How to calculate automation ROI honestly
Most automations that fail the arithmetic fail on the same thing: the task was not frequent enough to matter.
The short answer
Automation ROI is the annual labor cost saved minus the annual running cost of the automation, measured against what it cost to build. The frequency of the task dominates the result, and it is the variable people check last.
The formula
annual hours = task hours x times per week x working weeks
annual labor cost = annual hours x loaded hourly cost
annual running cost = API and tooling + maintenance hours x rate
annual saving = annual labor cost - annual running cost
payback years = build cost / annual saving
Loaded hourly cost, not salary divided by 2,080. Automating an hour of a person's day saves what that hour actually costs the business, which is considerably more than their wage.
Worked example
A task taking 45 minutes a day, five days a week, 46 weeks a year, done by someone costing $34 an hour loaded. Building it takes 40 hours at $60, the API costs $388 a month and it needs 6 hours of maintenance a year:
annual hours = 0.75 x 5 x 46 = 172.5 → 5,865 of labor
build = 40 x 60 = 2,400
running = 388 x 12 + 6 x 60 = 5,016
annual saving = 5,865 - 5,016 = 849
payback = 2,400 / 849 = 2.83 years
same automation on a 2 hour daily task: 460 hrs = 15,640
saving = 10,624 → payback = 0.23 years
Identical build, identical running cost. At 45 minutes a day it takes almost three years to repay and is not worth doing. At two hours a day it repays in under three months.
The costs that turn a good ROI into a bad one
Maintenance. Automations break when an upstream format changes, a model is deprecated or an edge case appears, and the fixing lands on someone. Six hours a year is optimistic for anything touching a third party API.
Verification. If a human still checks the output, you have not removed the task, you have changed it. An automation that saves 45 minutes of doing and adds 15 minutes of checking saves 30, and the model is usually built as though it saves 45.
The failure cases. Work that the automation cannot handle still has to be done, and it is now done by someone who has lost the routine of doing it.
Why saved time is not saved money
Freeing 172 hours a year across one person does not reduce a payroll. It gives that person 45 minutes a day, which is only a saving if the time goes to something that earns or if it prevents a hire.
That is worth saying plainly because it is where automation business cases quietly become fiction. The honest version names what the freed time is for, and if the answer is nothing in particular, the saving is a comfort rather than a return.
What makes a task worth automating
High frequency, stable format, and a tolerable failure mode. All three, not any one. A daily task in a rigid format where a mistake is caught immediately is the ideal case.
A quarterly task is almost never worth automating no matter how tedious it is, because four repetitions a year cannot repay a build. The tedium argues for it and the arithmetic does not, and the tedium usually wins the argument.
What this leaves out
- Build cost is a real estimate at a real rate, including the time spent specifying it, not just writing it.
- Assumes the automation works at the accuracy required. An automation that needs full human review has a very different saving.
- Excludes the value of consistency and speed, which are real benefits that do not appear as hours saved.
Common questions
- What payback period is acceptable for an automation?
- Under a year is comfortable, one to two years is a judgement call, and beyond two the tooling will likely have changed before it repays. Short paybacks also protect you from the risk that the process itself gets redesigned.
- Should I count my own time if I build it myself?
- Yes, at a rate you would accept for other work. Building it yourself changes who pays, not whether it costs. Ignoring your own time is how projects that were never worth doing get built.
- Why did my automation not save the time it promised?
- Usually a verification step nobody counted, or the task was less frequent than assumed. Measuring the real frequency for two weeks before building is the cheapest possible way to find out.
Spreadsheets that do this
The formulas above, already built and checked — so you fill in your numbers rather than the arithmetic.
Related guides
Last reviewed 22 August 2026