At the end of October, the Bank of Spain published a figure that matches what I see whenever I walk into a business: AI use among SMEs has nearly doubled in two years, reaching 30–35% depending on the technology, but most of those using it are still at an occasional or experimental stage, according to its 2026 analysis. Trying ChatGPT is easy; implementing AI in an SME so it gives you back hours every week is another matter. Here's how we do it, step by step.
Why does occasional AI use barely make a difference?
The same analysis holds the key: around 40% of businesses with moderate use of generative AI see productivity gains, compared with 14% of those using it only occasionally. These are the businesses' own perceptions, not an exact measurement, but the gap is wide.
CaixaBank Research (2026) points the same way: recent growth in Spain comes mainly from quick, cheap uses closely tied to admin tasks, whereas bringing AI into day-to-day operations requires integrating it and redesigning processes.
For your SME, this means something very specific: the leap doesn't come from a better tool. It comes from AI being part of how a task gets done, every day, without relying on someone remembering to open it.
Step 1: which task do you pick first?
Start with a single task, not a tool. I look for tasks that meet four conditions:
- It's repeated every day or every week.
- It almost always follows the same steps, even if there are a few exceptions.
- It takes up hours you can count.
- If it goes wrong, it's spotted and fixed quickly. That's why I wouldn't start with anything that reaches a customer without being checked.
Do the maths before deciding. For example, 150 supplier invoices a month at 4 minutes each is 10 hours a month. At a cost of €25 an hour, that's €250 a month and €3,000 a year, on that task alone.
Swap my numbers for yours. If the result is small, look for another task: not everything is worth automating.
Want to know which processes you can automate in your company?
Book a free assessmentStep 2: is it clear how it's done today?
AI won't fix a process that everyone does their own way: it repeats the mess, only faster. Before building anything, I sit down with whoever does the task and we write down how they really do it: what comes in, what they decide, where they record it and what they do when something doesn't add up.
The exceptions matter most. If that conversation reveals something only one person knows, you've already found a risk for your business, even if you end up automating nothing.
Step 3: how do you connect AI to your tools?
This is the difference between trying and implementing. If someone has to copy data from an email, paste it into a chat and then re-enter it in the ERP, the task is still manual. AI needs to receive the information and leave the result where your team already works: email, WhatsApp, CRM, ERP or a spreadsheet.
A real example: at a motorbike dealership, we connected a WhatsApp agent to live stock levels. In September it handled 77 conversations, and 42% were resolved without anyone stepping in. At the same company, purchase invoices are now recorded automatically: 129 that month, all correct.
Step 4: how do you check it works?
Test with real data, not prepared examples. For the first few weeks, a person reviews everything; once errors drop, they only review the doubtful cases. Set out three things from the start:
- What AI does on its own and what always goes through a person.
- What happens when it doesn't know something: it should flag it, not make up an answer.
- How you measure it: hours, errors and response times, before and after.
After that, someone has to look after it. Tools change, processes evolve and what works today may fail tomorrow if nobody keeps an eye on it. That maintenance is also part of implementation.
Where do I start this week?
Pick three repetitive tasks and note how many times they're done each month and how long they take. If you're not sure how they work, start with this article on how to map your processes. If you're torn between a chat tool and something connected, AI agents vs ChatGPT will help.
In our case studies you'll find examples by sector. And if you'd rather see it with your own numbers, in the free audit we'll tell you which task makes sense to tackle first.
My takeaway from this figure
What interests me most in the study isn't how many SMEs use AI, but how many have genuinely built it into their day-to-day work. I'd rather have one task running on its own every day than ten trials nobody remembers the following month.
Frequently asked questions
One that happens often, almost always follows the same steps and takes up hours you can count. Even better if mistakes are easy to spot and fix quickly.
Not necessarily. The usual approach is to connect AI to what your team already uses, such as email, WhatsApp, your CRM, your ERP or spreadsheets.
That's not the aim: it's about removing repetitive tasks so your team has more time. According to the Bank of Spain (2026), around 89% of businesses using AI see no significant effect on employment.
Written by
Carla Peñalver · Co-founder and Head of Technology
Designs the processes and systems for every ROMBIA project. Turns the way each business works into clear processes and automations that actually work.
