The ROI of AI automation: how to calculate it honestly
Every AI vendor has an ROI calculator, and almost all of them multiply an inflated time saving by a fully loaded salary to produce an implausible number. Finance directors have learned to discount these on sight.
A credible business case is built differently. It separates hard savings from soft ones, counts revenue as well as cost, and states its assumptions explicitly so they can be challenged. This article gives you that framework.
Four sources of value, in order of defensibility
Not all value is equally provable. Order your case so the hardest numbers come first and the softest are presented as upside rather than as the core justification.
- Revenue recovered — enquiries answered that previously went unanswered, missed calls captured, quotes delivered before a competitor's. Directly measurable and usually the largest line.
- Direct cost avoided — outsourced overflow support, temporary staff at peak, planned hires deferred. Verifiable against invoices.
- Error cost avoided — duplicate payments, mis-keyed orders, missed contract renewals. Historic incident data makes this defensible.
- Hours reclaimed — real but the softest, because saved time only becomes money if it is redeployed to something valuable.
Counting hours without exaggerating
Measure the current state before you automate. Sample a fortnight of the process and record volume and average handling time per case, splitting routine cases from complex ones. Guessing at this stage invalidates everything downstream.
Then apply an automation rate per case type rather than a blanket figure. A realistic model might automate eighty per cent of routine cases fully, reduce handling time on another ten per cent by half through drafting, and leave the remainder untouched.
Value the saved hours at the loaded cost of the people actually doing the work, not at an executive rate. And state plainly what the reclaimed time will be used for — if the answer is 'nothing specific', discount it heavily in the headline number.
The revenue side people forget
Most businesses do not know their answer rate. Pull the data: how many inbound calls went unanswered last quarter, how many web enquiries waited more than four hours for a reply, how many quote requests were never followed up.
Then apply your own conversion rate and average order value to that lost volume. For service businesses with phone-driven demand, this line item routinely exceeds the entire cost of the project, which is why voice automation pays back fastest.
Speed matters independently of coverage. Response time is strongly correlated with conversion in most B2B contexts, so moving from a same-day reply to a two-minute reply is worth modelling as an uplift on existing volume, not only on missed volume.
The full cost side
Include everything, or your credibility evaporates at the first question. That means the one-off build cost, the recurring platform and usage fee, integration work on your side, internal time for the workflow audit and knowledge preparation, review effort during the shadow period, and ongoing maintenance.
Internal time is the line vendors omit and clients feel. Budget realistically for the process owner's involvement during weeks one through eight; underestimating it is the most common cause of a delayed go-live.
A worked example
A thirty-person services business handles roughly 1,200 support emails and 600 calls a month. Support consumes about 2.5 full-time equivalents at a loaded cost of £45,000 each. Around fifteen per cent of calls go unanswered, average order value is £900, and the enquiry-to-sale conversion rate is twenty per cent.
Automating sixty per cent of email and answering every call recovers approximately 1.4 FTE of capacity — around £63,000 a year — and captures roughly ninety previously missed calls a month. Even assuming only a third of those were genuine sales enquiries, the recovered revenue is substantial and typically dwarfs the cost line.
Against a build in the low tens of thousands plus a monthly fee, payback lands well inside the first year, with the revenue component doing most of the work. Present it that way: cost saving funds the project, revenue recovery justifies expanding it.
How to present it to a sceptical board
Give three scenarios — conservative, expected and optimistic — with the automation rate as the variable, and make the conservative case stand on its own. If the project only works in the optimistic scenario, it is not a project, it is a bet.
State the payback period and the point of no return: at what stage could you stop and what would it have cost. Phased delivery with a proven first process makes this answer comfortable.
Finally, commit to measurement. Agree the baseline metrics before go-live and report against them monthly. Nothing kills the second phase faster than being unable to prove the first one worked.