Why companies do not use AI (and how to unlock it)

Eurostat’s enterprise ICT survey (dataset isoc_eb_ai, updated 2026-06-15) shows that in 2025, 9.1% of Spanish enterprises with 10+ employees said they do not use AI because of a lack of expertise—the top-cited barrier, above the EU-27’s 7.8%. Still, 20.3% of firms in Spain already use at least one AI technology in 2025, close to the EU-27’s 20.0%.

Key takeaways

• In 2025, 9.1% of Spanish enterprises (10+ employees) cite lack of expertise for not using AI vs 7.8% in the EU-27 (Eurostat, isoc_eb_ai).
• Privacy (5.9%), legal uncertainty (5.7%) and data quality/availability (5.3%) are clustered; costs (5.2%) and system incompatibility (3.9%) are lower in Spain.
• These percentages are over all enterprises with 10+ employees, not only non-users of AI.
• AI adoption in Spain reaches 20.3% of enterprises in 2025, near the EU-27’s 20.0%.
• The expertise gap is highest in mid-sized firms in Spain (11.7%), and relevant in large (9.8%) and small (8.6%) firms too.

What the Eurostat figures really say (and how to read them)

Enterprises in Spain are moving forward with AI, yet they still report very concrete barriers. According to Eurostat’s survey on enterprise ICT use and e-commerce (dataset isoc_eb_ai, updated 2026-06-15), in 2025, 9.1% of enterprises with 10 or more persons employed in Spain say they do not use AI due to lack of relevant expertise. Next come privacy (5.9%), legal uncertainty (5.7%), data availability/quality (5.3%), costs (5.2%) and incompatibility with existing systems (3.9%). These percentages are calculated over all enterprises with 10+ employees (financial sector excluded), not only among non-users.

At the same time, adoption is rising: 20.3% of Spanish enterprises use at least one AI technology in 2025, very close to the EU average at 20.0%. The picture is mixed: about one in five firms already use AI, while a smaller but meaningful share flags specific obstacles that can be removed with a practical plan.

How to interpret these shares correctly? A 5–9% figure means that fraction of all enterprises reports that barrier; it does not mean a majority among non-users. For growth with control, the focus should be on removing operational friction: building skills, running bounded pilots, strengthening privacy and data protection, clarifying legal guardrails, and integrating without disruption.

Practical first step: map which of these six barriers is most material in your company today and tie it to a single business objective (e.g., faster proposal turnaround or quicker customer replies). That is how you turn a statistic into a concrete roadmap.

SpainEU-27

1. Lack of expertise9.1 %7.8 %
2. Costs too high5.2 %4.2 %
3. Legal uncertainty5.7 %5.9 %
4. Data availability/quality5.3 %4.8 %
5. Data protection and privacy5.9 %5.8 %
6. System incompatibility3.9 %4.6 %
7. Using at least one AI20.3 %20.0 %

Source: Eurostat (isoc_eb_ai)

If you need a structured outside view to prioritise barriers and value, see our AI consulting approach to assessment and roadmapping in AI consulting for businesses: assessment and roadmap.

Barriers to not using AI and adoption in 2025 (enterprises with 10+ employees)

Indicator	Spain (10+ · 2025)	EU-27 (10+ · 2025)
Lack of expertise	9.1% (2025)	7.8% (2025)
Costs seem too high	5.2% (2025)	4.2% (2025)
Legal uncertainty	5.7% (2025)	5.9% (2025)
Data availability/quality issues	5.3% (2025)	4.8% (2025)
Data protection and privacy concerns	5.9% (2025)	5.8% (2025)
Incompatibility with existing systems	3.9% (2025)	4.6% (2025)
Using at least one AI technology	20.3% (2025)	20.0% (2025)

Lack of expertise: Spain’s top reason not to use AI

Lack of relevant expertise is the most-cited barrier in Spain: 9.1% vs 7.8% in the EU-27 (2025). By firm size, the pattern is clear: mid-sized firms (50–249) declare the largest gap in Spain (11.7% vs 10.5% in the EU), followed by large (9.8% Spain; 10.3% EU) and small (8.6% Spain; 7.2% EU). Spain’s figure has also risen from 2023 (4.0%) and 2024 (5.4%) to 2025 (9.1%), suggesting more firms are seriously assessing AI and discovering capability gaps.

30–60 day capability boost plan:

• Appoint an AI Champion per function to source use cases and circulate good practices.
• Run role-based training (sales, operations, marketing, finance): 90-minute sessions plus guided practice.
• Standardise prompts, human-in-the-loop checklists and responsible-use policies.
• Open an internal Q&A channel and a repository of approved examples.

Practical example: a mid-sized industrial company in Zaragoza (Spain) can upskill its technical office first on drafting documentation and summarising incident logs, with a fortnightly quality review by a senior engineer. In parallel, the sales team iterates bid templates with prompt libraries.

