AI Use Cases by Industry: What's Really Different in Your Sector
Most lists of industrial AI use cases look the same whatever the sector: visual inspection, predictive maintenance, automated quoting. What they leave out is a way to tell which idea will hold up in your sector and which will stall on missing data or evidence requirements. This guide gives you that filter. It describes sector patterns, not success stories, and reflects the situation as of October 2026.
What makes AI use cases differ between industrial sectors?
Three things vary a lot between sectors: how the core process runs (one-off orders, series production or batches), what data that process produces, and the rules you must follow to show how a decision was made. The agent technology itself is largely the same everywhere. A use case fits when it passes all three tests for your sector.
This is a different cut from looking at departments. The guide to AI use cases by department slices a company horizontally: sales, service, purchasing and finance exist everywhere, and the patterns there look similar at a toolmaker and at a chemicals plant. The sector view slices vertically. It asks what is different in and around the core process, because that is where most ideas get stuck.
Take two agents. One reads customer enquiries and prepares a draft quote; it behaves much the same at a special-purpose machine builder and at a plastics moulder. The other assesses deviations on the shop floor; at a drug manufacturer it looks nothing like the version at a foundry. It needs different data, different sign-off and different records.
Adoption varies widely too. According to Eurostat, around 20% of EU enterprises with ten or more employees used at least one AI technology in 2025. The figure was 62.5% in information and communication and 10.8% in construction. These numbers tell you where companies are experimenting. They do not tell you which use case will pay off for you.
How does the sector filter work?
The filter is three questions you put to every idea. How does the process run in your sector? Is the data it needs readable and accessible? Which evidence and approval rules apply? An idea that fails any one of them goes back on the list for later.
Building block 1: process type
The way an order moves through the business decides where an agent can help.
- Engineer-to-order and small batches (special-purpose machinery, plant engineering, tooling): every order is a little different. Much of the effort sits in enquiries, specifications, design reviews and documentation, so agents mainly help with reading, comparing and drafting text.
- Series production (automotive suppliers, electrical equipment, plastics): the process is stable; deviations are the issue. Complaints, engineering change requests and supplier correspondence generate a steady stream of repetitive text.
- Batch and process manufacturing (chemicals, pharmaceuticals, food): recipes, batch records and release decisions are central, and every deviation has to be recorded and assessed.
- Project business (construction, building services, energy installations): many parties, many documents, changing sites. The bottleneck is often keeping track of drawings, change orders and meeting minutes.
Building block 2: data
Sectors produce different data. Machinery makers keep their knowledge in manuals, bills of materials, service reports and design-related documents. Series producers add inspection data, 8D reports and supplier files. In process industries the key information sits in batch documentation, specifications and laboratory systems. An agent can only work with what it is allowed to read.
Building block 3: regulation and evidence
The more regulated a process, the more weight falls on validation, logging and human approval. That does not rule AI out. It means the work before the first productive use is larger and lands in different places.
How do the main sectors compare?
The table below applies the three filters to common industrial sectors. Treat it as a starting point rather than a complete map; plenty of companies in each sector work differently. Check your own process against the row for your sector, not the other way round.
| Sector | Process type | Typical data | Regulatory frame | Obvious first step |
|---|---|---|---|---|
| Machinery and plant | Engineer-to-order | Enquiries, specs, manuals, service reports | Machinery Regulation, CE marking | Quotes, field service |
| Automotive suppliers | Series | Inspection data, complaints, customer requirements | Customer requirements, TISAX | Complaints, change management |
| Chemicals and pharma | Batch | Batch records, specifications, lab reports | GMP (pharma), Annex 11 | Document search outside GMP |
| Food and beverage | Batch, series | Recipes, hygiene records, supplier evidence | Food law | Supplier documents, specifications |
| Medical devices | Series with approval | Technical documentation, risk file | Medical Device Regulation | Documentation search |
Machinery and plant engineering
Every order is a small project. Enquiries arrive as PDFs, specifications or emails with a sketch attached. An agent can extract the requirements, compare them with past orders and prepare a draft, as described on the page about the quotation agent. The second common lever is field service: technicians search manuals, fault databases and old service reports, and a technical service agent can do that legwork first. A dedicated guide to AI agents for machinery manufacturers is in preparation.
