Is it worth automating? Automation ROI calculator
For anyone deciding whether a task is worth automating: estimate the hours it takes now, the hours a workflow could give back, and the most a build could cost and still break even.
Rule-basedInstant, rule-based result.
Check it yourself: every request this page makes
Every request this page has made since you opened it
The list fills in once the page has loaded.
Downloads bring the tool to you. Nothing has been sent. Your numbers are not in any of these.
Or open your browser's developer tools, choose Network, change a number and watch: your numbers never appear.
Your numbers
Your data. Runs in your browser. Your numbers stay here unless you press Send this estimate. How we handle data
Task 1
Stays in this browser. Never in the link or anything you send.
Money and upkeep
For all tasks. Leave the money fields empty to see hours only.
Optional. What an hour of the people doing it costs.
Named beside every amount. Never converted.
Work it out from a yearly salary
Salary × 1.3 ÷ 2,080. The 30% overhead is a rule of thumb. The salary is not kept.
Someone looks after every automation. Only for tasks above 0%.
Optional. Needed for payback.
Optional. Hosting, subscriptions, model calls.
Example result from made-up data
Example: Fernway, a made-up small software company. 3 people spend 12 minutes on a task that comes up 40 times a week; 50% of it could be automated, with 1 hour of upkeep a month, at 30 USD an hour. Build 3,000 USD, running 50 USD a month.
Worth a closer look.
About 51.0 hours back each month, 30.2 in the conservative case.
What this means
Most you could spend building it and still break even within 12 months: 10,272 USD. Within 3 years: 30,816 USD.
Your estimate from your own numbers, not a measurement or a quote.
| Hours a month now | 104.0 h |
|---|---|
| Hours back a month, expected | 51.0 h |
| Hours back a month, conservative | 30.2 h |
| Value a month | 1,530 USD |
| Value a year | 18,360 USD |
| Net a month, after running cost | 1,480 USD |
| Payback | 2.0 months3.5 months conservative |
| Full-time equivalent | 0.29of one person at 173.3 h a month |
Change any number to see your own.
01How it works
Arithmetic on your numbers, worked out on this device.
You enter one to five tasks: how many minutes each takes, how often it happens, how many people do it, and how much of it could be automated. The calculator turns that into hours a month now and hours a month back, first as expected, then as a conservative case with the automatable share 20 points lower.
Add a cost per hour and it shows the value per month and per year in the currency you pick, never converted. Add a build cost and a running cost and it shows the payback period, or says plainly that the task does not pay back at these numbers. The verdict is worked out from the conservative case, so a borderline task is called borderline.
Everything is arithmetic in your browser, with the formulas printed in the Method block below. No model is called and nothing is sent anywhere. Your inputs live only in the page and in the link fragment, which never reaches a server, until you choose to press Send this estimate.
| Input | Range |
|---|---|
| Minutes each time | 1 to 480 |
| How often, per working day | 0 to 200 |
| How often, per week | 0 to 1,000 |
| How often, per month | 0 to 4,330 |
| How often, per year | 0 to 52,000 |
| People doing it | 1 to 500 |
| Automatable share | 0 to 95% |
| Cost per hour, optional | 0 to 1,000 |
| Upkeep, hours a month per automated task | 0 to 40 |
| Build cost, optional | 0 to 10,000,000 |
| Running cost a month, optional | 0 to 1,000,000 |
Method
- Times a week
- how often × 5 (per working day), × 1 (per week), ÷ 4.33 (per month) or ÷ 52 (per year)
- Hours a month now
- minutes × times a week × people × 52 ÷ 12 ÷ 60
- Hours back a month
- hours now × automatable share, minus upkeep, never below 0. No upkeep on a task at 0%.
- Conservative case
- the same, with the automatable share 20 points lower, never below 0
- Value a month
- hours back × cost per hour. Value a year: value a month × 12.
- Net a month
- value a month minus running cost a month
- Payback
- build cost ÷ net a month, only when the net is above 0
- Full-time equivalent
- hours back ÷ 173.3 (40 hours × 52 weeks ÷ 12 months)
- Break-even budget
- 12 × the conservative net a month (36 × for 3 years). Without a cost per hour, the same in hours.
- Cost per hour from a salary
- yearly salary × 1.3 ÷ 2,080
The verdict
Worked out from the conservative case, so a borderline task is called borderline. The first rule that matches wins.
- Probably not worth automating at these numbers.
Conservative hours back under 2 a month; or, with a cost per hour, a conservative net of 0 or less; or, with a build cost, a conservative payback over 24 months.
- Borderline: worth it only if the build is small.
Conservative hours back from 2 to under 8 a month, or a conservative payback over 6 and up to 24 months.
- Worth a closer look.
Every other case.
Assumptions
- A month is 4.33 weeks.
- A working day is 1 of 5 a week.
- A full-time month is 173.3 hours.
- The automatable share is your estimate.
- Upkeep is real: someone looks after every automation.
- Not counted: fewer errors, faster responses and the first weeks of setup.
- The 30% salary overhead in the cost helper is a rule of thumb.
Three worked examples
Example: Fernway, the example result above
- 12 minutes × 40 times a week × 3 people × 52 ÷ 12 ÷ 60 = 104.0 hours a month now.
- At 50% with 1 hour of upkeep: 104.0 × 0.5 − 1 = 51.0 hours back a month.