To accelerate this with structured upskilling designed for teams, see AI and Claude training for businesses. ARTIC supports organisations that want to start safely with clear business cases and hands-on practice.

1. Germany10.8 %
2. Portugal9.2 %
3. Spain9.1 %
4. Netherlands7.8 %
5. EU-277.8 %
6. Italy5.5 %

Source: Eurostat (isoc_eb_ai), 2025

SpainEU-27

1. 10–49 staff8.6 %7.2 %
2. 50–249 staff11.7 %10.5 %
3. 250+ staff9.8 %10.3 %
4. All (10+)9.1 %7.8 %

Source: Eurostat (isoc_eb_ai), 2025

Perceived costs: funding value while capping risk

In 2025, 5.2% of enterprises in Spain say AI “seems too costly,” above 4.2% in the EU-27. The concern is stronger among mid-sized firms (7.5% in Spain), where cross-department projects emerge under limited budgets. Spain’s recent evolution (3.3% in 2023; 3.2% in 2024; 5.2% in 2025) suggests that as teams identify use cases, questions arise on licensing, integration or internal effort.

How to dismantle the cost barrier without big upfronts:

• Define an 8–12 week pilot with one business KPI (e.g., cycle time for one task) and minimal technical scope.
• Prioritise tools that offer controlled trials and enterprise options to ease future scaling.
• Measure hours saved, errors avoided or response times improved with a before/after baseline.
• Reuse existing infrastructure and data; avoid custom builds until fit and value are proven.

Practical example: a small B2B distributor in the Madrid region can launch an internal assistant to answer catalogue and logistics FAQs using existing documentation. The pilot tracks the share of queries resolved without escalation and time saved per order.

If you need a crisp, value-first roadmap to pick low-cost/high-impact cases and justify budgets, explore our approach in AI consulting for businesses: assessment and roadmap. A milestone-based plan helps defend investment and reduces risk.

Legal uncertainty: control and traceability from day one

Legal uncertainty is cited by 5.7% of enterprises in Spain (5.9% in the EU-27) in 2025. The figure is fairly even by size, with a slightly lower incidence among large firms in Spain vs the EU. The upward trend vs 2023 and 2024 mirrors broader experimentation: as more potential use appears, teams need simple, clear rules.

A practical way to keep work flowing without legal gridlock:

• Short internal usage policy: what data can/cannot be used, mandatory human review, and a lightweight log of prompts/outputs when needed.
• Light traceability: keep representative samples of inputs/outputs for critical cases to audit decisions and improve processes.
• Clear roles: who approves use cases, who reviews risks, and who maintains templates and guidance.
• Vendors and contracts: centralise tool ownership, document configurations and permissions.

Practical example: a professional services firm in Bilbao (Spain) can roll out a two-page guide with permitted/restricted scenarios, a simple intake form to register new use cases and a monthly sample review by a joint operations–legal committee.

Enterprise editions help configure permissions, limits and data isolation at workspace level. See how to plan it in Claude for business: Claude Team and Enterprise deployment. If your website or ecommerce is part of the data flow, align policies and public texts with a review of Data protection for websites and ecommerce.

Data availability and quality: the silent bottleneck

Data availability/quality difficulties affect 5.3% of enterprises in Spain (4.8% in the EU-27) in 2025. It is not the loudest barrier, but often the most decisive: without located, current and understandable data, pilots stall or yield uneven results. In mid-sized and large firms, the challenge is as much about locating sources as agreeing shared definitions across functions.

Four-week “good-enough data” checklist:

• Inventory: list the 10–20 sources behind the use case (documents, databases, spreadsheets, CRM, etc.).
• Access: define who needs read access to each source and how they access it (avoid shared credentials).
• Minimum quality: identify key fields and simple rules (format, required fields, duplicates) and assemble a small training/test batch.
• Maintenance: set a cadence to refresh and version the datasets used by AI.

Practical example: a mid-sized ecommerce in Barcelona (Spain) can start with a customer service answer selector powered by help articles and shipping policies. Structure the articles into categories, remove duplicates and point each topic to a single “official” source.

If you need light engineering to connect sources, automate cleaning or build internal AI helpers without replacing core systems, see how we approach this in AI software development with Claude Code.

Privacy and security: clear rules and controlled environments

Concerns about data protection and privacy are cited by 5.9% of enterprises in Spain (5.8% in the EU-27) in 2025. This is a cross-cutting concern that rises when teams begin using AI with sensitive documents or internal contexts.

Lower risk without blocking innovation:

• Sensitive data: label what must never be uploaded to external tools and define alternatives (summaries without personal data, templates with placeholders, anonymisation).
• Controlled environments: use enterprise workspaces with team-based permissions and basic activity logs when the case warrants it.
• Human review: establish checkpoints before sending content to clients or publishing AI-generated pieces.
• “First-aid” privacy training: short sessions to recognise common risks and how to avoid them.

