Calculator · nothing is sent
Work out whether it pays off
Enter your numbers. If you don't know one, see below how to estimate it: a rough figure from your own company is better than a perfect one from someone else's.
Hours freed up per year
—
Net saving per year
—
You recover the investment in
—
Return over 3 years
—
The calculator above tells you, with five figures, whether an AI project recovers what it costs and how long that takes. This guide explains what it does with each figure, how to get good figures and what is left out of the sum. The formulas are additions, subtractions and divisions. The hard part is measuring well: if you give it optimistic numbers, it gives you back an optimistic decision. If before calculating you want to understand why there is no average ROI and how to spot an invented figure, we explain it in the ROI of automating processes with AI.
What exactly does the calculator work out?
It starts from a simple idea: an AI project gives you time back. That time has a cost to your company. If the time it gives back is worth more than what you pay for the project and its maintenance, it pays off.
You give it five figures:
| Figure | What it is |
|---|---|
| H | Hours per week your team spends on that process today |
| P | Percentage of that work that gets automated (%) |
| C | Cost per hour of that work to the company (€/h), including salary and Social Security |
| I | Project cost (€) |
| M | Annual maintenance cost (€/year) |
And it makes five calculations, in this order:
- Hours freed per year = H × P/100 × 46 working weeks.
- Gross annual saving = hours freed × C.
- Net annual saving = gross saving − M.
- Months to recover the investment = I ÷ (net annual saving ÷ 12).
- 3-year return (%) = (3 × net annual saving − I) ÷ I × 100.
The payback period tells you how long the money takes to come back. The three-year return tells you how much you gain, in proportion to what you put in, once it is recovered. Look at both: a project that pays for itself in 30 months leaves you almost nothing at three years.
How do you measure the hours your team spends (H)?
Do not estimate it from memory. Measure it over a normal week.
Ask whoever does the work to note down, each day, how much time they spend on the process: entering orders, answering the same enquiries, reconciling invoices, whatever it is. A sheet with three columns is enough: day, task and minutes. At the end of the week, add it up.
Three things to take care of:
- Choose a normal week. Not the quarter-end week or the one in August. If your business is very seasonal, measure one high-season week and one low-season week and use the average.
- Count everyone who touches the process. If the administrator enters the orders but the salesperson checks them, those are hours for both.
- Include interruptions. The minutes that go every time someone asks about an order are also part of the process.
That measured week is useful afterwards: it is the starting picture to compare against.
What percentage of the work really gets automated (P)?
This is where people go wrong most. The temptation is to put 100%. It is almost never like that.
In almost any process there is a repetitive part, which AI does well, and a part of exceptions: the customer who asks for something odd, the invoice with an error, the case that needs judgement. That part stays with a person. And someone has to review what the system does, especially at the beginning.
To estimate P, go back over what you noted during the measuring week and split the tasks into two piles: those that are always the same and those that are different each time. The first pile is your ceiling. Then take something off for the review. If you are not sure which part can be automated, the AI self-assessment for businesses helps you sort it out.
Putting in a cautious percentage is not being pessimistic. It is making sure the sum holds up once the project is running.
What does an hour of work really cost you (C)?
It is not the salary the person earns. It is what they cost the company.
Take that person's total annual cost: gross salary plus the Social Security contributions the company pays. You have it in the payslips or your accountant can give it to you. Divide it by the hours they really work in a year. If they work 40 hours a week, the maximum ordinary working week set by Article 34 of the Workers' Statute (Estatuto de los Trabajadores), and you take off holidays and public holidays, you get about 1,840 hours (40 × 46).
As a reference, Spain's National Statistics Institute (INE) put the average labour cost per effective hour at €26.18 in the second quarter of 2026, according to the Quarterly Labour Cost Survey published on 17 September 2026. It is an average for the whole of Spain and all sectors. Your figure may be well above or below. Use your own.
If several people on different salaries touch the process, calculate an average weighted by the hours each one spends.
Why 46 weeks and not 52?
Because nobody works all 52 weeks of the year. The Workers' Statute sets a minimum of 30 calendar days of holiday (Article 38) and up to 14 public holidays a year (Article 37). Thirty calendar days is a little over four weeks; fourteen working days, almost three. You are left with about 45 weeks.
The calculator uses 46 as a round figure. If your company closes more (bridge days, the whole of August), lower the weekly hours a little to compensate.
What goes into the project cost (I) and into maintenance (M)?
I is everything you pay to have the system running: development, integration with your programs and going live. The quote you are given sets it. If a grant pays part of it, enter what comes out of your own pocket, but only once it has been awarded. The ones that exist and the ones coming are in the guide to AI grants for businesses in 2026.
M is what it costs each year it keeps running: servers, use of the AI models, support and adjustments. Ask for it in writing, itemised. A project that is cheap to build with expensive maintenance may never pay off.
What hidden costs are often forgotten?
The calculator does not see them unless you give them to it. Add them to the project cost (I):
- Your team's time during implementation. Someone has to explain how the process works, review tests and give the go-ahead. Those are hours not spent on their own work.
