How to calculate the payback of automation
Most articles about automation ROI end with a percentage you have no way to verify. This one ends with a number you work out yourself — on your own hours and your own rates.
The payback of automation is simple to calculate: monthly saving = hours saved × your hourly cost rate, minus the solution's monthly running costs. Payback period = deployment price ÷ monthly saving.
The whole trick is putting real numbers into it — not an estimate of hours, which is usually overstated, and not a salary instead of a cost rate.
A practical threshold: if the payback period comes out above 18 months, the project usually does not make sense — the process will change in that time anyway. Under 12 months it is a good investment, under 6 months a very good one.
Payback calculator
Enter your numbers. Nothing is sent anywhere — the calculation runs in your browser.
The calculator is a quick estimating aid, not financial advice. It does not account for tax, depreciation or the value of fewer errors — which in practice is a further, harder-to-quantify benefit.
The formula and what each part means
- Hours saved per month = hours per week × 4.33 × the share automation takes over.
- Gross saving = hours saved × cost rate.
- Net monthly saving = gross saving − monthly running costs.
- Payback period (months) = deployment price ÷ net monthly saving.
- Net benefit in year one = (net monthly saving × 12) − deployment price.
The 4.33 is the average number of weeks in a month (52 ÷ 12). I use it because counting a month as four weeks systematically understates the saving by roughly eight per cent.
How to put in numbers that are not made up
Hours of manual work
Do not estimate. Log every interruption and every data retyping for three days. Estimates are wrong in both directions — for tedious tasks they are overstated, for fragmented ones understated, because nobody counts the time it takes to get back to the original work.
Cost rate
Not gross salary. Cost rate = total annual cost of an employee (salary, contributions, equipment, space) ÷ hours actually worked in a year. It usually comes out markedly higher than you expect — and it is the right number for this calculation.
The share automation takes over
This is the field where the biggest mistake gets made. It is almost never 100%. There will always be checks, exceptions and cases the system hands to a person. A realistic range is 60–85%. If somebody promises you a hundred per cent, they either do not understand the process or have not told you about the exceptions.
Monthly running costs
- The tool subscription, or the server if self-hosted
- The language model API if one is used (by volume of text)
- Maintenance time — a few tens of minutes a month for a simple solution, but not zero
What this calculation does not capture
The calculator counts saved time only, because that is the one quantity you can verify. In reality these benefits are also in play — and I deliberately do not quantify them, so the result stays honest:
- Fewer errors. One badly retyped order or a missed due date can cost more than a whole month of running costs.
- Faster response. With enquiries, who answers first often decides. This can only be measured in hindsight, in jobs won.
- Capacity without hiring. Handling twice the volume without another person is a saving that never shows up in hours.
- Less dependence on one person. When a process lives in a system, it does not stop for a holiday.
I recommend calculating payback from time alone. If it looks good on that basis, the other benefits are a bonus. If a project only pays off once hard-to-measure effects are counted in, that is a warning sign.
When it is not worth it: the calculation comes out badly
Payback over 18 months. In that time the process, the tools or the suppliers will change. Postpone the automation or narrow the scope.
A saving under 5 hours a month. Too small a volume for deployment and maintenance to pay for themselves.
An automation share under 50%. That means the process has too many exceptions. Simplify it first.
Running costs close to the saving. Then the whole project is a pointless risk — you are swapping manual work for a monthly bill.
If one of these is your case, the honest answer is not to automate. That is a legitimate result of the analysis, and I will tell you so too when you ask about a specific process.
Frequently asked questions
How is the payback of automation calculated?
Monthly saving = hours saved × hourly cost rate, minus monthly running costs. Payback period = deployment price ÷ monthly saving. Hours saved are calculated as hours per week × 4.33 × the share automation actually takes over.
What rate should I put into the calculation?
Not gross salary but the cost rate: total annual cost of an employee including contributions, equipment and space, divided by hours actually worked. It comes out markedly higher than the salary and it is the right number, because that is the money the company actually pays.
How much of the work does automation really take over?
A realistic range is 60–85%. A hundred per cent almost never — there will always be exceptions, checks and cases the system hands to a person. If somebody promises the complete removal of manual work, they are probably not counting the exceptions.
What payback period is good?
Under 6 months is very good, under 12 months is good. Above 18 months a project usually does not make sense, because the process or the tools used will change in that time anyway.
Should I count fewer errors and faster response into the saving?
Into the decision yes, into the calculation better not. I recommend calculating payback from saved time alone — if it looks good on that basis, the other benefits are a bonus. If a project only pays off once hard-to-measure effects are counted, that is a warning sign.
What are typical monthly running costs?
For a simple solution, on the order of single-digit to tens of euros — a server or a tool subscription. If a language model is used, add the API cost by volume of text processed. An estimate should be part of every proposal.
How do I know how many hours a process takes us?
Log every interruption and retyping for three days — one line in a spreadsheet. It is the cheapest part of the whole project and the one that most affects whether the decision is right.
Related services
Related articles
Got a number that makes sense?
Send me the process description together with the inputs you used. I will check whether they are realistic — and if they are not, I will say so before you put money into it.
Get a free consultation →