Many impressive automations cost more than the task they replace. The calculation that reveals it takes five minutes, and almost nobody does it before starting.
Definition
Automation ROI is the ratio between what it saves you and what it costs you, counting building, subscriptions and supervision.
The full formula fits on one line: annual gain = (unit time × annual frequency) minus (build time + annual supervision time), compared against tool costs.
The term everyone forgets is supervision. An automation is never set and forgotten: it breaks, it needs fixing, it needs adapting when a tool changes. Count one to two hours a year per automation, more if it is critical.
A worked example
You manually enter orders received by email into your management tool. Three minutes per order, forty orders a month.
| Item | Calculation | Result |
|---|---|---|
| Current time | 3 min × 40 × 12 | 24 hours a year |
| Building | Half a day | 4 hours, once |
| Supervision | Fixing and adapting | 2 hours a year |
| Tools | Subscription and calls | About €300 a year |
First year: 24 hours saved against 6 hours invested, so 18 hours net. Following years: 22 hours a year. The calculation is favourable, and clearly so.
Redo the same table with a task done three times a year: 15 minutes saved annually against 4 hours of building. You will never break even, and that is exactly the kind of appealing project worth refusing.
Tool costs are converted into hours at €50 an hour, so everything comes back to one unit. Build time is almost always underestimated: double your first instinct.
What the hours calculation does not show
Reliability. Manual entry produces errors. If an error costs a customer or a credit note, the automation is worth more than the time it saves.
Timing. A task done continuously rather than weekly sometimes changes service quality, and that is not measured in hours.
Mental load. No longer having to think about it has real value, hard to quantify but you will feel it.
Conversely, the cost of checking. An automation whose every result must be reviewed saves almost nothing. That point sinks most projects built on an AI model without guardrails.
Beware the optimistic stopwatch. The build time you estimate is almost always for the normal case, without edge cases or trials. Double your initial estimate and you will be close to reality.
Frequently asked questions
At what threshold should you automate?
A simple rule: if the task comes back at least weekly and takes more than two minutes, the calculation is almost always favourable. Below that, do the maths honestly.
How do you count the cost of an AI model?
By estimating monthly consumption in Token then applying the rate. It is often a small share of the total against human time, but it can run away with a poorly capped AI agent.
Should learning the tool be counted?
On the first automation, yes, and it often makes that one unprofitable on its own. That learning then spreads across all the following ones, which is an argument for not stopping at the first.
How do you choose where to start?
By listing your repetitive tasks with their frequency and duration, then sorting. Our n8n course starts with that quantified inventory before any building.