What UK SMEs Get Wrong About AI ROI

Ask an SME owner how much time their new AI tool has saved and you'll usually get a guess dressed up as a fact. Something like "loads" or "a few hours a week, probably." That answer tells you the business never captured a baseline. If you don't know how long the invoice run took before AI, you can't say how much faster it is now. The same applies to customer response times, quote turnaround, or the hours a marketing team spent drafting content.

I've run technology businesses for over 30 years, and the lesson repeats in every sector: measurement starts before the project does, not after. Pick the one process you're changing. Time it, cost it, and count the errors in it, for two to four weeks, before you switch anything on. If your finance team takes three days to close a set of monthly accounts, write that number down. If your sales team takes six days to turn round a proposal, write that number down too. Without a starting point, every claim about AI's return is an opinion dressed as a result.

The dashboards that come free with most AI tools measure activity. Logins, queries sent, documents processed. None of that tells you whether the business is better off. A team can generate hundreds of AI-drafted emails a week and still take the same number of days to close a deal. Activity is easy to track and easy to mistake for progress.

Outcome metrics ask a harder, more useful question: has the thing you actually care about changed? Has invoice processing time dropped from four days to one? Has the error rate on data entry fallen? Has a member of staff been freed up to do work that generates revenue instead of retyping figures? If your reporting only shows usage, you are measuring adoption, not return.

AI dropped into an unchanged process rarely pays for itself. If a member of staff still checks every AI output line by line, the way they checked their own work before, you've added a step, not removed one. The return comes from redesigning the workflow: who reviews what, what gets automated outright, and what still needs a human decision.

This is the part most SMEs skip, because it takes a conversation with the team doing the work, not just a subscription to a tool. It's also the part that makes the biggest difference to the numbers. A council I worked with expected an AI drafting tool to speed up correspondence. It didn't, until they removed a duplicate approval step that had nothing to do with the AI itself. The tool got the credit. The workflow change did the work.

Complicated ROI models fail in SMEs for the same reason forty-tab budget spreadsheets fail: nobody keeps updating them. Keep it to three numbers, tracked monthly, against your baseline.

Before your next AI purchase, or before you judge the one you've already made, write down the baseline number for the process you're changing. Not a rough guess. An actual measurement, taken this week.

Then agree, in one sentence, what success looks like. Not "AI adoption" or "efficiency gains." A number: two days off invoice processing, a 20% drop in quote errors, three hours a week returned to a named member of staff. Review the three numbers above every month for the first two quarters. If the tool isn't moving them, fix the workflow around it before you spend more money fixing or replacing the tool.