AI use cases by sector: real examples for SMEs

·

The question we hear most often at Summum IA since 2007 has not changed in form, only in protagonist: «What does this actually do for my business?». Fifteen years ago it was the ERP; five years ago, the cloud. Today it is artificial intelligence. And the answer remains the same: it depends on what problem you have and what sector you operate in. This article goes through the AI use cases that are generating measurable results in Spanish SMEs during 2025 and 2026, sector by sector, without empty promises or unnecessary jargon.

Why sector matters more than technology

AI is not a single tool: it is a family of techniques (language models, computer vision, numerical prediction, workflow automation) applied in very different ways depending on context. A defect-detection model on a manufacturing line has nothing in common with a copilot that drafts legal reports, even if both carry the label «AI».

That is why the right question is not «what can AI do?», but «what specific problem in my sector can AI solve with the data I already have?». Below we break down the most active sectors in real deployments during this cycle.

Manufacturing: computer vision and predictive maintenance

The manufacturing sector is, globally, the sector with the highest demonstrated return on AI projects. Three use cases concentrate the bulk of real investment:

Logistics and transport: routes, forecasting and document management

The logistics sector operates on tight margins where every extra kilometre and every hour of waiting counts. The most mature use cases for transport and distribution SMEs are:

Professional services: the copilot as a capacity multiplier

Law firms, accounting practices, tax advisories and HR consultancies share a common bottleneck: the time their professionals spend on repetitive, high-volume, low-value tasks (contract review, case-law research, drafting standard documents, document classification). AI does not replace the lawyer or the advisor; it amplifies their capacity so they can focus their time on decisions that genuinely require human judgement.

The most widely deployed use cases in 2025-2026 in this segment are:

If you manage a law firm or advisory practice, the most efficient starting point is our copilot service for professional practices, designed specifically for this business profile.

Hospitality and food service: demand, bookings and operations

Hospitality operates with perishable products, variable staffing and demand heavily tied to external factors (weather, events, seasonality). The use cases with the greatest real impact are:

Retail: personalisation and stock management

Physical retail and e-commerce share the challenge of managing a wide product range with limited resources. AI adds value on three fronts:

Healthcare and clinics: clinical documentation and triage

In the private healthcare sector (dental clinics, physiotherapy centres, aesthetic clinics), the most widespread use cases in 2025 are automatic consultation transcription (the doctor speaks; the system generates a draft clinical record) and smart appointment reminders that reduce no-shows. The processing of health data is subject to Article 9 of the GDPR (special categories) and requires additional safeguards; any deployment must be validated with the clinic's DPO or data protection officer.

Comparative table: use cases by sector and maturity level

Sector Primary use case Core AI technology SME maturity Typical reference data point
Manufacturing Visual quality inspection Computer vision (CV) High Very high detection rates under controlled conditions (recent European industrial deployments)
Manufacturing Predictive maintenance Time series + IoT Medium-high Prediction window 48-168 h before failure
Logistics Route optimisation Optimisation algorithms High Viable from fleets of 10 vehicles
Logistics Data extraction from documents Intelligent OCR / LLM High Eliminates manual entry from CMRs and delivery notes
Professional services Contract review and summarisation LLM + RAG High Hours saved per contract in due diligence
Hospitality Occupancy prediction Demand forecasting Medium Reduces food waste and adjusts staffing
Retail Automatic replenishment Demand forecasting Medium-high Reduces stockouts for fast-moving references
Private healthcare Consultation transcription ASR + LLM Emerging Requires GDPR Art. 9 compliance (health data)

What successful projects have in common

After accompanying AI deployments in SMEs across multiple sectors, we have identified four factors that separate projects with demonstrated ROI from pilots that never scale:

  1. The data exists and has minimum quality. AI does not create data; it processes it. An order history with inconsistent fields or a half-filled CRM produces useless predictions. The first step is always to audit the quality of the available data.
  2. The use case solves a concrete business problem, not a technology curiosity. «I want to use AI» is not enough. «I want to reduce contract review time in our audit process from 4 hours to 30 minutes» is.
  3. There is an internal project owner. Without someone inside the company who understands the use case, validates results and manages change with the team, the project does not land.
  4. Start small and scale with evidence. Projects that try to transform the entire company at once fail. Those that choose one process, automate it, measure the result and then replicate the pattern have far higher success rates.

