Insights — AI Workflow & Commercial Automation — 4 min read
How to Measure ROI from AI Automation (With a Worked Example)
Knowing what automation costs is only half the picture. Measuring what it actually returns is where most businesses lose track.
In short
ROI from AI automation is measured by comparing the total benefit (time saved, error reduction, faster response, additional capacity or revenue protected) against total cost (implementation, licensing and ongoing maintenance) over a defined period, then expressed as a percentage or payback period. Some benefits are measurable directly from records; others, like revenue impact, can only be reasonably estimated and should be labelled as such.
Cost is usually the easier number to find: a quote, an invoice, a monthly subscription. Return is harder, because it is spread across time saved, fewer errors, faster responses and sometimes revenue that would otherwise have been lost.
Measuring ROI properly means being honest about which of these benefits can be measured directly, which can only be estimated, and being clear about the difference when reporting results internally.
Why ROI on automation is harder to measure than it sounds
Automation rarely produces a single clean number. A business might save admin time, reduce errors, respond to enquiries faster, and free up a person for higher-value work — all from one change. Some of these translate directly into measurable cost savings; others are softer and need a reasonable, clearly labelled estimate rather than a precise figure.
The building blocks of automation ROI
Time saved and capacity gained
The most measurable benefit. Compare time spent on a task before and after automation, multiplied by an hourly cost figure appropriate to who was doing it. Capacity gained — the ability to take on more work without hiring — is related but should be tracked separately, since saved time is only valuable if it is redirected to something useful.
Error reduction
Measured by comparing error or rework rates before and after, where records exist (returned invoices, corrected orders, complaint volumes). Where no baseline was recorded, this benefit should be estimated conservatively rather than invented.
Response speed and conversion
Faster response to enquiries can plausibly improve conversion rates, but the link is indirect and influenced by many other factors. This should be reported as a plausible contributing factor, not a guaranteed causal result, unless genuinely isolated through testing.
Revenue protected or gained
The hardest to attribute cleanly. Automation that prevents missed enquiries or follow-ups can protect revenue that would otherwise have been lost, but this should be estimated with caution and never presented as a guaranteed outcome.
Cost reduction
Direct reductions — fewer hours of overtime, reduced need for temporary staff, lower error-correction costs — are usually the most defensible figures to use in an ROI calculation.
The costs side of the equation
Implementation cost (build and setup), ongoing licensing or subscription cost for any third-party tools, and ongoing maintenance (monitoring, adjustment as processes or systems change) should all be included. Omitting maintenance cost is the most common way ROI calculations overstate the benefit.
A simple ROI formula
ROI (%) = ((Total benefit − Total cost) ÷ Total cost) × 100, measured over a defined period, typically 12 months. Payback period = Total cost ÷ Average monthly benefit, which tells you how many months until the investment is recovered.
Worked example
Suppose a business automates its enquiry response and basic quoting process. Before automation, one team member spends roughly 10 hours a week on manual data entry and chasing information, at an estimated cost of £18/hour including overheads (£180/week, or around £9,360/year). Implementation is costed at £5,500 (within Evans' from £4,950 + VAT implementation range) and ongoing managed support at £995/month (£11,940/year).
| Item | Illustrative figure |
|---|---|
| Time saved (annual, estimated) | £9,360 |
| Reduced quote errors (annual, estimated) | £1,200 |
| Total estimated annual benefit | £10,560 |
| Implementation cost (year 1) | £5,500 |
| Managed support (year 1) | £11,940 |
| Total year 1 cost | £17,440 |
| Year 1 ROI | ((£10,560 − £17,440) ÷ £17,440) × 100 ≈ −39% |
| Year 2 ROI (no implementation cost) | ((£10,560 − £11,940) ÷ £11,940) × 100 ≈ −12% |
This illustrative example shows a negative first-year return once implementation and ongoing support are both included honestly — which is a realistic outcome for some automation projects and a useful reminder that not every process justifies full managed support. A business in this position might reasonably move to lighter-touch maintenance after the first year, or automate a higher-value process where the time saving is larger relative to cost.
Why some figures should stay as estimates, not forecasts
Businesses sometimes want a single confident number to present internally. Resist collapsing revenue-related or conversion-related benefits into a hard figure unless they can genuinely be isolated and measured — a confident but unfounded number is worse than an honest range.
Risks and limitations
ROI calculations are only as good as the baseline data used. Businesses without clear records of time spent or error rates before automation will need to estimate a baseline carefully, ideally conservatively, rather than retrospectively inflating the 'before' picture to flatter the result.
How Evans approaches this
An AI Workflow Audit (£1,495 + VAT) establishes a realistic baseline before any automation is built, so that any ROI measured afterwards is compared against an honest starting point rather than an assumed one. Evans does not promise guaranteed savings or revenue outcomes, and will say plainly where a process is unlikely to justify the cost of automating it.
Want to see how this would work in your business?
AI Workflow & Commercial Automation looks at the commercial and operational work around your sales, customers and administration, and automates only what is worth automating — with people approving what matters.