Automotive suppliers
Series production means stable processes and demanding customer requirements. A lot of time goes into complaints, aligning engineering changes and working through customer-specific standards. Agents can structure incoming complaints, find similar past cases and draft reports. For information security, the automotive industry has TISAX, an assessment framework governed by the ENX Association and based on the ISA catalogue published by the German automotive association VDA. If your customers require a TISAX label, any agent that handles design or prototype data has to fit inside that security concept.
Chemicals and pharmaceuticals
In process industries, batches and recipes sit at the centre. For medicinal products, EU GMP Annex 11 requires computerised systems used in GMP activities to be validated, and states that replacing a manual operation should not increase the overall risk of the process. The same text recommends, based on risk, an audit trail of GMP-relevant changes and deletions. In 2025 the European Commission also put a draft Annex 22 on artificial intelligence out for consultation, which closed on 7 October 2025. A sensible first step is therefore work outside GMP activities, such as searching specifications or preparing supplier enquiries.
Food and beverage
Food manufacturers handle recipes, hygiene and cleaning records and a large volume of supplier evidence such as specifications and certificates. Much of the office work consists of requesting, checking and updating that evidence. It is text-heavy and repetitive, which makes it a natural fit for agents. Decisions about product safety stay with people.
Medical devices
Medical devices fall under the Medical Device Regulation (EU) 2017/745, which appears in Section A of Annex I to the AI Act. AI that serves as a safety component of a medical device, or is itself a device, can therefore count as high-risk. An agent that helps staff search the company's own technical documentation is not a medical device. Where exactly the line falls needs a legal assessment in each case.
What do you need in place, sector by sector?
Every sector needs the same foundation: a clearly scoped process, readable documents, a way for the agent to access them and a named person who approves the output. Depending on the sector, validation, safety evidence or customer requirements come on top. Budget for those before you pick a use case, not after.
Data and systems. Find out where the documents for the process live: in the ERP, a document management system, SharePoint, a laboratory system or shared mailboxes. The more systems involved, the more access rights matter. The page on the knowledge management agent explains how agents work with knowledge spread across systems.
Organisation. In regulated sectors, quality assurance and, where relevant, information security should be involved from the first shortlist. In less regulated sectors the business team and IT are often enough. Either way, each use case needs an owner who signs off results.
Where the model runs. If design data, recipes or prototype information are your most valuable assets, the location of the model matters more. Options where the model runs in your own data centre then come into play, as described under sovereign AI.
How should you run a sector-based use case search?
Work through five steps: collect ideas, assign the process type, check the data, check the rules, and pick one use case where value and effort are well balanced. Run the sector filter before you score value, because it removes weak ideas fastest.
- Collect ideas. Ask each team about repetitive text work: what gets read, retyped, compared or summarised again and again? Add the ideas from the department view.
- Assign the process type. Does the idea sit in the core process (design, production, batch) or at its edge (quoting, purchasing, service)? Ideas at the edge are usually quicker to deliver.
- Check the data. Pull ten real cases and ask whether a person could reach the result using only the documents that are accessible. If they could not, an agent will not either.
- Check the rules. Does the use case touch GMP-relevant data, a machine's safety functions, a medical device or customer information security requirements? If so, bring in quality assurance or legal advice early.
- Choose and test. Pick one use case and test it on real documents. A small test tells you more than any scoring matrix.
A word on sequencing. Many companies start with the most exciting idea in the core process, such as AI-assisted quality inspection. In regulated or safety-critical areas that is the case with the heaviest evidence burden. Starting at the edge of the core process builds experience with data, permissions and approvals that you will need later anyway.
What are the limits and risks?