- Conservative (30%): 30.2 hours back a month.
- At 30 USD an hour: 1,530 USD a month, 18,360 USD a year (conservative 906 USD a month).
- Build 3,000 USD, running 50 USD a month: net 1,480 USD a month, payback 2.0 months (conservative 3.5 months).
- Full-time equivalent: 0.29. Verdict: Worth a closer look.
Example: invoice entry, the preset from invoice data extraction
- 6 minutes × 150 times a week × 2 people × 52 ÷ 12 ÷ 60 = 130.0 hours a month now.
- At 80% with 1 hour of upkeep: 130.0 × 0.8 − 1 = 103.0 hours back a month.
- Conservative (60%): 77.0 hours back a month.
- Full-time equivalent: 0.59. Verdict: Worth a closer look.
Example: first-line ticket sorting, the preset from support ticket triage
- 3 minutes × 400 times a week × 3 people × 52 ÷ 12 ÷ 60 = 260.0 hours a month now.
- At 50% with 1 hour of upkeep: 260.0 × 0.5 − 1 = 129.0 hours back a month.
- Conservative (30%): 77.0 hours back a month.
- Full-time equivalent: 0.74. Verdict: Worth a closer look.
- Version
- Cite it as
- Todkett, Automation ROI calculator, version 25 Oct 2026, https://todkett.com/tools/automation-roi-calculator/
02Limits
What it does not do
- It estimates from your own inputs. It does not measure your work.
- It does not count fewer errors, faster responses or the first weeks of setup.
- It never converts currencies.
- Results are estimates, not financial advice.
03Stress test
We try to break every tool before you rely on it.
Automation ROI calculator, check by check
Not yet0 of 4 checks that apply have passed
No recorded test run yet.
Checks marked Not yet are still open.
| Check | What it checks | Status |
|---|---|---|
ST1Test set Not yet The worked example and the unit tests for every formula, range edge and verdict boundary, with the right answer written down first. | The worked example and the unit tests for every formula, range edge and verdict boundary, with the right answer written down first. | Not yet |
ST2Prompt attack Not yet The shared injection strings as every value in the link: each one is ignored and the defaults load. | The shared injection strings as every value in the link: each one is ignored and the defaults load. | Not yet |
ST3Run twice Not needed Runs in your browser: no server Sending the same input twice, or rerunning after a crash, never doubles a record, a message or a payment. | Sending the same input twice, or rerunning after a crash, never doubles a record, a message or a payment. | Not needed Runs in your browser: no server |
ST4Failure recovery Not needed Runs in your browser: no server A failed call is retried a set number of times. If it still fails, the item is parked for a person. Nothing is dropped silently. | A failed call is retried a set number of times. If it still fails, the item is parked for a person. Nothing is dropped silently. | Not needed Runs in your browser: no server |
ST5Load and limits Not yet A 10,000-character link, five tasks at the largest values, and a recompute under 16 ms for every change. | A 10,000-character link, five tasks at the largest values, and a recompute under 16 ms for every change. | Not yet |
ST6Failure alert Not needed Runs in your browser: no server If a run fails, a named person gets an email saying which step failed. | If a run fails, a named person gets an email saying which step failed. | Not needed Runs in your browser: no server |
ST7Human sign-off Not needed Runs in your browser: no server Nothing reaches a customer, a candidate or a money system without a person, unless a fixed rule allows it. | Nothing reaches a customer, a candidate or a money system without a person, unless a fixed rule allows it. | Not needed Runs in your browser: no server |
ST8Rules around the AI Not needed No AI step Fixed rules run before the AI and check its answer after. If the AI fails or is down, the workflow falls back to a safe result. | Fixed rules run before the AI and check its answer after. If the AI fails or is down, the workflow falls back to a safe result. | Not needed No AI step |
No upload Not yet The network log while you estimate: no number leaves the browser unless you press Send this estimate. | The network log while you estimate: no number leaves the browser unless you press Send this estimate. | Not yet |
0 passed, 5 not needed, 4 not yet. We show what we tested, never the test data.
What we test04Changes
What changed
- First version, from the free tools architecture.
- Built the calculator, its worked example and its unit tests.
05Questions
Questions people ask before they trust a calculator.
Does a model or chatbot work out the result?
No. It is fixed arithmetic that runs in your browser. The formulas are printed in the Method block on this page.
How is the conservative case worked out?
It uses the same inputs with the automatable share 20 points lower, never below 0. The verdict is based on it.
Do my numbers leave my browser?
No. They stay in the page and in the part of the link after the # sign, which browsers never send to a server. They leave only if you press Send this estimate.
What does the verdict mean?
Under 2 conservative hours saved a month: probably not worth automating. From 2 to under 8 hours: borderline, worth it only if the build is small. Anything more is worth a closer look. With costs entered, the payback period counts too.
Why is there no ROI percentage?
A percentage hides the hours and the money behind it. The calculator shows hours, value, net per month and payback instead, each labeled as your estimate.
From an estimate to a workflow you can copy.
Each workflow page walks through the whole job step by step: the fixed rules, the one AI step and what a person checks. Build it yourself in n8n, and book a call if you get stuck.
07Credits
Credits
The question this tool asks, how long you can spend on automating a task before it stops paying, is drawn well in xkcd 1205. Nothing from it is copied here. The formulas and code are Todkett's own, with no outside library.