The regulatory framework already in force: the European AI Act

Regulation (EU) 2024/1689 on Artificial Intelligence (AI Act) entered into force on 1 August 2024 and is directly applicable EU law in Spain, with a phased implementation schedule. The most relevant part for SMEs in the near term is not the high-risk systems (which affect sectors such as credit, employment or critical infrastructure), but the transparency obligations under Article 50, which will fully apply from 2 August 2026 to any system that interacts with people and could be mistaken for a human (chatbots, voice agents). Any deployment must document the system's use, inform the user when they are interacting with AI and not use personal data without a valid legal basis under the GDPR.

For SMEs using general-purpose AI tools (models such as those integrated into productivity copilots), the main obligation is responsible use: not using AI to make automated decisions that significantly affect people without human oversight, and respecting the rights of data subjects. Our sister company Summum Consultoría handles full legal advisory on the AI Act for businesses; from Summum IA, we cover the technical side of governance and deployment.

How to prioritise the first use case in your company

The method we apply at Summum IA to identify the highest-potential use case for an SME follows a structured process:

  1. Inventory of high-volume, low-variance processes. Repetitive, predictable processes with historical data are the best candidates. Examples: invoicing, email classification, in-line quality control, generation of periodic reports.
  2. Costing the current process. Person-hours per week × hourly cost. This gives the maximum possible ROI ceiling of the automated alternative.
  3. Technical feasibility assessment. Does the data exist? Is it digitalised? Is there an API or programmatic access to the system where it lives? How long would integration take?
  4. Select the case with the best value-to-complexity ratio as the first project. Not the most spectacular, but the fastest to demonstrate a result.

We carry out this process within our AI advisory service, where in a few weeks we identify the three or four use cases with the highest potential for your business and design the deployment roadmap.

Frequently asked questions

Which sector has the most mature AI use cases for SMEs?

Manufacturing and logistics concentrate the greatest number of deployments with demonstrated ROI, mainly because structured data (sensors, orders, routes) has been digitalised for years. Professional services are the service sector with the fastest adoption in 2025-2026, driven by large language models that allow working on text without requiring large proprietary datasets. Hospitality and retail are progressing at a medium pace, limited in some cases by the quality of available historical data.

Do I need a lot of data to start with AI?

It depends on the use case. For custom predictive models (demand forecasting, predictive maintenance) a history of at least 12 to 24 months with reasonable coverage of seasonal variations is recommended. For use cases based on pre-trained language models (copilots, document extraction, internal chatbots), the volume of proprietary data required is much lower: the model comes pre-trained and only needs your documentation to contextualise itself. This second approach is what allows small SMEs to start extracting value from AI with a low initial investment.

Will AI replace jobs in my company?

In the vast majority of SME deployments, AI eliminates tasks, not roles. A professional who spends four hours a week entering invoice data into the ERP will spend those four hours on higher-value tasks, not become redundant. The sectors where substitution is most real are those with highly repetitive tasks and little variation, such as high-volume manual data entry. However, change management with the team is a critical success factor: the deployments that work are those that involve the affected people from the very beginning of the system design.

What about GDPR when using AI with customer data?

The use of personal data in AI systems is subject to Regulation (EU) 2016/679 (GDPR) and, for the specific aspects of automated systems, also to the AI Act. The key principles are: specified purpose (not using data for purposes other than those declared when collecting it), minimisation (using only the data necessary for the use case) and a valid legal basis (consent, legitimate interest or contract, as appropriate). If the system makes automated decisions that significantly affect people (credit, staff selection, personalised pricing), Article 22 of the GDPR applies, requiring additional safeguards. When in doubt, consulting the DPO or a data protection advisor before launching the project is the most prudent approach and the one that avoids subsequent fines.