The biggest risks come from sector rules that surface too late: an agent that turns out to touch GMP-relevant data, an AI feature that turns out to be a safety function, or customer security requirements that rule out the chosen hosting model. Language models also have their usual limits: they can be wrong and sound confident.
The EU AI Act and its revised deadlines. The AI Act has applied in general since 2 August 2026. The omnibus amendment, Regulation (EU) 2026/1744, which entered into force on 27 July 2026 according to Cooley, pushed back the high-risk obligations: to 2 December 2027 for systems under Annex III and to 2 August 2028 for systems under Annex I. Annex III includes safety components in the supply of water, gas, heating and electricity, which matters more to utilities and plant operators than to a typical manufacturer.
Machinery. The Machinery Regulation (EU) 2023/1230 becomes mandatory on 20 January 2027 and, according to the Commission, covers machinery with AI-powered safety functions. Under the omnibus amendment it has moved from Section A to Section B of Annex I to the AI Act, so requirements for AI in machinery are meant to come mainly through the Machinery Regulation itself. A quoting or service agent does not raise these questions; AI inside a safety function does. Treat this paragraph as orientation, not legal advice.
Practical limits. An agent is only as good as the documents it reads. Outdated manuals, conflicting specifications or missing revision levels lead to wrong drafts. In every sector the rule is the same: the agent prepares, a person decides, and the decision stays traceable.
When the sector filter does not help. If your bottleneck is a pure back-office task such as processing supplier invoices, the sector hardly matters. The department view is the better starting point.
Where should you start?
Start with a list of five to ten ideas and apply the three filters: process type, data, regulation. Pick the use case that passes all three with the least extra effort and test it on real documents from your own operations. The result will tell you faster than any study whether the approach holds.
For many manufacturers the first use case sits at the edge of the core process: quoting, field service or knowledge search. Data there is usually accessible, the rules are manageable, and mistakes surface before anything reaches a customer. The overview of AI agents shows how such agents are built and which packages exist. To decide which department goes first, the guide to AI use cases by department complements this sector view.
In our own operations one simple rule has held up: only once an agent has prepared ten real cases well enough that the experts check rather than rewrite them is it worth thinking about the next use case. That rule applies in every sector. What differs between sectors is the route to those ten cases.
Frequently asked questions
Are there AI use cases that work in every manufacturing sector?
Yes, mostly text-heavy tasks such as preparing quotes, triaging service requests or searching technical documentation. These depend more on the department than on the sector. Sector differences start to matter once AI touches production, quality or regulatory approval.
Does an AI agent at a machinery maker count as high-risk under the EU AI Act?
Not by default. An agent that drafts quotes or searches manuals is not a safety component of a machine. AI that performs a safety function is a different matter and needs a case-by-case legal assessment.
Why is pharmaceutical manufacturing a harder place to start?
Computerised systems used in GMP activities must be validated, and changes to GMP-relevant data should be traceable. Agents used outside GMP activities, for example in sales or procurement, are usually not affected.
Should you copy what your sector is doing with AI?
Only to a point. Adoption figures show where others are experimenting, not what will pay off in your plant. Volume, data access and the cost of an error in the specific process still decide.
Sources
- Eurostat: Use of artificial intelligence in enterprises (2025 data) ec.europa.eu
- AI Act Explorer: Article 113, entry into force and application (consolidated with the omnibus amendment) artificialintelligenceact.eu
- AI Act Explorer: Annex I, list of Union harmonisation legislation artificialintelligenceact.eu
- AI Act Explorer: Annex III, high-risk areas artificialintelligenceact.eu
- Cooley: Digital AI Omnibus delays key deadlines, introduces new rules cdp.cooley.com
- European Commission: Machinery (Machinery Regulation (EU) 2023/1230) single-market-economy.ec.europa.eu
- EudraLex Volume 4, Annex 11: Computerised Systems (revision 1) health.ec.europa.eu
- European Commission: consultation on GMP Chapter 4, Annex 11 and new Annex 22 (AI) health.ec.europa.eu
- ENX Association: TISAX enx.com