What AI can realistically do today in sales, marketing, customer service, finance, HR, operations and IT, what each use case needs, and where the risks are.
In short
Realistic AI use cases in a mid-sized company are mostly about reading, writing, classifying and predicting: drafting sales and marketing text, routing customer requests, matching invoices, forecasting demand and answering internal questions from approved documents. Each needs good data, a clear owner and a human check. HR and credit decisions carry extra legal requirements under the EU AI Act.
Lists of "100 ways to use AI" rarely help a leadership team decide what to do on Monday. The useful question is narrower: which tasks in our departments are frequent, pattern-based and checkable, and do we have the data for them? This explainer from Devies, a Gothenburg-based product and engineering consultancy, gives department heads and IT leaders a realistic map of AI use cases (ai användningsområden) for companies with roughly 50 to 500 employees, with what each requires.
What is an AI use case?
An AI use case is a specific task in a specific process where an AI model does part of the work, with a defined input, output, user and measure of success. "Use AI in finance" is an ambition; "extract fields from incoming supplier invoices and suggest the cost account" is a use case.
The term is often blurred with related ideas:
Term
What it is
Example
AI use case
What it is
One task in one process, with data, user and success measure
Example
Draft first replies to support emails for an agent to approve
AI tool
What it is
Software that offers AI features
Example
A chat assistant or an AI feature in the CRM
AI project
What it is
Work to build or introduce one or more use cases
Example
A three-month pilot in customer service
AI strategy
What it is
Choices about where AI should create value and in what order
Example
Focus on finance and customer service in year one
Thinking in use cases makes value and risk visible. A tool licence is a cost; a use case has a before-and-after you can measure.
How widely do Swedish companies use AI?
About a third of Swedish companies used AI in 2025, and use rises steeply with company size. Statistics Sweden reports that 35 percent of companies with ten or more employees used some form of AI in 2025, up ten percentage points from 2024, with 31 percent of small, 50 percent of medium-sized and 72 percent of large companies doing so.
According to the same survey, marketing and sales was the most common purpose among companies using AI, followed by business administration and management. Lack of relevant skills was the most common reason why companies that considered AI chose not to use it. For a mid-sized company, that suggests the bottleneck is less the technology than knowing where to apply it and who will run it.
Which AI use cases fit each department?
The most realistic AI use cases in each department are tasks that are frequent, based on text, documents or historical data, and easy for a person to check. The table lists examples with the data each needs.
Department
Realistic use cases
Data needed
Watch out for
Sales
Realistic use cases
Summarise accounts before meetings, draft follow-up emails, score leads
Data needed
CRM history, emails, won and lost deals
Watch out for
Outdated CRM data, invented details in drafts
Marketing
Realistic use cases
Draft and adapt copy, generate image variants, analyse campaign results
Data needed
Brand guidelines, past content, campaign data
Watch out for
Brand consistency, copyright, marking of AI content
Customer service
Realistic use cases
Classify and route requests, draft answers, summarise long cases
Data needed
Ticket history, knowledge base
Watch out for
Wrong answers sent unchecked, personal data
Finance
Realistic use cases
Extract invoice data, suggest accounts, flag anomalies, draft report commentary, forecast cash flow
Data needed
Invoices, ledger, payment history
Watch out for
Accuracy of figures, audit trail
HR
Realistic use cases
Answer policy questions, draft job ads, summarise survey comments
Data needed
Policies, handbooks, survey data
Watch out for
Recruitment and evaluation are high-risk uses
Operations and supply chain
Realistic use cases
Forecast demand, predict delivery delays, read emailed orders, inspect quality from images
Data needed
Order and stock history, sensor or image data
Watch out for
Data quality, changes in demand patterns
IT
Realistic use cases
Answer internal support questions, assist developers, summarise logs and incidents
Data needed
Ticket history, documentation, code
Watch out for
Security of code and credentials, access rights
Management
Realistic use cases
Draft board material, summarise market and competitor news
Data needed
Reports, approved data sources
Watch out for
Unverified figures in summaries
Two patterns cover most of the table. Generative AI handles text-heavy work: reading, drafting and summarising. Predictive models handle numbers: forecasts, scores and anomaly detection. Many companies start with the first because tools such as Microsoft Copilot are already licensed, and move to the second when they have the data platform for it.
Where are the legal and practical limits?
The legal limits are tightest where AI affects individuals, especially in HR and credit, and the practical limits are data quality, access rights and the cost of mistakes.
Annex III of the EU AI Act lists employment uses as high-risk, including AI for recruitment and selection, for decisions on promotion or termination, for allocating tasks and for monitoring and evaluating performance. As of October 2026, the European Commission states that obligations for these Annex III systems apply from 2 December 2027. An HR policy assistant is low-risk; a tool that ranks job applicants is not.
Practical limits matter in every department:
- Data: a use case is only as good as the documents or history behind it. Outdated policies give outdated answers.
- Access: an assistant must not show salary data or contracts to people who should not see them.
- Mistakes: generative models can produce confident errors. Keep a human check wherever an error reaches a customer, a ledger or an employee.
- Ownership: someone in the department must own the use case after launch, not only IT.
How do you choose where to start?
Choose the first use case where a frequent, checkable task meets available data and a department head who wants it; then measure it before scaling. Enthusiasm without data, or data without an owner, rarely produces results.
- Collect two or three candidate use cases per department, written as specific tasks.
- For each, estimate volume and time spent today, and the cost of a mistake.
- Check the data: does it exist, is it current and who may see it?
- Pick one or two with high volume, low risk and an engaged owner.
- Define the success measure before the pilot, such as handling time or error rate.
Our guide on how to prioritise AI use cases provides a scoring method, and AI automation of business processes shows how to turn a chosen use case into a working process with a human check.
Keep reading
From our work
Client Case: Stena LineStena Line modernised its Azure data platform - improving Power BI reliability, reducing pipeline runtimes and scaling with medallion architecture.Read the caseDevies
Want a second opinion?
Devies, a Gothenburg-based product and engineering consultancy, works on questions like this with Nordic product and IT teams. Tell us where you are and we will suggest a sensible first step.
Frequently asked questions
What are the most common AI use cases in companies?
Among Swedish companies using AI, Statistics Sweden reports that marketing and sales is the most common purpose, followed by business administration and management. In practice that means drafting and adapting text, summarising information, answering internal questions, classifying incoming requests and handling documents such as invoices. Forecasting and anomaly detection are common where reliable historical data exists.
Which department should start with AI?
Start where frequent, checkable tasks meet good data and an engaged department head. In many mid-sized companies that is customer service, finance or sales: request routing, invoice handling and meeting preparation are high-volume and easy to review. HR is a weaker first choice for anything beyond policy questions, because recruitment and evaluation uses are high-risk under the EU AI Act.
Is using AI in HR allowed?
Yes, but with care. Low-risk uses such as answering policy questions or drafting job ads are generally fine with normal GDPR safeguards. AI used for recruitment, selection, promotion, termination, task allocation or performance evaluation is classed as high-risk under Annex III of the EU AI Act, with strict obligations that, as of October 2026, apply from 2 December 2027.
How do we measure the value of an AI use case?
Measure the task before and after: time per case, volume handled, error or correction rate, and customer or employee satisfaction where relevant. Agree the measure before the pilot starts and compare against the manual process over the same period. Include licence, model and maintenance costs, and the time people spend reviewing AI output, in the calculation.
