Contents
At a glance
- Category
- Business and buying
- Formula
- (Value produced minus total cost) divided by total cost, over a stated period
- Strongest evidence
- Traceable revenue, then costs that genuinely stop being paid
- Weakest evidence
- Time saved, unless the time is demonstrably redeployed
- Most under-counted cost
- Maintenance - it is never zero
- Rank candidates by
- Absolute value, not percentage return
What ROI means for an automation
Return on investment is net gain divided by cost. For an automation, the gain is the value it produced - time saved, jobs won, errors avoided - and the cost is everything in total cost of ownership, not just the build. Expressed as a percentage over a stated period, it is the number that decides whether the second automation gets approved.
The reason automation ROI cases have a poor reputation is that most of them are written before the fact by whoever wants the project approved. The version that is worth doing is measured afterwards against a baseline recorded beforehand.
Building an honest case
Four categories of gain, in descending order of defensibility. Revenue you can trace: jobs booked that would not have been, measurable when you can attribute them - the strongest evidence there is. Cost you actually stop paying: an overtime line that ends, a contractor no longer needed. Real, but only if the money genuinely stops leaving.
Time saved: minutes per run times runs per month times a loaded hourly rate. This is the most quoted and the weakest, because saved time is only worth money if it is redeployed to something valuable - twenty minutes a day returned to someone who was not fully occupied is worth approximately nothing. Say so. Errors avoided: quantifiable where you know the cost of the error, such as a missed appointment or a double booking.
The costs to count
Build cost, whether paid or internal time. Platform subscriptions in whichever billing unit applies. Model tokens. Connected service subscriptions. And maintenance, which is the line most often set to zero and is never zero - APIs change, credentials expire, requirements move. Budget a realistic share of someone's time per month per automation.
Also count the cost of getting it wrong: what happens if the automation misfires, and what it costs to check its output. An automation requiring an hour of daily review has that hour in its cost base.
Measuring rather than projecting
The method that works is boring. Pick two or three metrics before you build - response time, no-show rate, cost per booked job, hours spent on the task. Measure them for two to four weeks. Build. Measure the same metrics for the same period afterwards. Report the difference, including anything that got worse.
Two honest caveats to include: other things changed during the period, so attribution is imperfect, and seasonality may explain part of the movement. Stating those makes the rest of the case more credible, not less.
What you should know
Be sceptical of anyone quoting a standard ROI figure for AI automation, including vendors and including consultants. The variance between businesses is enormous, the published figures are almost all supplier-produced, and the number that matters is yours. Our ROI calculator is a modelling tool for your own inputs rather than a source of benchmarks, and what AI automation costs gives realistic ranges for the cost side.
Finally, size matters more than percentage. A 400% return on a process worth £50 a month is a rounding error; a 60% return on your largest recurring cost is a real result. Rank candidate automations by absolute value, not by ratio - which is exactly what the free automation audit is for.
ROI FAQ
How do I calculate ROI on an automation?
Take the value it produced - traceable revenue, costs that genuinely stopped, time redeployed, errors avoided - subtract everything it cost to build and run including maintenance, and divide by that cost. State the period, because ROI without a timeframe is meaningless.
Is time saved real ROI?
Only if the time is redeployed to something valuable. Twenty minutes a day returned to someone who was not fully occupied is worth close to nothing in cash terms. It is still worth reporting, but label it honestly as capacity rather than as savings.
What is a typical ROI for AI automation?
There is no useful typical figure. Variance between businesses is enormous and nearly all published numbers come from suppliers. Measure two or three metrics for a few weeks before you build, then measure the same ones afterwards - your own number is the only defensible one.
What costs do people forget?
Maintenance above all, which is never zero because APIs change, credentials expire and requirements move. After that: the platform's billing unit at real volume, model tokens, and the human time spent checking output on any automation that needs review.
How long should I measure before and after?
Two to four weeks each side for a busy business, longer if your volume is lumpy or seasonal. Use the same period length and the same metrics both times, and note explicitly what else changed - imperfect attribution stated openly is more credible than a clean number nobody believes.
Should I pick automations by highest ROI percentage?
No, by absolute value. A 400% return on a process worth fifty pounds a month is a rounding error, while a 60% return on your largest recurring cost is a real result. Percentage returns favour trivial processes because their denominators are small.
Primary source
Harvard Business Review - research on measuring lead-response returns
Checked August 2026. Anything priced or rate-limited moves, so verify against the source before you budget.
Want this built rather than explained?
Knowing what roi means is one thing; having it running in your business is another. We scope one workflow at a time, quote a fixed price in writing after a free consult, build it, and hand it over with documentation you own.
