Nalum · Report Spain & Mexico · 2026
— AI Strategy

AI adoption in companies:
from experimentation
to strategic chaos

Over the past 24 months, the adoption of artificial intelligence in companies has grown exponentially.

01 Introduction

Just two years ago, only 7-8% of large companies used AI in any form. Today, that number has jumped to 40-50%. It's unprecedented growth.

But here's the problem: that growth hasn't come with strategy. Companies are experimenting. They're using AI because they know it matters, but they don't really know how, why, or where they're heading. Their employees use ChatGPT, Claude, Copilot. They do it in isolation, without governance, without measurement, without a plan. It's organized chaos.

Meanwhile, in the United States, hundreds of companies —from startups to large traditional corporations, and not just the tech ones with their own models— are already several steps ahead. They've moved from experimentation to strategic implementation. And the result is visible: optimized processes, lower costs, layoffs because automation makes certain roles unnecessary. This isn't fiction. It's happening now.

The question that matters: where are you on this spectrum? Are you experimenting, or do you have a strategy?

And more importantly: do you have time to course-correct before it's too late?

02 The current state

What companies do today with AI

In most Spanish and Mexican companies, the answer to how they use AI is almost always the same: "our employees use tools like ChatGPT". That's it. No more structure. No more plan.

The reality is that, of the 40-50% of large companies that claim to be using AI, the distribution is as follows:

Prompting · basic stage 85%
RAG · exploring 10%
Fine-tuning / local models · exceptional 5%

The 85% is at the most basic stage: prompting. Employees write questions and get answers. It's manual, improvised, with no integration into systems. Each department does its own thing. Accounting uses AI to generate reports; marketing to draft content; HR to analyze candidates. No one talks to anyone. No data governance, no policies, no impact measurement.

The 10% is exploring RAG, which means connecting AI to their own documents and databases. Still experimental, still slow to implement.

The 5%, very few companies, is doing fine-tuning or local models. These are exceptional cases with advanced technical teams.

The result: chaos. Questionable security. Unknown ROI. Regulatory risk left unassessed. And worst of all: while this happens in Spain and Mexico, in the United States hundreds of companies are already at more advanced stages, seeing concrete results in efficiency and cost.

03 The problems we find

Four obstacles that show up every time

First, lack of governance. Employees use AI without clear policies. What data can they upload to ChatGPT? Who monitors usage? Are there security risks? Nobody knows. Confidential documents end up in public prompts. Intellectual property exposed. Client data processed without control.

Second, disconnection from business strategy. AI is used as an individual tool, not as a lever for transformation. An employee uses it to save an hour on a document. Fine. But where's the impact on the business? How much money are you really saving? What's the ROI? Without measurement, you don't know whether this matters or not.

Third, fear of replacing employees. Companies know AI automates tasks. Does that mean layoffs? That question paralyzes. The answer is nuanced: it doesn't replace, it transforms. Some roles disappear, others evolve, new ones appear. But yes, there is headcount reduction in companies that do it well. It's a structural change, not an apocalypse.

Fourth, lack of clarity on where to start. Do we need our own model? Should we use APIs? RAG or fine-tuning? The options are overwhelming. And here comes the most critical part: the companies that have historically outsourced all their technology infrastructure and development to external firms are the ones with the biggest problem now. They believe they need to follow that same model, outsourcing to external vendors to implement AI. The result: expensive, slow projects and external dependency.

A small internal product and technology team, working in an agile way and with the right leadership and guidance, can manage this entire AI evolution far more efficiently, more cheaply and more nimbly.

With the right direction, that small team is infinitely more productive than any external vendor. The tools exist, they're accessible, and they don't require massive infrastructure. The companies that get this save tens of thousands of euros and gain implementation speed.

04 The three stages

Strategic adoption, in phases

There's a clear path to implement AI in your company without chaos or unnecessary risk. But before taking the first step, you need to do something most companies skip: rethink your organizational structure. The three stages that follow only work if your company is ready to change.

You can't implement AI on a rigid structure designed years ago for a world without automation. You need a deep structural analysis: how is your company organized today? Which roles will change? Which teams will need new skills? Where is there redundancy that AI can eliminate? How will humans and machines collaborate? Only once you're clear on this can you move forward with the three stages. They're progressive: you don't need to skip any, and each builds on the previous one.

Stage 1
Prompting
The most basic and the fastest. Well-designed instructions to the model, a concrete answer. No complicated infrastructure. It can be employees using the web app manually, or it can be scaled with APIs integrated into your systems for automation. This is where almost every company starts. And that's fine: low risk, low cost, immediate result.
85% of companies using AI
Stage 2
RAG · Retrieval-Augmented Generation
Here you connect the model to your documents, your database, your knowledge base. The model queries your real data and answers based on it. You don't retrain anything; the model just retrieves. Ideal when you have your own documentation and want precise answers based on your data. You scale without retraining costs.
10% of companies are exploring it
Stage 3
Fine-tuning
Here you do train the model with your specific data so it behaves exactly as you need. It's more expensive, slower, and requires more infrastructure. But the result is a model fully customized to your business.
5% of companies are here

The key message: start with prompting, move to RAG when needed, and only consider fine-tuning when you truly justify it. But most important: adapt your organizational structure in parallel with these stages. Companies that do this see results from day one. Those that implement AI without changing their internal structure stay just as inefficient, only with more expensive tools.

05 Where to start

A contained pilot, without risk

The question everyone asks is: where do I start? The answer is: with a contained pilot. Don't transform your whole company at once. Pick a small, specific process with a clear problem. For example: automating responses to customer queries, automatically analyzing employee feedback, or generating operational reports without manual intervention. Something measurable in weeks, not months.

The pilot has three goals. First, to validate that AI works for your specific use case. Second, to train your team and shift mindsets. Third, to generate data and results that justify future investment.

A well-run pilot doesn't cost much. It can run on pure prompting, without complex infrastructure. And in 4-8 weeks you have answers: does it work? Is it worth scaling? What did we learn? With that data, you decide whether to move to RAG, scale to more processes, or adjust the strategy.

Low risk, fast result, and clarity for the next steps. The companies that win are the ones that start today, even small. The ones waiting for a perfect plan lose time others are already using.

06 Final reflection

Questions for your company

In 24 months, AI adoption in companies has grown from 8% to 40-50%. In two more years, probably 70-80% of companies will be using AI in some form. But using AI isn't the same as using it well. The companies that win will be the ones that start now with strategy, clear structure and a medium-term vision. Those that arrive late or do it chaotically will lose competitive advantage.

Before we close, these questions are for you:

01Where is your company today on this spectrum? Are you experimenting with AI without governance, or do you have a clear strategy?
02Do your employees use AI without policies, or is there control?
03Do you know your competitive risk in 18 months if you don't act?
04Is your organizational structure ready to change, or is it still rigid?
05Do you have a small internal team that can lead this, or do you believe you need to outsource everything?
06Have you identified a contained pilot to start with, or will you keep waiting for the perfect moment?

The answers to these questions will say a lot about whether you're ready or not.

Want to save it or share it with your team?
Download as PDF
— Nalum

If your company wants to transform and you're not sure of the best path, write to us.

We're founders, not career consultants. We've built and sold tech companies, and we work from that experience: with our own judgment, honesty, and the candor of someone who's been in your seat.

We work in
Spain · Portugal · LATAM
Email
Web
Nalum · AI Strategy Report · Spain & Mexico · 2026