Reading and understanding documents
Invoices, delivery notes, contracts, emails. Pulling the data that matters out of something that is not organised, which is where conventional software gives up.
best returnArtificial intelligence · Valencia
AI applied to your business, not to a demo
I do not sell artificial intelligence: I sell you no longer losing hours. AI is the tool when it fits, and I tell you when it does not.
History kept in Europe No training on your information No lock-in contract
Why so many AI projects stall halfway
Almost always for the same reason: they start with the technology instead of with the problem. Someone decides they need to “add AI”, a prototype is built that impresses in a meeting, and six months later nobody uses it because it did not solve anything that really hurt.
The right order is the opposite, and it is a good deal less glamorous:
- First, where the time goes. Half an hour with the people who do the work, not only with the people who manage it.
- Then, whether to automate. Often the answer is to simplify the process, and that costs nothing.
- Then, whether AI is needed. If the data is organised, an ordinary connection between systems is cheaper, faster and does not fail.
- And only then, which AI. It is the last decision, not the first.
AI is expensive and imprecise compared with conventional software. It only pays off where ordinary software cannot reach: when something messy has to be interpreted (an email written by a person, a scanned document, an enquiry in natural language). There it has no rival. Everywhere else, it is using a jackhammer to hang a picture.
Where it fits
Six uses that do hold up in a small business
They all have something in common: they start from disorganised information, which is exactly where classic software falls short.
Handling the first contact
Answering frequent questions, booking appointments and filtering what deserves your time, on WhatsApp or on your website.
see chatbotDrafting
Replies, reports, product descriptions, meeting summaries. The person reviews and approves instead of writing from scratch.
direct savingSorting and prioritising
Which email is urgent, which customer each incident belongs to, which enquiry is a sale and which is a question.
every morningSearching your own material
Asking in plain language and getting the answer with a reference to the document it comes from. Your manuals, your prices, your contracts.
in high demandSpotting what is out of the ordinary
An unusual order, a price that does not add up, a customer who has not bought for a while.
preventiveThe three questions I am always asked
“What if it makes things up?”
It is the legitimate concern, and the answer is not “it never happens”, because it can. The answer is how it is built so that it does not matter: the assistant answers only from your documents, says where it took each piece of data from, and when a question goes beyond what it knows, it stops and alerts a person instead of improvising.
And above all that, there is always a point where someone in the business approves before anything goes out. Where to place that point is up to you, depending on what is at stake.
“Where does our data end up?”
With providers that guarantee by contract not to train on the content sent, and whenever it is feasible, on my own infrastructure within the European Union. We sign the data processing agreement required by the GDPR and I hand you documentation of which data passes through where and for how long it is kept. No “just trust me”.
“Is this legal?”
It is, with conditions. The EU AI Act already requires transparency: anyone talking to an automated system has the right to know it, and generated content must be identifiable. Everything I build complies as standard: the assistant introduces itself as what it is from the first message.
Interestingly, that works better commercially than hiding it: people cooperate more with an assistant that identifies itself than with one that pretends to be a person and ends up giving itself away.
What I am not going to sell you
- A model trained from scratch. It costs millions and no small business needs one. Existing models are used, connected to your data.
- AI where it is not needed. If your data is already organised, an ordinary integration is cheaper and more reliable.
- A six-month project without seeing anything. The first piece working in two weeks, or something is badly planned.
- Replacing your people. These projects pay off when they free up time, not when they cut jobs. If you are after the latter, I am not your provider.
How I work
From the problem to up and running, step by step
Half an hour listening
With the people who do the work. At no cost. Sometimes it is already clear there that AI is not needed, and I tell you.
A proposal with the numbers
What gets solved, how many hours a month it recovers, what it costs and in how many months it pays for itself. Within 24 hours.
First version in two weeks
Working with real cases, not with examples. It is the only way to know whether it is useful.
It is fine-tuned with real use
I review the conversations of the first month and make corrections. That is where it goes from correct to good.
Common questions
About implementing AI in a business
What exactly is an artificial intelligence agency?
In practice, someone who takes AI models that already exist and applies them to a specific problem in your business: reading your documents, answering your WhatsApp, sorting your email. Nobody trains a model from scratch (that costs millions and makes no sense for a small business); a good one is connected to your data and your rules. The work lies in the integration and the judgement, not in the model.
Will it make things up about my business?
It can, if it is built badly. That is why what I implement answers from your documents and not from whatever the model remembers, cites where it took each piece of data from, and alerts a person when a question goes beyond what it knows instead of improvising. And there is always a point where someone approves before anything goes out to the customer. Where that point sits is up to you.
Will our data end up training an American company’s models?
No, if the providers are chosen well. I work with services that guarantee by contract that the content sent is not used to train their models, and whenever it is feasible, processing takes place on my own infrastructure within the European Union. We sign the data processing agreement required by the GDPR (RGPD in Spanish) and I document for you which data passes through where.
Is my business too small for this?
It is usually the other way round. The smaller your team, the more repetitive work costs you, because it is done by people who should be selling or serving customers. With five people, removing two hours of manual tasks a day between everyone gives ten hours a week back to the business. What is true is that if the volume is very low, it does not pay off: we work that out at the first meeting and I tell you even if it leaves me without a project.
Is there any new regulation to comply with?
Yes. The EU AI Act (the European regulation on artificial intelligence) introduces transparency obligations that already apply: when someone talks to an automated system they have the right to know it, and generated content must be identifiable as such. Everything I build complies as standard: the assistant introduces itself as what it is and the conversations are logged. It is not only an obligation; it also builds more trust than pretending there is a person.
How long does it take and when do you see the return?
Between three and six weeks for a specific implementation, and the first working version is usually ready in two so that you can try it with real cases. The return is calculated before starting: how many hours a month are recovered and in how many months it pays for itself. If the numbers do not work, there is no project.
Specific cases: the assistant that answers on WhatsApp and on your website and process automation when AI is not needed. If you also need an online store or local SEO, I do that too.
Start here
Tell me what is eating up your time.
I will tell you whether AI has a role to play.
Half an hour at no cost. We leave with the numbers worked out and a clear answer, including “do not automate this” if that is the case.