S.03 / CONSULTING
AI consulting for businesses that doesn’t end in a slide deck: a diagnosis of which processes suit AI, feasibility on your data, risks and the EU AI Act, a pilot and putting it into production.
02 / PROBLEM
Recommending and building shouldn’t be separated: whoever recommends should be able to answer for what gets built.
Many companies have already paid for an AI study with an opportunity matrix and a phased plan. Months later, nothing is running: the analysis never reached the real data, nobody measured the cost per operation and the team that had to build it wasn’t part of the decisions.
I approach consulting differently: the diagnosis is validated on your data, the recommendation includes what it will cost to run, and the first piece is built and put into production.
03 / PROCESS
Six phases. Any of them can end in “not worth it”, and that’s a result too.
Which processes eat up time, where the data lives and which decisions depend on it.
Your real data shows which step suits AI, which is solved with rules and which isn’t worth it.
What data goes to the provider, what the AI Act says and what happens when the model gets it wrong.
One bounded piece running with real users, with cost and quality measured.
Integrated into your systems, with permissions, logging and someone accountable.
Check whether it saves what it promised. If not, fix it or retire it.
05 / DELIVERABLES
Six concrete deliverables. None of them is a document that gets filed away: they’re all for deciding or building.
A short list of processes ranked by impact and feasibility, with what each one needs: rules, AI, data or a person.
Not a demo with sample data: real questions and cases run against your information and evaluated before anything is promised.
EvidenceAtalayaIQ
What data travels to the AI provider, which AI Act risk category the use falls into and which controls it needs.
What each query or run will cost in production, before anything is built, so the savings aren’t eaten by the model bill.
The piece with the highest return, built and integrated with real users. The consulting ends with something that works.
EvidenceRocio.com
A hands-on session so your team knows what AI does in their work, what it doesn’t do and when to distrust it.
06 / FIT
Consulting works when there are specific processes and data to work on.
07 / CONSULTANCY OR SPECIALIST
Both make sense. It depends on the size of the change and on whether you need to decide or to build.
08 / AI ACT
The European AI Act classifies uses by risk. These are the dates that matter today.
The AI practices listed in Article 5 are prohibited, such as social scoring or manipulation that exploits vulnerabilities.
Article 50 transparency obligations: anyone interacting with an AI system has to be told so, unless it’s obvious.
Requirements for Annex III high-risk systems (employment, credit, education, among others), after the delay introduced by the Digital Omnibus on AI.
Requirements for Annex I high-risk systems, those built into regulated products.
Sources: Regulation (EU) 2024/1689 and Regulation (EU) 2026/1744 (Digital Omnibus on AI). This isn’t legal advice: for your specific case, ask your adviser.
09 / EVIDENCE
An assistant evaluated against real data before anyone relied on it, and a public assistant whose cost and quality are measured.
Business intelligence
Feasibility was tested on the company’s real data, and the assistant was evaluated before it reached management.
STACK Nuxt / Vue / MySQL
Case study: AtalayaIQ
Editorial platform
A public AI assistant whose cost, latency and quality are measured conversation by conversation.
STACK Nuxt / Vue / MySQL
Case study: Rocio.com10 / QUESTIONS
Identifying which of a company’s processes improve with AI, checking feasibility on real data, assessing risk and cost, and taking the proposal to a working pilot. Good AI consulting also tells you where AI isn’t needed.
It depends on scope. Diagnosing and testing the feasibility of one specific process takes weeks, not months. The pilot depends on what has to be integrated.
A consultancy usually delivers analysis and recommendations; a developer builds what someone else has decided. My work joins the two: I decide what makes sense to build and I build it, so the recommendation and the result don’t drift apart.
It depends on the use. Most business uses, such as assistants or document automation, mainly carry transparency obligations. High-risk uses, such as recruitment or credit scoring, have much stricter requirements. The diagnosis checks which category each case falls into; for the legal interpretation, ask your adviser.
Then the consulting did its job. Often the real improvement is a rule, an integration or tidier data, and that’s cheaper and more reliable than a model.
When NOT to use artificial intelligenceWhichever best fits the task, the cost and the data terms. The provider is designed as a replaceable part of the system, not a hidden dependency.
11 / SAME SYSTEM
A business problem rarely lives in a single piece. These usually appear in the same project.
Tell me the process, the data you have and what worries you. I’ll tell you whether it’s worth a diagnosis, whether it can be solved without AI or where I’d start.