- Training. Learning to use the new system and to know when not to trust it takes time.
- Process changes. Sometimes you have to reorganise how things are done so the automation fits: a different order template, a different order of steps.
- The first weeks at half speed. While it is being adjusted, the old method and the new one live side by side.
Work out those hours and multiply them by the same cost per hour (C). That way they do not turn up later as a surprise.
Worked example, step by step
Example with made-up figures. It does not correspond to any client or any real quote. It is there to show how the sum is done.
A distributor receives orders by email and one person enters them by hand into the management software. They measure a normal week and find they spend 15 hours on that task. Going back over what was noted, they see that 60% of the orders always arrive in the same format; the rest are orders with changes, doubts or errors that a person has to look at.
That person costs the company €36,800 a year, counting salary and Social Security. Divided by 1,840 hours (40 × 46), that is €20 an hour.
| Figure | Example value |
|---|---|
| Hours per week (H) | 15 h |
| Percentage automated (P) | 60% |
| Cost per hour (C) | €20/h |
| Project cost (I) | €12,000 |
| Annual maintenance (M) | €1,800/year |
The sum:
| Step | Operation | Result |
|---|---|---|
| Hours freed per year | 15 × 0.60 × 46 | 414 h |
| Gross annual saving | 414 × 20 | €8,280 |
| Net annual saving | 8,280 − 1,800 | €6,480 |
| Net saving per month | 6,480 ÷ 12 | €540 |
| Months to recover | 12,000 ÷ 540 | 22.2 months |
| 3-year return | (3 × 6,480 − 12,000) ÷ 12,000 × 100 | 62% |
Now, two variations that show why measuring well matters.
If they had put in 100% automation. The hours freed would go up to 690, the net saving to €12,000 and the investment would be recovered in 12 months, with a three-year return of 200%. It looks like a much better project. It is not: it is the same project with a false figure.
If they add the hidden costs. Suppose implementation takes 40 hours from the team and training another 20. That is 60 hours at €20, €1,200 more. The project cost goes to €13,200:
| Step | Without hidden costs | With hidden costs |
|---|---|---|
| Project cost (I) | €12,000 | €13,200 |
| Months to recover | 22.2 | 24.4 |
| 3-year return | 62% | 47% |
It still pays off, but it takes over two years to get the money back. With that figure in hand, the company decides with its eyes open.
What benefits does the formula not capture?
The calculator only counts hours. There are benefits that cannot be measured that way and are worth valuing separately:
- Fewer errors. A mistyped order means a wrong shipment, a return and an annoyed customer. If you know how many errors you have a month and what each one costs, you can add it in.
- Faster response. Answering in minutes instead of hours can change how many quotes get closed.
- Out-of-hours sales. A system that answers at eleven at night picks up orders or appointments that are lost today.
- Growing without hiring at the same pace. If volume goes up, the system absorbs part of the increase.
Do not put them into the calculator by eye. If the project is already profitable without them, they are margin in your favour. If it only pays off by adding them, measure them before signing.
And a warning in the opposite direction: a freed hour is only a saving if it is used for something. If that time goes to serving customers better or to another task, the saving is real. If nothing changes, it stays on paper.
When does it not pay off?
There are cases where the honest answer is that it is not worth it, at least not yet:
- The net saving is zero or negative. If annual maintenance costs more than the hours it frees, the project never pays for itself. The calculator will show you that with no payback period.
- The process takes few hours. Automating something that takes two hours a week rarely justifies custom development.
- The process changes every time. If there is no stable, repeated part, there is nothing to automate with any guarantee.
- The process is messy. Automating a chaotic process makes it faster, not better. First you tidy it up.
- The payback period is longer than your company can wait. If recovering the investment takes longer than the process will last as it is today, the sum does not hold.
If you are in one of these cases, it is not that AI is not for you: it is that this process is not the first one. In how to choose your first AI project you have criteria for picking one that does pay off.
What does this have to do with the AI voucher?
The IA360 Plan, presented by the Government on 21 September 2026, foresees that rolling out the Artificial Intelligence Voucher (Bono de inteligencia artificial) will include a prior maturity diagnosis, the identification of one specific use case and the subsequent measurement of the impact on productivity. It also foresees payments by milestones.
The rules are not yet published and it is not known how that impact will be measured. But to measure a before and an after you need the before. The hours you measure today with the calculator are exactly that: your starting point.
And after you calculate?
If the sum works out, the next step is to check that the saving actually happens once the system is running. Measure the process hours again at three and six months and compare them with your starting week. A business dashboard that collects that data only saves you doing it by hand.
If you are not sure which process to measure or you are missing data, a business audit helps locate where the time goes before investing in anything. And if you want to go over your numbers with us, write to us.
Sources
Xora IntelligenceAI consultancy and custom software in Los Barrios (Cádiz). We write about what we build for businesses in the province. If you have a question about your case, write to us.




