AI agents
grounded in your business.
We build AI that does real work in your company.
The foundation. A data warehouse or a data lake where the data from all your systems arrives on its own every night, clean and under the same rules. Without it, nothing you put on top can be relied on.
See and anticipate. Business Intelligence for the figure everyone trusts. And machine learning models that predict demand, bad debt or breakdowns before they happen.
Answer and act. An AI knowledge base (RAG) that answers questions about your contracts and procedures and cites the document. And AI agents that take a goal, decide within the limits you set and finish the task.
The whole process. AI workflow automation: the process runs end to end, and a person decides only the exceptions.
With models from OpenAI, Anthropic, Google or open models, depending on what the case needs and where your data has to stay. If you are not sure which product is yours, then what you are looking for is our AI consulting.
Someone in accounts pays for ChatGPT out of their own pocket. Marketing subscribes to another tool that nobody else knows about. And in several teams, customer data gets pasted into personal accounts just to get the job done.
That is not a company that has adopted AI. It is a collection of uncontrolled individual initiatives, each with its own tool, its own judgement, its own data and its own risks. What they learn is kept in no process and no system, so when that person moves to another role, it leaves with them. And every month there is one more subscription, and one more place where company data has ended up.
Adopting AI is something else. It means AI that is predictable, reliable and repeatable, inside a process, with someone responsible for supervising it and a record of what it has done. And it means doing things in order: an agent built on data that contradicts itself amplifies the mess instead of fixing it. That is why we are not an AI agents company and nothing more. Dashboards, predictive models, data warehouses, knowledge assistants and agents are all pieces of the same path.
Choosing where to start is half the work. It is decided in a discovery session, before any budget is committed.
Lots of isolated trials Each with its own data, rules and owner
A few projects, in sequence Each leaves the house tidier for the next
01One integrated source
02One agreed metric
03One documented process
We start with the process that costs money, not with the technology
Before we build, we agree how we will measure whether it worked
People stay in the process as supervisors, never outside it
We check that your data can carry what we are going to build on it
Progress valuations, cost overruns on each project and the paperwork that goes back and forth between the technical office and the site. Here AI usually starts with the documents (tender specifications, contracts, site minutes), not with the dashboards.
Paperwork first, numbers second.
Maintenance before the breakdown, batch traceability and waste. It is the industry where shop-floor data already exists and is almost never set against what things cost.
The data is there; it just needs a price on it.
Occupancy forecasting, pricing and the conversation with guests in several languages, at any hour and without keeping anyone waiting.
Anticipate occupancy and reply in any language.
Consultancies, advisory firms and professional practices where what you sell is the document: reports, opinions and proposals that repeat what has already been written. It is the industry where the firm's own archive is worth more than any new system.
Here the product is the document.
Sensitive data and traceability. The first decision is not which model to use, but where the information stays and who can see what.
Architecture before the model.
Shipyards, marinas, yacht clubs and service companies: refits that are badly planned, berths managed by hand and a history of past work that nobody looks at.
Every refit is different, and the records already know how.
Stock that is calculated rather than estimated, lead times and routes, and orders that arrive by email and are keyed in by hand. It is where an agent shows its value soonest.
From inbox to ERP, without retyping.
No company needs the whole catalogue. It is ordered along the path data follows inside a company: explore, organise, decide, innovate, lead. Filter by your stage and keep what applies to you.
The way in: an assessment of where you stand, use cases ranked by return and a roadmap with timelines.
See the product
Data warehouse, data lake, automated loading, quality and governance. It is what makes everything on top of it tell the truth.
See the product
Dashboards for each role, a metrics dictionary and reports that send themselves. The figure everyone trusts.
See the product
Models that predict demand, bad debt, customer churn or breakdowns from your historical data, built into day-to-day operations.
See the product
Ask your manuals, tender documents and technical data sheets, and get the answer with the source document alongside it.
See the product
Business tasks carried out from start to finish: they query your systems, decide within their limits and act, with oversight where it matters.
See the product
Complete processes that reason over your data, make decisions and run themselves. Replenishment, month-end close, orders, claims.
See the productWhat people ask us on the phone before the first meeting.
Let's talk about your projectWe are a data and AI consultancy for mid-sized companies. We build three kinds of thing. AI agents that carry out business tasks from start to finish. Assistants that answer questions about your company's documents and cite their source. And the data platforms and dashboards that the first two rest on. We are a European team based in Palma de Mallorca and Barcelona, working in English, Spanish and Catalan under European rules (GDPR and the EU AI Act), and we have worked with more than 50 companies on projects delivered or under way.
It is designed for companies of 20 to 500 people, in any industry. The signal we look for is not size or industry: it is that you already have an ERP and that someone spends hours every week moving information around by hand. If that is happening, there is a case. What does change from one industry to another is the project. The same agent will not work for a hotel and for a factory: the technology is similar, but the process, the vocabulary and who signs off the decision are not.
No, and that is the most common situation: most companies that come to us have nothing set up. You start by sorting out the bare minimum the first useful project needs, not by building a complete platform that takes a year to show any sign of life. Skipping a whole stage is what fails: an agent that reads contradictory data spreads it at full speed. How much has to be sorted out before building is one of the questions the initial assessment answers.
With a knowledge assistant over a limited set of documents, or with an agent that does a single repetitive task. They are the two projects that show visible results soonest and do not require everything else to be in order first. If you are not sure which of the two, that is exactly what AI consulting is for.
A program that receives a goal instead of instructions. It queries your systems, decides what to do within the limits you have set, carries out the task and keeps a record. The difference from a chatbot is that a chatbot tells you what you should do, and an agent does it.
Retrieval augmented generation: the technique that lets a language model answer from your documents instead of from what it learned on the internet. It finds the relevant passages in your contracts, procedures or technical data sheets and writes the answer, citing where it comes from. It is what keeps an assistant from making things up, and it is the basis of our AI knowledge base.
That depends on the option you choose, and the choice is yours: in your own cloud, in ours or on your own servers. We work with closed models and open models so that this choice is possible. In projects that use third-party models, we use contracts that exclude training on client data.
An assessment with a roadmap is measured in weeks. A first system in production (a dashboard, an assistant or an agent on a specific task) takes months, not years. The timeline depends on the state of your data, so it comes out of the assessment, not out of a project template.
The system is yours: the code, the data model and the documentation are handed over, in a form that lets another team carry on. If you would rather we looked after maintenance, that is agreed separately. But the option of not depending on anyone has to be on the table from the start: if it isn't, the project turns into a subscription.
