Supplier invoices at an accountancy firm
Staff only review the uncertain fields. Tax number, date, net amount, VAT and withholding are lifted from the PDF or photo and arrive as a draft in the accounting package.
Forget how impressive AI looks on stage. What matters is which specific work it takes off your team, what it costs each month and where the data you feed it ends up. We work with small firms, law and accountancy practices, clinics and public bodies that want answers in figures: which tasks are worth it, how to trial them safely, where to host models and documents inside the EU, and how to teach staff to use these tools with sound judgement. Everything happens remotely, with no site visits.
You can start with any of them. If you are starting from nothing, the sequence that wastes least money runs like this: decide what is worthwhile, test it against genuine daily cases, host it somewhere secure and train the people who will use it.
What your company already uses, which opportunities pay back, which data is missing and what risks the GDPR and the European AI Act raise. It ends in a phased roadmap that management can approve or cut back.
A pilot on a genuine task, timed before and after, with AI placed inside the programs you already run: Outlook, Teams, a3ERP, Holded or your PrestaShop store. If it fails to save what was forecast, it stops.
A sealed environment in your own account, in AWS, Azure or Google Cloud regions in Spain or with European providers. GPUs hired by the hour, search over your documents and spending kept under watch. We sell no hardware.
Online workshops per department, built on cases from your own work. Every session drills the same two habits: checking an answer before relying on it and knowing which data never goes into an assistant.
A good candidate comes round many times a week, a one-off mistake is cheap, and the output passes a person before it reaches a customer or the accounts. If any of those three is missing, you will hear it from us before anything is quoted.
Staff only review the uncertain fields. Tax number, date, net amount, VAT and withholding are lifted from the PDF or photo and arrive as a draft in the accounting package.
An agent signs off the reply. The customer's email is matched to the order and the SEUR or Correos tracking, and the system drafts an answer showing the real delivery status.
Every answer links to its source. Clinic staff ask how instruments are sterilised or what to do about a cancellation, and the assistant quotes the exact paragraph of the protocol. It never diagnoses.
Whether to bid is still your call. A construction firm gets a digest of solvency requirements, deadlines, award criteria and BIM demands for each tender published on the Spanish public procurement platform.
Nothing goes live unchecked. A first draft is produced from the technical sheet and shaped to the format of Amazon.es, PcComponentes or Miravia, each with its own rules.
Your customers' data should not end up in a free assistant opened on someone's personal account. Ahead of the first test, you and we settle which information may be sent to a third-party provider and which is kept inside Microsoft 365 or a private environment in a Spanish or EU region. Once personal data comes into play, three things are required: a processor agreement, an entry in your processing register and, where the risk justifies it, an impact assessment. It is all written down for your DPO, because that is what lets you justify the use to the AEPD or, should it arise, to AESIA as Spain's AI supervisory authority.
Nothing is rolled out until a brief trial with genuine data has proved its worth. Workshops, configuration and testing take place over Teams or Google Meet, Monday to Friday, 9:00-18:00 Spanish time (CET).
We clock the task as it stands and set a pass mark together, for example «nine emails in ten reach the right department without edits».
A handful of users apply it to their everyday workload for a set number of weeks. Someone approves each result while we log accuracy and minutes saved.
Results are set against the baseline. Should the pass mark be missed, we say so plainly and close the pilot, since an unsuccessful test is much cheaper than software that sits unused.
Everyone else concerned gets access, along with one page of usage rules, periodic quality sampling and the paperwork your DPO needs.
The Act is not aimed solely at model developers: obligations depend on what the system is used for and on the role your company plays. In Spain, oversight falls to AESIA. Our contribution is practical: mapping the tools people actually use, the reasons and the data involved, then helping draft in-house rules. Judging your legal position is a job for your lawyers or DPO, not for us.
Wherever you choose, within what each provider offers. The major clouds have regions in Spain (AWS in Aragon, Azure and Google Cloud in Madrid), and there are European providers such as OVHcloud, Hetzner, Arsys and IONOS. We pick services whose business terms let processing be pinned to the EU and forbid using your material to train models, and we recheck those terms before starting because they change over time.
Yes, though never by copying it into a free chatbot. The groundwork is a lawful basis, a stated purpose, a processor contract with the provider, a processing register entry and, where appropriate, an impact assessment. In healthcare, health data is a special category and a private environment is advisable. Often the simplest fix is to send less: sorting a complaint needs the order number and the reason, not a name or address.
For drafting, summarising and searching your own documents it sometimes is, and we will say so. But Copilot sees whatever each user can see, so if SharePoint and Teams permissions are a muddle, it will surface files it should not. Those permissions ought to be reviewed and staff trained before it is enabled. Tasks tied to your ERP or web shop usually call for something more tailored.
Each pilot covers one process and is either priced as a fixed project or charged at €75 per hour + VAT, capped in advance. Model consumption comes on top, invoiced by the provider by volume; we forecast it before work begins and keep an eye on it throughout. For continuing support, AI work can be folded into the Start, Business or Premium plans, while a dedicated group of specialists is quoted per project.
Tell us which task your team repeats every day and which program it lives in. We will review the available data and suggest a pilot with a measurable goal.
Your request is with us
Expect an answer within one working day. A reported fault that has halted your team is handled first.
No match found. Try another spelling, or go with the closest provincial capital: every job is done remotely, so the location makes no difference to what we deliver anywhere in Spain.