Practical example: a technology consultancy in Valencia (Spain) can set a proposal-prep flow: a standardised template, a library of pre-approved paragraphs and a mandatory reviewer sign-off before sending. The team learns what to include, omit or summarise.

For help configuring workspaces, permissions and responsible-use guidelines from the outset, you can lean on ARTIC, Claude Partner, which supports deployments aligned with enterprise controls.

Compatibility with existing systems: integrate without rebuilding

Incompatibility with existing equipment, software or systems is cited by 3.9% of enterprises in Spain (4.6% in the EU-27) in 2025. Although the smallest measured barrier, it can still block pilots if teams try to “replace” core platforms instead of complementing them.

Good practices to integrate AI into your current stack:

• Start “on top” of critical systems: assistants reading exported documentation, standard APIs or connectors that avoid deep changes.
• Automate peripheral tasks: drafting, summarisation, classification, data enrichment or contextual search over already-governed content.
• Open standards: prioritise widely supported formats and protocols to avoid lock-in.
• Security and permissions: ensure the integration respects existing controls and least-privilege principles.

Practical example: a logistics company in Málaga (Spain) can implement automatic incident summaries from the current ticketing system without touching the core app. AI produces drafts that get validated in the usual workflow and archived with the ticket.

When integration needs custom development or specific connectors, a pragmatic engineering approach helps avoid long projects until value is proven. See how we build with AI in AI software development with Claude Code.

From barrier to rollout: a 90-day plan (with examples)

With the data in hand, the key is to turn each barrier into an action. Example 90-day plan for an industrial SME in Valencia (Spain, 50–120 people):

• Weeks 1–2: quick use-case assessment by function and risk map (focus on training and data).
• Weeks 3–4: role-based training and AI Champions appointed. Publish responsible-use guidelines and prompt templates.
• Weeks 5–8: one pilot with a single KPI (e.g., reduce offer-prep time). Define datasets, review steps and usage limits.
• Weeks 9–10: before/after evaluation (quality, time, internal satisfaction) and decide to scale or iterate.
• Weeks 11–12: controlled scale-up, documentation, light integration and governance formalisation (roles, policies, review cadence).

Operational tips:

• Communicate progress with real examples; celebrate learnings and record best practices.
• Keep a living backlog of cases prioritised by impact and complexity.
• Avoid “compliance paralysis” with clear rules and risk-proportionate reviews.

If you want expert support on diagnosis and roadmapping, ARTIC details how we work in AI consulting for businesses: assessment and roadmap. For deployments that require enterprise controls, permissions and isolation out of the box, see Claude for business: Claude Team and Enterprise deployment.

Frequently asked questions

What is the main reason companies in Spain do not use AI?

In 2025, the most-cited reason is lack of expertise: 9.1% of Spanish enterprises with 10+ employees say so. Source: Eurostat enterprise ICT survey (dataset isoc_eb_ai, updated 2026-06-15).

Are these percentages based on all companies or only on non-users of AI?

They are calculated over all enterprises with 10 or more employees (financial sector excluded). They do not represent the share within the group of non-users but within the entire business population covered by Eurostat.

Is Spain far behind the EU in AI adoption?

No. In 2025, 20.3% of enterprises in Spain use at least one AI technology, very close to the EU-27’s 20.0%. Adoption is advancing alongside the need to clear specific barriers.

Which company size most often reports ‘lack of expertise’?

In 2025, mid-sized firms (50–249 people) report it most in Spain (11.7%), followed by large (9.8%) and small (8.6%) firms. The EU-27 shows a similar pattern (10.5%, 10.3% and 7.2%, respectively).

Which other barriers weigh almost as much as lack of expertise?

Privacy (5.9%), legal uncertainty (5.7%) and data availability/quality (5.3%) form a close cluster. Costs (5.2%) and incompatibility (3.9%) are slightly lower in Spain (2025).

How have these barriers evolved in Spain since 2023?

Several have increased: lack of expertise rose from 4.0% (2023) to 5.4% (2024) and 9.1% (2025). Costs, legal uncertainty, data and privacy concerns, and incompatibility also grew, reflecting broader exploration of AI and rising awareness of prerequisites.

Where should a company start to overcome barriers with limited resources?

Run an 8–12 week pilot tied to one KPI, deliver role-based training, publish responsible-use rules and prepare a minimal data inventory. Measure before/after and document learnings to scale safely. ARTIC can help with assessment and a practical roadmap.

Sources and methodology

• Eurostat, survey on ICT usage and e-commerce in enterprises: companies with 10 or more persons employed, excluding the financial sector (database).
• The linked client examples are real ARTIC projects. Calculators are indicative.