The push to prove AI's return on investment in 2026 is intensifying, but according to a new article by Pedro Varela, Head of AI at Slalom UK & Ireland, measuring ROI at the individual tool level is fundamentally misguided. Research suggests up to 70% of UK businesses are using or planning to use AI, and after last year's panic about an "AI bubble", the consensus has swung the other way: 2026, we're told, is the year AI ROI gets real. However, Varela argues that the fixation on license-by-license returns often masks a deeper failure: the original investment was made without a clearly defined problem.
The Danger of Tool-Level Metrics
According to Varela, the demand to demonstrate immediate ROI at the tool level is not always a sign of financial discipline. "Sometimes it's a sign that the original investment decision was made without a clear problem to solve. You can't calculate the return on an answer when nobody agreed on the question," he wrote. A leader pushing hard for tool-level ROI is often revealing a failure of definition rather than a failure of measurement. Varela argues that AI implementation is not like rolling out a new CRM; it "reorders how people make decisions, and it tends to expose whatever was already broken in those decisions."
| Frame | Example |
|---|---|
| Tool-first | "We need to deploy generative AI across customer service" |
| Problem-first | "Our agents spend 40% of every call searching three systems for policy information, and we want to cut that in half" |
The table above, derived from Varela's examples, contrasts two approaches. In the tool-first frame, the ROI question becomes unanswerable because the goal is deployment itself. In the problem-first frame, the return is built into the brief, and the AI either delivers or it doesn't.
From Tool-First to Problem-First
Varela emphasizes that most organizations brought AI tools in to innovate, solve specific problems, improve productivity, or keep up with competitors. Others were meeting employee demand for best-in-class tools. But the pressure to demonstrate immediate ROI, layered on top, causes the framing to break down. The better question, Varela suggests, isn't "what did this tool return?" but "has this investment created genuine value, broader than a figure on a page?"
Skipping the groundwork—the honest assessment of which problems are AI-shaped and which aren't—and then asking the spreadsheet to retrofit a justification leaves teams confused and leadership defensive. Varela warns that treating AI like any other piece of software, handed to staff because it might increase profits, not only fails to deliver results but builds friction and skepticism that make the next AI investment harder to justify.
Winning with AI in 2026
The businesses that win with AI in 2026, according to Varela, will not be the ones with the most sophisticated ROI dashboards. They will be the ones that started with a sharply defined problem and worked forward to a solution, rather than starting with a tool and working backwards to a justification. The difference shows up in how the conversation begins. Varela notes that task-level ROI is often the right place to start: if an AI system reduces the time to summarize a case, draft a response, check a contract, classify an incident, or retrieve policy information, that gain should be measured. But the overarching question must remain focused on the problem solved. For CTOs and digital leaders, the takeaway is clear: stop asking what the AI tool returned and start asking what business problem it was meant to solve.
Sources:
