How do I work out the ROI of automating a task?

The arithmetic is simple. The assumptions are where people go wrong.

The calculation

Annual hours saved = (minutes per task × times performed per year) ÷ 60

Hourly cost = annual salary ÷ annual working hours

Annual saving = hours saved × hourly cost

Our savings calculator runs this for you. It assumes 46 working weeks a year, which strips out holiday and bank holidays, and 1,725 annual working hours (46 weeks at 37.5 hours). A task done daily therefore happens 230 times a year, not 365. Using 52 weeks and 2,080 hours, as a lot of online calculators do, overstates the saving by roughly a fifth.

A worked example

A finance assistant on £45,000 spends 2 hours a day keying invoice data.

• 2 hours × 230 days = 460 hours a year
• £45,000 ÷ 1,725 = £26.09 an hour
• 460 × £26.09 = £12,000 a year

Against a £6,000 build, that is payback inside six months on salary alone. On those numbers, build it, and the case only strengthens once you count what those 460 hours get redeployed into.

Change one input and the picture flips. If the same person spends 15 minutes a day, the annual saving is about £1,500 and payback stretches to four years. Time per task and frequency are what make automation worth doing.

Three ways these numbers get inflated

Counting saved time as saved money. Freeing 460 hours does not reduce payroll unless you actually redeploy or reduce headcount. It is real value, but it is capacity, not cash. Be honest with yourself about which one you need.

Ignoring the residual. Almost no automation removes 100% of a task. Exceptions still need a human. Assume you remove 70% to 90% and model on that.

Forgetting running costs. Platform subscriptions, API usage and model calls are ongoing. So is maintenance when a system you depend on changes its API.

The value that does not fit the formula

Some of the strongest cases are not time savings at all. Faster follow-up wins deals that would otherwise go cold. Consistent reporting makes decisions better rather than cheaper. One client used AI to analyse thousands of customer accounts and surfaced revenue patterns nobody had the hours to find manually, which is value that no time-saved calculation would ever have predicted.

Quantify what you can, name what you cannot, and do not pretend the second category is zero.