AI ROI cannot be measured: myth or reality?

"We can see AI is doing something. We just cannot prove it is worth the money."

That is the honest position most Managing Directors and Finance Managers are in right now. It is not an unreasonable place to be. AI vendors promise productivity gains. Marketing case studies show impressive numbers. But when a Finance Manager asks what the business actually got back for its investment, the honest answer is often a shrug.

Research covering Irish and UK SMEs puts a number on this discomfort. Three quarters of businesses believe AI will matter to long-term growth. Yet only 6% say they can measure AI return on investment confidently. That gap between belief and measurability is not a failure of ambition. It is a failure of method.

The myth this article addresses is not that AI cannot deliver value. It is that ROI measurement is somehow optional, or that it can be done retrospectively once the tool is already embedded. Both assumptions are expensive. This article is for Managing Directors, Operations Managers, and Finance Managers who want to move from "it seems to be helping" to a figure they can defend in a board meeting.

Why the measurement gap exists

The belief that AI ROI cannot be measured has a reasonable origin. Most SMEs encounter AI through a free trial, a team member's personal account, or a subscription that comes bundled with existing software. There is no formal project. There is no baseline. There is no defined outcome. Within a few weeks, the tool is in use across several functions, and no one recorded what the process looked like before.

When leadership then asks whether the investment is paying off, the honest answer is that there is nothing to compare against. That is not evidence that AI has no ROI. It is evidence that the measurement infrastructure was never built.

NatWest research found that 34% of UK SMEs cite uncertainty around ROI as a barrier to further AI adoption. A separate finding shows 40% identify AI skills or capability as a barrier. These two numbers are connected. Teams that lack the skills to design an AI pilot properly also lack the skills to measure one.

The result is a credibility problem that stalls investment. Leadership sees the tool in use but cannot quantify what it is replacing. Finance cannot approve further spend without a number. Operations cannot escalate a success they cannot document.

What the evidence actually shows

The measurement problem is solvable. The academic and practitioner evidence points consistently to the same approach: define the process cost before you deploy, then measure the same process after.

The first step is selecting one operational process with a visible, repeatable cost. The cost might be measured in hours per week, error rate, turnaround time, or headcount hours per output unit. The metric needs to be something an Operations Manager can record without a specialist tool. If it cannot be measured before deployment, it cannot be measured after.

The second step is running a controlled pilot rather than a broad rollout. Research from the Economic and Social Research Council points to management structure as a significant factor in whether AI adoption produces measurable gains. Organisations that scope AI projects deliberately, with clear objectives and defined timeframes, are better positioned to isolate what changed and why. A broad rollout with no control group makes attribution almost impossible.

The third step, and the one most SME pilots miss, is including the full cost in the calculation. KPMG's September 2026 research noted a shift in UK business focus toward AI governance as tools move into core operations. That shift reflects a growing recognition that AI output requires human review, correction, and oversight. A Finance Manager calculating ROI needs to include review time, any remediation when the output is wrong, data handling costs, and the cost of training team members to use the tool competently. Licence fees and raw time saved are not the whole picture.

When those costs are included honestly, some AI pilots will fail the test. That is not a problem. That is the measurement working as intended.

A realistic approach for SME finance and operations teams

The commercially rational path from curiosity to investment is a time-limited pilot on a single process, with a go or no-go gate at the end.

For a Finance Manager, a practical pilot might look like this. Choose one document-heavy process: supplier invoice processing, monthly management account preparation, or contract review. In an accountancy or law practice, client onboarding and engagement letters are strong candidates. Record how long it currently takes, who is involved, and what the error rate is. Run the AI-assisted version for six to eight weeks. At the end, record the same metrics. Include the time spent prompting, reviewing output, and correcting errors. Compare the two.

For an Operations Manager, the same logic applies to any repeatable process with a measurable output: response drafting, data entry reconciliation, shift reporting. The process itself matters less than having a clean before-and-after comparison.

For a Managing Director, the useful question to ask before any pilot starts is: "What would need to be true for this to have paid back in six months?" If no one can answer that question, the pilot is not ready to run.

One honest caveat is necessary here. Some productivity gains from AI are genuinely hard to quantify in the short term. Team members who spend less time on repetitive tasks may produce better quality work, catch more errors elsewhere, or simply be less fatigued. Those benefits are real, but they do not show up in a six-week time-and-motion comparison. A rigorous pilot captures what it can capture and is honest about what it cannot.

gecco's Training and consultancy programme gives SME teams the practical grounding to scope AI projects properly before committing budget. The ESSENTIALS tier covers AI fundamentals, the GRAFT methodology and hands-on assistant building across three days. By building internal competence first, teams can set baselines, define success criteria and evaluate whether a pilot is genuinely saving time, rather than simply producing more output that still needs review.

What to include in the ROI calculation

This section is for Finance Managers who want a working framework rather than a principle.

Costs to include on the investment side: software licence or subscription cost, internal time spent on setup and configuration, time spent training team members, ongoing review and quality-checking time per output, remediation time when output is wrong, and any data handling, storage, or compliance costs.

Benefits to measure on the return side: reduction in hours per unit of output, reduction in error rate or rework, reduction in turnaround time, and any reduction in third-party spend the AI replaces.

Regulatory and compliance costs deserve particular attention. GDPR obligations do not disappear because a task is now AI-assisted. If the process involves personal data, the cost of compliant data handling, human review requirements, and potential remediation if something goes wrong belongs in the calculation. Omitting these costs does not make the ROI look better in the long run. It makes it look better until something goes wrong.

The point of including all costs is not to make AI look bad. It is to make the comparison honest. A pilot that shows a genuine net benefit after full-cost accounting is a business case that a Finance Manager can take to a board. A pilot based on partial costs is a business case waiting to collapse under scrutiny.

Next step

If you are a Managing Director, Operations Manager, or Finance Manager trying to move from "we think AI is helping" to a number you can defend, the AI Readiness survey is designed to surface exactly where your measurement gaps are.

Take the AI Readiness survey. You will get access to 65+ free resources and a custom AI Readiness report. We then offer a free 30-minute showcase to walk through your results.

gecco's Quick-Start programme is built for SMEs that want to scope and measure AI properly before committing to a wider rollout. If you want structured support to build that competence in-house, that is what we do.

Website · LinkedIn · Case Studies · Newsletter