What it costs to automate an SME

“What does automating my company cost?” is one of the first questions that comes up, and the honest answer is: it depends. But that does not mean you cannot get a clear picture. In this article we explain what the price depends on and, above all, how to work out whether it pays off.

The honest answer is that it depends, and anyone who gives you a figure without knowing your operation is selling you smoke. But “it depends” does not help you decide. So let us get to the useful part: what exactly it depends on, what is inside the price, which recurring costs almost nobody mentions and how to work out whether it pays off before spending a euro.

What the price actually depends on

Automating is not buying an off-the-shelf product. Two companies in the same sector and of the same size can receive very different quotes for reasons that have nothing to do with size. These are the factors that really move the number:

1. How many systems must be connected

This is the heaviest factor and the least anticipated. Automating within a single tool is cheap. Automating a process that starts in a web form, goes through email, is logged in a spreadsheet, is reflected in the CRM and ends in the invoicing software is another matter: every hop between systems is a point where something can fail and must be anticipated.

2. Whether those systems have an API

Modern tools offer an official way to connect. Older management software, very common in Spanish firms and SMEs, often does not. When there is no API, you have to find alternatives ( scheduled exports, file reading, automating the interface itself ) and all are more fragile and more expensive to maintain. If your ERP is fifteen years old, say so in the first conversation: it changes the quote.

3. How many exceptions the process has

Here is the most common trap. Everyone describes their process as if it were linear: “the order comes in, it gets checked, it gets invoiced”. The reality is “the order comes in, it gets checked, unless it is a long-standing client in which case Marta handles it, or it is over a certain amount in which case the boss must approve it, or it comes from France in which case the VAT changes”.

Those exceptions are written down nowhere and they decide whether the automation holds. A process with three exceptions costs a fraction of what one with fifteen costs.

4. What volume it handles

A flow that runs fifty times a month and one that runs fifty thousand are not built the same way. The second needs error handling, retries, queues and active monitoring. The first does not.

5. How much judgement is required

Sorting emails by sender is deterministic: the same input always gives the same output. Drafting a reply that sounds like your company requires a language model, testing, tuning and a review mechanism. The more judgement a task needs, the more work it takes to make it reliable.

The three pricing models on the market, and what each hides

When you ask for a quote you will meet one of these three. None is bad; what is bad is not knowing which one is being applied to you.

Fixed-scope project

A fixed price for a defined scope. For: you know what you are paying from the start and the risk of it dragging on sits with the provider. Against: it forces you to define the scope precisely before starting, and anything that appears later is a separate quote. With poorly documented processes, things appear.

By the hour

For: flexible, and it fits well when neither you nor the provider yet knows how deep the problem goes. Against: the risk sits with you. Always ask for an agreed cap and a warning when a certain percentage is reached.

Monthly subscription

For: it spreads the cost, includes maintenance and aligns incentives: if the automation breaks, the provider has reason to fix it quickly. Against: it is worth reading the small print: what happens if you leave, who owns the flows, where the accounts live. If cancelling leaves you with nothing running, that was not a maintenance subscription: it was a rental.

What is inside the price

An automation quote that just says “automation development” lets you compare nothing. This is what you are actually paying for:

• Diagnosis. Understanding the process, its exceptions and which systems it touches. It is usually free, and if it is not, it should be deducted from the project.
• Design. Deciding what gets automated, what stays with a person and what happens when something fails. It is the part that most determines whether the result is any good.
• Build. Building the flow. It is usually less than half the total work, even though it is the only visible part.
• Testing. Breaking it on purpose: what happens if a field is missing, if the customer replies something odd, if the destination system does not answer. If a quote does not mention this phase, ask about it.
• Rollout. Moving it into the live environment and running alongside the manual process for a few days before switching it off.
• Documentation and handover. Having it written down what it does and where it lives. Without this, you depend on the provider forever.
• Maintenance. Processes change. An unmaintained automation degrades on its own.

The recurring costs almost nobody tells you about

This is where companies get the most surprises, because the initial quote rarely itemises them and they show up the following month.

Automation platform licences

Tools like Make or n8n charge per operation run. A five-step flow triggering two hundred times a month uses a thousand operations. Free plans run out sooner than you would think, and when they do the scenario stops. Always ask how many operations what they are building will consume.

AI model usage

If the automation uses a language model, every run costs money. Little per run, but it multiplies with volume. And there is an enormous difference between models: choosing a powerful one for a task a basic one handles can multiply the bill tenfold without improving the result. It is one of the technical decisions with the greatest long-term cost impact.

Maintenance

It is not optional. A tool changes, new people join, an unforeseen case appears. You can absorb it internally if you have someone capable, or contract it out. What you cannot do is assume it does not exist.

How to work out whether it pays off

The useful question is not “what does it cost” but “what does it give back”. And that can be estimated before hiring anyone.

Step 1. Count how many times a month the task runs.
Step 2. Measure how many minutes it takes each time. Measure it, do not estimate: almost everyone underestimates.
Step 3. Work out the hourly cost of whoever does it, adding gross salary and employer costs.
Step 4. Multiply: times × 12 × minutes ÷ 60 × hourly cost.

An example with round numbers, just to see the mechanics. A task done 200 times a month, taking 12 minutes, performed by someone costing €22 an hour: that is 480 hours a year and about €10,500. If automation removes 85% of that time, you recover around 400 hours and close to €9,000 every year, indefinitely.

You have that calculation set up in the savings calculator: move three sliders and it gives you the figure for your case.

And note one nuance: the saving is recurring, the cost is almost all upfront. A project that takes eight months to pay back looks expensive in year one and is a bargain by year three.

A real case with figures

Santa Coloma de Gramenet City Council had to anonymise around 5,000 documents a year before publishing them, removing names, ID numbers, addresses and bank accounts to comply with the GDPR. Each document took about 20 minutes of manual work: over 1,600 hours a year, the equivalent of one full-time person doing nothing else.

Con RelevX Redactor, the time per document went from 20 minutes to 2-3 seconds. This is not a cherry-picked extreme: it is what happens when a task has a clear pattern and high volume. Those two conditions together are what make automation pay for itself.

The cost of not automating

There is an expense that appears on no profit and loss account: carrying on by hand. Every hour your team spends copying data is an hour not spent selling or serving better. And there are two more costs that are almost never counted.

The first is errors. A repetitive manual process fails, and every failure costs: an invoice issued wrong, a customer nobody answered, a deadline missed. The second is turnover. Nobody signs up to spend the day copying data from one place to another, and the good people tire of it first.

Signs that a quote does not add up

• A fixed price without having seen your operation. If they have not asked about your exceptions, they do not know what they are quoting for.
• It does not mention recurring costs. Licences and AI usage always exist. If they do not appear now, they will appear later.
• It does not say who owns the accounts. If the flows live in the provider's accounts, the day you part ways you are left with nothing.
• It promises to automate 100%. No real process automates entirely. Whoever promises it has not looked at yours.
• Well below the rest. It usually means they have skipped design and testing, which is exactly where it is decided whether it holds.

So where do I start?

With whatever makes sense given the return, and starting small. Pick one contained process with high volume and a clear pattern, build it and measure it. If the result is there, you move to the next with data in hand rather than hunches.

Starting with a large project touching five departments stretches the time to the first result and makes it impossible to know what worked. Besides, the second process always comes out cheaper than the first: the connections are already built and the map of how you work is already drawn.

If you want to see where it usually makes sense to start, here are five processes you can automate today. And if you would rather we looked at your case, the diagnosis is free: in 45 minutes we tell you what to automate first and what is not worth touching.