human side of the rollout includes capability building, time to experience, and workflow redesign. Teams that were involved in scoping an AI tool, defining what it should and shouldn’t do, tend to use it more selectively and more effectively than teams handed a finished product with a training session. That selective use is where the return lives.
When the entire conversation is anchored solely to financial returns, this human side looks like an overhead. But it isn't. In reality, it’s where the return compounds. Employees who understand what an AI system is good at (and what it's not) are able to make better decisions about employing it, when to override it, and when to escalate. That human judgement is the asset that sees real returns. Tool licenses aren't.
A workforce that has been included in the design of an AI rollout, rather than handed the output alone, is more likely to be receptive to the disruption that follows. Morale stays high. Adoption is real rather than performed. And the return that everyone was so anxious to measure starts to appear, often somewhere different to where the original spreadsheet was looking.
Track results, not subscriptions
The C-suite fixation on tool-level measurement has pulled attention away from the question that we should have started with: what is the problem, and for whom are we solving it?
Returns from AI are happening, but they show up at the level of capabilities and outcomes. Fewer handoffs, quicker decisions, better judgement at the edge of the business. Not at the level of individual tool subscriptions.
This question will become more pressing as businesses move from copilots to agents. With copilots, it is tempting to measure productivity task by task: how quickly a document was drafted, or someone searched for an answer. Those measures can be useful, but they’re limited.
In an agentic world, where AI systems can plan, act across tools, trigger workflows and only need to involve humans at specific decision points, those metrics quickly become inadequate. Now, the question will not be whether one task becomes 20% faster. Instead, it will be whether an entire process becomes more resilient, auditable, responsive or fundamentally different. If AI changes the unit of work, ROI measurement must change too.
In practice, this means shifting from measuring task speed to measuring process outcomes. Did the claims process complete with fewer human touchpoints? Did the compliance review catch more issues in less time? Did the customer get a resolution without being handed off three times? These are the questions that agentic ROI demands. They require different data, different baselines, and a different conversation between technology and finance teams than
the one most organisations are currently having. The point is not to stop measuring. It is to measure at the right level. Leaders who do that will find the returns they’re looking for.
Just not necessarily where everyone was expecting them.
