In most companies someone spends part of every week copying data from one place to another. A supplier invoice arrives by email and gets keyed into the accounts. A delivery note arrives as a photo on WhatsApp and ends up in a spreadsheet. Our AI process automation in Barcelona takes that part away: AI reads the document and pulls out the data, a person checks it and approves it in one click, and it lands in your system. Your team stops typing and starts approving.
Which processes are worth automating
Not all of them. If AI doesn't improve the solution, we don't add it. Business process automation pays off when you hear lines like these:
- "One person spends most of their time keying this into the system."
- "Delivery notes come in on paper, by email and on WhatsApp."
- "Typos slip through and the order goes out wrong."
- "Everything has to be checked by one person, and she's swamped."
If any of those sounds familiar, there's something to automate. And it can nearly always be done without changing the software you already use.
Invoices, orders and delivery notes: from document to system
This is the clearest place to start. The document comes in wherever it comes in (email, WhatsApp, a shared folder, a form), AI extracts the information that matters and structures it. A person reviews and approves it, and the data lands in the tool where it belongs. Some typical examples:
| Document | How it arrives | Where it goes |
|---|---|---|
| Supplier invoices, often in Spanish | Email, PDF attachment | Accounting: Holded, Sage, A3, or the sheet your gestoría asks for |
| Delivery notes from site | Photo on WhatsApp, paper | Cost control for each project |
| Purchase orders | Email, Excel | Your ERP or order sheet |
| New customer onboarding | Forms, scanned documents | Your CRM |
The detail that changes everything: a person still validates. We don't sell taking people out of the loop. We sell people reviewing instead of retyping.
If your software has no easy way in, which is more common than vendors admit, that's not the end of it. We look at how it works first, then pick the simplest reliable route. There's more on that in systems integration.
AI document review: a second pair of eyes
The other half of AI automation isn't capturing data, it's checking documents. You hand over the document and the criteria, and it returns a justified assessment, point by point. We don't ask a single AI: we ask several and cross-check their answers, more like a panel than a single examiner. The result doesn't hinge on one model's judgement, and there's a written record of why each thing was approved.
For example:
- Marketing material checked against the brand guidelines before it goes to print.
- Content checked against the regulations of your sector.
- Supplier bids compared against what the tender asked for.
- Deliverables checked before they go to the client.
It's useful wherever everything funnels through one overloaded person, or where someone could ask you tomorrow to prove why something was signed off.
People still make the decisions
Good automation doesn't remove humans from every decision. The split is this: software searches, organises, extracts, checks and suggests. People review, decide and act. Anything unusual always reaches someone. Automation without losing control.
A real case: Los Vasos de Agua Clara's contacts
Los Vasos de Agua Clara make hand-painted glassware. Their orders and contacts arrived from everywhere and were entered into Holded by hand.
"Orders and contacts came in through Instagram, email and in person, and all of it went into Holded by hand. Now every contact lands in Holded on its own, sorted by how that customer found us."
Nobody types those contacts any more, and they know where each customer came from. It's the kind of thing we do for ecommerce automation and integrations. On a building site, the equivalent is automating delivery notes for construction companies.
How we get it running
We start with a single process, the one that costs you the most time. We look at it with whoever does it every day, send a proposal with a fixed price in writing, and build the smallest useful version. Real people test it and, once it works, you decide what comes next. We don't start from scratch: the engines have been running for years, which is why it's weeks and not months. More on how we work.