AI Adoption in Mid-Size and Large Companies: The First-Year Guide
Most digital transformation plans I've seen inside mid-size and large companies don't fail on the technology. They fail in the jump from the slide to the repository: someone approves an AI strategy, someone presents it to the board, and six months later there isn't a single line of code in production.
Lately, the conversation I keep having at Conectia isn't with startup founders. It's with executives at established companies — gyms, logistics, industrial property, insurance — who have just decided AI is no longer optional, and are discovering that the decision was the easy part.
This guide is what we tell them before we try to sell them anything: the numbers, the timelines and the mistakes, in that order.
Where we are: the gap is no longer conviction — it's execution
The Spanish data tell a very specific story. According to CaixaBank Research, AI adoption among companies with more than 10 employees has more than doubled between 2021 and 2025, from 8% to 21%. But the headline buries the part that matters: 3 out of 5 large companies already use AI, against 31% of mid-size companies and 18% of small ones.
Translation: if you run a mid-size company, half of your large competitors are already in production, and two thirds of the companies your size aren't yet. That is exactly the window where moving pays off most — and where standing still should no longer be an option.
And it isn't slowing down. Deloitte's State of AI in the Enterprise study finds that 85% of Spanish companies plan to increase their AI investment next year, and roughly a third expect increases above 20%.
The 2026 question is no longer «should we adopt AI?». It's «who builds it, how fast, and at what real cost?».
The three execution paths (and when each one makes sense)
When a mid-size or large company decides to execute, it has three roads. All three work. What rarely works is picking one for the wrong reason.
1. Build your own team
Hiring AI and data engineers onto payroll. This is the right road when AI is going to be a central, permanent capability of the business — not a project, a department.
The problem is the clock and the cost. As I laid out in the guide to hiring senior developers in Spain, a senior hiring process takes two to three months, and the total payroll cost lands at 1.4-1.5 times gross salary once you add employer contributions. With AI profiles, add a market where you're bidding against banks, consultancies and international remote offers for the same people.
None of that rules it out. What rules it out is doing it first: standing up the department before a single use case has reached production is building the garage before you know whether you need a car.
2. Bring in a large consultancy
The classic digital transformation route. It's the right one when the problem is organizational at scale: process redesign across twenty departments, change management for thousands of employees, a multi-year program with a steering committee.
Just be clear about what you're buying: the plan, the governance and the paperwork. The technical execution usually comes later, with a different team, at a different rate, and with a rotation of profiles you don't get to choose. If your problem is "I need this running in production this quarter", a large consultancy's cycle works against you.
3. Deploy a technical squad
A team of senior external engineers, embedded in your organization, building with you from week one. This is the right road when you already know what you want (or you have someone in-house who does — a CIO, a head of AI, a digital transformation director) and what you're missing are the hands to build it.
It's the model we run at Conectia, so here are the numbers you can hold us to: senior engineers at 16-22 €/h, senior leads at 22-28 €/h, first candidate presented within 72 hours, and an intake filter that lets through 3% of applicants. The MintID case study shows what that looks like on a real project.
The combination we see work best in mid-size companies isn't choosing one road: it's sequencing them. A deployed squad to reach production in the first quarter, your own hiring in parallel to internalize what already works, and staff augmentation to absorb whatever the company needs along the way.
Where AI projects go to die (the six patterns that repeat)
After enough conversations with companies halfway through their adoption, the failures all fit on this list.
1. Pilot purgatory. The demo worked, leadership was pleased, and eight months later the pilot is still a pilot. Nobody budgeted the road to production: legacy integration, security, monitoring, support. A pilot with no production plan isn't progress — it's internal marketing.
2. Buying a tool instead of fixing a process. Copilot and ChatGPT Enterprise licenses for the whole workforce, without redesigning a single workflow. I've written about what actually works with ChatGPT in companies: licenses without process quietly fade out.
3. Ignoring the state of your data. The use case was viable; the data feeding it lived in four systems that don't talk to each other and two private spreadsheets kept up to date by email. Half the real work of AI adoption is data engineering, and almost nobody puts it in the initial budget.
4. Letting mid-level profiles make architecture decisions. The same mistake as in any software project, amplified: in AI, a bad early call — model, platform, data schema — compounds. Expensive decisions need seniority, not hours on a calendar.
5. Not defining ROI before starting. If nobody wrote down which number has to move — hours saved, tickets resolved, margin per order — the project will be "interesting" forever and a priority never.
6. Leaving governance for the end. With the EU AI Act rolling out in phases, use cases that touch personal data, decisions about people or regulated sectors need the risk analysis before you build, not as an audit afterwards. Retrofitting compliance costs twice what designing it in does.
The first year, quarter by quarter
The realistic calendar for a mid-size or large company that's starting in earnest. It's not the most ambitious one — it's the one that survives contact with reality.
Quarter 1 — Diagnosis and a first win. Inventory your use cases, ranked by value and data viability — not by spectacle. Pick one — one — with measurable ROI and accessible data. Get it into production, even in a minimal version. That early win buys the internal credibility for everything that follows.
Quarter 2 — From one to three. With the first case in production, scale to two or three more, reusing what you've learned: the infrastructure, the integration patterns, the data pipeline. This is where the team that built the first one delivers the second at twice the speed.
Quarter 3 — Internalize. Decide which capabilities move onto payroll and hire them without the pressure of the launch phase, with the squad still running. A documented transition — handover written from day one — is the difference between internalizing and starting over.
Quarter 4 — Governance and measurement. Formalize the AI committee, the use-case register, the data policy and regulatory compliance. And publish the year's numbers internally: what was promised, what moved. Year-two budgets are won here.
Eight questions for any «AI transformation» vendor
The same ones I'd ask Conectia if I were on the other side of the table.
- How many of your AI projects are in production — not in pilot? Ask for named examples.
- Who will write the code: the people you're showing me, or a team I haven't met?
- What percentage of candidates passes your technical filter? How do you evaluate them?
- How soon is the first engineer working in my repository?
- What happens when a profile doesn't fit? Replacement in how long, at what cost?
- How do you document the handover so I can internalize the team without starting over?
- What share of the budget goes to data engineering? (If the answer is "none", they haven't done this before.)
- How do you handle the EU AI Act in the design? (If the answer is a slide, same conclusion.)
FAQ
Do we need a Chief AI Officer before we start?
You need someone with a mandate — a CIO, a head of AI, a digital transformation director with budget and the authority to prioritize. The title matters less than the power to decide. If you've just appointed that person and they have no team, we wrote a guide for exactly that scenario: your first CIO or AI lead, executing without a team.
What does a serious first AI adoption project cost?
A first use case in production, with a squad of two or three senior engineers over a quarter, sits in the 60,000-120,000 € range depending on integration and the state of your data. For reference: the total annual cost of a single senior AI engineer on payroll exceeds that figure — and takes three to six months to even start.
Isn't it cheaper to wait for the technology to mature?
The technology will keep changing; what builds slowly is your organization's capacity to absorb it. The companies that started in 2023 don't have better models than you — they have two years of ordered data, redesigned processes and in-house judgment. That is what licenses can't buy, and it depends, more than ever, on the senior software engineers everyone has spent two years trying to retire — who have never had more work.
Ready to go from strategy to production?
Talk to a technical partner and deploy CTO-vetted engineers in 72 hours.
