AI use cases for SMEs: 12 examples with benefits
AI use cases in small and mid-sized companies are mostly about text: according to the Eurostat survey for 2025, 26 percent of enterprises in Germany with at least ten employees use artificial intelligence, most often to analyse written text and to generate images, video or audio (13.5 percent each). Small firms with 10 to 49 employees stand at 23 percent, mid-sized firms with 50 to 249 at 36 percent, large firms at 57 percent. The gap is an opportunity: most worthwhile AI use cases need no research of their own, just clean data, a clear process and a tool that fits existing software. This guide shows twelve use cases by department with effort and benefit, a suitability check and a worked example. As of: November 2026.
AI use cases in numbers: where SMEs stand
Official figures show what companies actually use AI for. The values come from the European survey on ICT usage in enterprises for Germany, reference year 2025, enterprises with ten or more employees excluding agriculture, forestry, fishing, mining and the financial sector.
| Purpose | Share of enterprises |
|---|---|
| Marketing or sales | 9.1 percent |
| Organising business administration or management | 7.2 percent |
| Accounting, controlling or finance | 7.1 percent |
| Production processes | 5.5 percent |
| ICT security | 5.4 percent |
| Research, development or innovation | 3.3 percent |
| Logistics | 1.2 percent |
Among the reasons against AI, lack of expertise comes first (10.8 percent of all enterprises), followed by unclear legal consequences (9.3 percent) and data protection concerns (9.0 percent). Only 4.8 percent cite high costs. That argues for starting with well-understood cases whose legal position is clear.
AI use cases by department: twelve examples with effort and benefit
The table groups typical cases by department. Effort refers to a first pilot in a mid-sized company using existing standard software: low means a few weeks without building interfaces, medium means one or two interfaces, high means custom training or sensors.
| Department | AI use case | Data | Effort | Benefit measured as |
|---|---|---|---|---|
| Customer service | Sort emails and prepare draft replies | Mailbox, knowledge articles | low | handling time per request |
| Customer service | Phone assistant for standard questions and bookings | Phone system, calendar | medium | calls answered outside hours |
| Sales | Prepare quotes from request and price list | ERP, templates | medium | time to quote |
| Marketing | Product descriptions and translations | Item master data | low | texts per week |
| Accounting | Read incoming invoices and pre-assign accounts | Invoices, chart of accounts | low | documents per hour |
| Controlling | Explain variances in the monthly report | Financial accounting | medium | days to report |
| Purchasing | Compare supplier offers | Offers as PDF | low | time per comparison |
| HR | Draft job ads and replies to applicants | Templates, pay scale | low | time to reply |
| Administration | Meeting minutes and intranet knowledge search | Documents, recordings | low | search time, rework |
| Production | Visual inspection by camera | Images of good and faulty parts | high | scrap, complaints |
| Maintenance | Predict maintenance needs from machine data | Sensor data, fault history | high | unplanned downtime |
| Logistics | Plan demand and stock | Sales history, lead times | medium | stock level, shortages |
The low-effort cases share one trait: they work with text and documents that already exist. How to automate such a process step by step is shown in our guides on process automation for SMEs and AI invoice processing.
Assessing AI use cases: the five-question suitability check
Before money goes into a project, a sober assessment pays off. Give each question zero to two points. From seven out of ten points a case is a good pilot candidate.
| Question | 0 points | 2 points |
|---|---|---|
| How often does the task occur? | a few times a month | daily, many cases |
| Is the data digital? | on paper or scattered | in one system, consistent |
| How costly is an error? | high damage, no review | low, a person checks before sending |
| Is someone responsible? | nobody | the business unit drives it |
| Is the benefit measurable? | gut feeling only | time, volume or rate measured beforehand |
A case with many repetitions, digital data and a person before the final step almost always beats the spectacular one-off.
Worked example: what an AI use case in customer service delivers
A specialist retailer receives 1,200 customer emails a month; each takes six minutes on average today. An AI assistant sorts the mail and prepares a draft reply for 60 percent of requests, checked before sending. Assumption for the example: for these requests time drops to two minutes.
| Item | Value |
|---|---|
| Requests with draft per month | 720 |
| Time saved per request | 4 minutes |
| Time saved per month | 2,880 minutes, i.e. 48 hours |
| Value at 45 euros per working hour | 2,160 euros |
| Running costs for tool and operation (assumption) | 400 euros |
| Net saving per month | 1,760 euros |
| One-off setup (assumption) | 12,000 euros |
| Payback | just under 7 months |
The calculation is only as good as its inputs. Measure how long a request takes today before you start, and again after eight weeks. Realistic setup and running costs are covered in How much does an AI agent cost, a fully calculated service project in Automating customer service with AI.
AI use cases and the AI Act: what matters legally
Most AI use cases in the table trigger no special obligations under the AI Act. For all of them, businesses using AI must take measures to support their staff's AI literacy, such as training and clear rules (Art. 4 as amended by the Digital Omnibus); the law does not require a specific level. Two groups need more attention:
- Customer contact: chatbots and phone assistants must make clear that an AI is speaking. Our guide on the AI labelling obligation shows how this works in practice.
- HR: AI that filters applications or evaluates employees counts as high-risk under Annex III. After the postponement by the Digital Omnibus, those obligations apply from 2 December 2027. Drafting the text of job ads is not affected; using AI to target job ads or filter applications is.
For personal data the GDPR applies as well: a data processing agreement with the provider, a clear storage location, no customer data in tools without a contract. This section is not legal advice for individual cases.
Which AI use cases make a good start
The order decides how well AI is accepted in-house. A three-stage path has proven itself:
- First four weeks: a low-effort text case such as draft replies in service or meeting minutes. The goal is a measurable gain and trust in the team.
- Second quarter: a case with an interface, such as quotes from the ERP or invoices straight into accounting.
- After that: cases with custom training such as visual inspection or predictive maintenance, once the data is in place.
Each stage ends with a measurement and a decision: roll out, improve or stop.
Common mistakes with AI use cases
- Starting with the hardest case because it impresses most.
- Not measuring the baseline and then being unable to prove a benefit.
- Involving the business unit only after the tool has been bought.
- Putting customer data into a tool without a data processing agreement.
- Never moving the pilot into regular operation because nobody owns it.
Checklist for your AI use cases
- Repetitive tasks collected per department.
- Each candidate scored with the suitability check, seven points as threshold.
- Baseline measured: time, volume or error rate.
- Legal position checked: customer contact, HR, data protection.
- Owner and measurement date after eight weeks fixed.
- Decision after the pilot documented.
Questions about AI use cases
What is an AI use case?
An AI use case is a concrete task in a company that artificial intelligence takes over fully or partly, such as reading invoices or preparing replies to customer requests. It has a clear start, a clear result and a metric.
How many companies in Germany use AI?
According to Eurostat, around 26 percent of enterprises with ten or more employees in Germany used at least one AI technology in 2025; among large enterprises with 250 or more employees it was 57 percent.
Which AI use case is best for getting started?
Cases with text and documents that occur daily and where a person checks before sending: draft replies in customer service, reading invoices, writing minutes.
What does an AI use case cost an SME?
A pilot with standard software often costs a low five-figure amount for setup plus running fees. What matters is the comparison with working time saved, as in the worked example above.
Want to know which AI use cases save the most time in your business? Go through your list with us in a short introductory call. How we automate processes together with your team is described on our AI automation page.
Transparency notice: This article was created with the assistance of artificial intelligence.