How to measure the ROI of AI in a small business
If you can't tell whether AI is paying off, you'll either over-invest in hype or quit something that was working. Here's how to measure it without a finance team.
AI spend is easy to start and easy to lose track of. Without a simple way to tell whether it's paying off, you end up in one of two bad places: pouring money into things that sound impressive but change nothing, or killing something that was quietly working. You don't need a finance team to avoid both — you need a bit of honest arithmetic.
Count the whole cost, not just the subscription
The licence fee is the visible, and usually smallest, part. The real cost includes the hours spent setting things up, the time your team takes to learn the new way of working, and the ongoing effort of reviewing what the AI produces. Add those in. A £30-a-month tool that eats five hours of setup and an hour a week of checking is not a £30-a-month tool.
Value the benefit honestly
Benefits usually fall into three buckets: time saved, revenue gained, or cost avoided. The trap is "time saved" that doesn't actually go anywhere. If AI frees up two hours a week and those hours go into better work, more customers, or genuinely reclaimed capacity, they count. If they just evaporate, they don't.
Record a baseline before you start
This is the step almost everyone skips, and it's the one that makes the whole exercise trustworthy. Before you roll anything out, write down the current reality: how long the task takes today, how many enquiries you handle a week, what your response time is. Without that "before" picture, every later claim is a guess dressed up as a result.
Pick a sensible period
Judge simple, task-level uses over one to three months — the costs are low and the effect is quick. Bigger, custom builds deserve a longer horizon, because the upfront cost is front-loaded and the payoff accrues over time. Match the measurement window to the size of the bet.
Do the sum — and act on it
Put it together: value created, minus full cost, over your chosen period. If it's comfortably positive, expand. If it's marginal, look at whether better habits or a cheaper setup would tip it. If it's negative and you can't see a path to positive, stop — and count the lesson as cheap. The businesses that win with AI aren't the ones that spend the most; they're the ones who know which of their bets are actually working.
Frequently asked questions
- How do you measure ROI on AI?
- Compare the full cost of the AI (subscriptions, set-up time, and ongoing effort) against the value it creates (time saved, revenue gained, or cost avoided) over a set period. The discipline that makes it honest is recording a baseline before you start, so you're comparing against reality rather than a guess.
- What's a realistic ROI timeframe for AI in a small business?
- For simple, task-level uses — drafting, summarising, handling routine enquiries — you can often see a return within one to three months because the costs are low and the time savings are immediate. Larger custom builds take longer and should be judged over a longer horizon.
- What mistakes make AI ROI look better or worse than it is?
- The three big ones: ignoring the hidden costs (set-up, training, checking output), valuing 'time saved' that doesn't actually free up capacity or revenue, and having no baseline to compare against. Fix those and your numbers reflect reality.