Edition 1 · July 2026
Implementa Report · AI Operations
The State of AI Operations in Spain 2026
Spain already uses AI. Almost nobody operates it. This report measures that gap —the one between having AI and AI actually working— and puts a name, a structure and numbers on the function that closes it.
In one line
Adopting AI stopped being the problem. Operating it is. Spain plugs in models at a good pace, but the value stays in the pilot: the tech isn't what fails, the function that puts it into daily work is missing. That's the job of AI Operations, and almost nobody has it set up today.
1 · The picture
Spain is adopting AI fast
It's not a future promise; it's happening, and accelerating. In Q1 2025, 21.1% of Spanish companies with 10+ employees used AI —8.7 points more than a year earlier—. Services lead (25.7%), followed by industry (17.5%) and construction (11.4%).
The talent market confirms it: AI job postings have multiplied by nine in five years, and the profiles who master it earn up to 25% more. Spending intent follows: 85% of companies have already invested or plan to invest in AI.
The takeaway is simple: adoption is not the bottleneck. The money, the tools and the will are already there.
2 · The problem
Using AI is not operating it
Using AI and integrating it into the operation are two different things, and Spain is stuck between them. Most companies use AI for isolated tasks —marketing, admin, sales—, but integration into the core (production, logistics, product) advances slowly; roughly eight in ten companies still don't use it regularly.
And the figure that makes it clear isn't Spanish, it's global —but it's paid the same in Spain: 95% of organizations see no measurable return on their P&L despite $30–40 billion in spend. Only 5% of custom AI tools reach production.
What matters is why it fails, and MIT is blunt: the differentiator is not the model, the infrastructure or regulation. It's that most systems don't retain feedback, adapt to context or improve over time. The bottleneck is embedding behaviour into workflows and process orchestration.
Translated: the 95% seeing no return don't have an AI problem. They have an operations problem. They bought the engine and never built the car.
3 · The function
What AI Operations is —the function that closes the gap
If the problem is operating, the answer is a function, not another tool. AI Operations is the discipline that moves AI from pilot to daily work and keeps it running: identify which processes to operate, deploy on the real systems, run it with an owner and rules, measure what it resolves and what escalates, and improve with the data it generates.
It's not a lone role or an "AI expert" you hire and you're done. It's a team with a mandate: someone who leads, someone who automates, someone who trains the rest, and governance that decides what AI runs on its own and what escalates to a person. Without an owner and metrics, a company's AI is a cost center with a new name.
The gap between "we use AI" and "AI works" now has, for the first time, a function name and an org chart. It's what the 5% extracting value have set up and the 95% don't.
4 · The map
The AI Operations roles and their compensation
The function is made of concrete roles, with their own responsibilities and salary bands. Implementa keeps the map live —definition, skills and market salary by country— and updates it with every data point the community contributes.
AI Operations Manager
Turns pilots into daily operation; decides what runs, at what priority and how it's measured.
Head of AI Operations
The mandate at leadership level: budget, governance and the function's results.
AI Enablement Lead
Trains the rest of the organization to use AI well and safely.
AI Automation Specialist
Builds the automations that remove mechanical work.
AI Workflow / Agentic AI Engineer
Orchestrate workflows and agents that execute end to end.
AI Governance Lead
Defines limits, control and compliance for what AI is allowed to do.
AI Solutions Architect
Designs how it all fits on the company's real stack.
5 · Methodology
Why you can trust the number
A hard rule, the same one that governs all Implementa data: no figure without a source. Every salary data point carries its origin and date, and every band carries a visible label: "public-sources estimate" (job postings and public reports) or "community data" (real salaries contributed anonymously in the calculator).
A band moves from public estimate to community data when it accumulates enough real contributions. An estimate is never published as if it were observed data. That transition —from public to proprietary, labelled in plain sight— is what makes this report more precise each edition.
This Edition 1 starts mostly from public sources: it's the baseline. The proprietary layer is being built now, contribution by contribution. Cadence: annual.
What to do with this
Shall we set it up in your company?
Spain has the adoption. It's missing the operation. Measure before you hire, and build the function with an owner and metrics.
Sources
- INE — Encuesta sobre el uso de TIC y comercio electrónico en las empresas, datos 1T 2025
- MIT NANDA — State of AI in Business 2025
- Randstad — Las ofertas de empleo con IA se multiplican por nueve en España (2025)
- KPMG — Perspectivas España 2025
- Solunion — Más empresas la utilizan, pocas la integran (análisis sobre datos INE)