AI agent examples: 30 use cases by department
Where do AI agents pay off in a company? This collection shows 30 examples from six departments. Each case lists what the agent does, the knowledge it needs, who approves and how autonomously it may work.
Vendor-neutral, with an approval checklist as a PDF.
Key points
- Many
AI agent examples show it: a good first agent reads and suggests. Acting without approval is only an option for small, reversible steps. - Every agent needs
knowledge from your sources : policies, manuals, product data. Without it, it stays with generic answers. - Where
money, contracts or people are involved, the agent prepares and a person decides.
What do these AI agent examples show?
The examples are typical tasks that come up in many companies. They are sorted by department and always follow the same pattern: task, knowledge, approval, level. That way you see quickly which cases suit a first project and where more preparation is needed.
The levels follow the classification in the agentic AI guide: 0 answer, 1 suggestion, 2 act with approval, 3 act within limits. Higher levels rarely make sense in companies.
In customer service, requests repeat and the answers are usually already in policies and product documents. That makes it a good place to start.
See examples →In HR, agents help with questions and workflows around policies. Decisions about people stay with a person; the works council should be involved early here too.
See examples →IT has many standard cases with clear solutions. Agents suggest fixes and handle routine work; changes to systems and permissions stay behind an approval.
See examples →In sales, agents save time on preparation and follow-up. Anything that goes to customers is checked by a person before sending.
See examples →Where money is involved, the agent checks and prepares, and a person approves. Payments and postings should never run without approval.
See examples →In legal and compliance, every reference counts. Agents find and compare; the assessment stays with experts.
See examples →AI agents in customer service.
In customer service, requests repeat and the answers are usually already in policies and product documents. That makes it a good place to start.
| Use case | What the agent does | Knowledge | Approval | Level |
|---|---|---|---|---|
| Classify requests | sorts emails by topic and urgency and routes them | topic list, responsibilities | spot checks | 3 · within limits |
| Draft reply with sources | writes a reply from policies and manuals, with references | returns, warranty, delivery terms | before sending | 1 · Suggestion |
| Look up order status | reads status and tracking data and summarises them | order system, read only | none, read only | 0 · Answer |
| Prepare a complaint case | checks the terms and suggests a credit note or replacement | complaints policy, order data | team lead | 2 · with approval |
| Call note | summarises a call and creates open items | call transcript, customer record | staff review | 1 · Suggestion |
AI agents in HR.
In HR, agents help with questions and workflows around policies. Decisions about people stay with a person; the works council should be involved early here too.
| Use case | What the agent does | Knowledge | Approval | Level |
|---|---|---|---|---|
| Policy questions | answers questions on leave, travel expenses, parental leave with sources | handbook, works agreements | none, answer only | 0 · Answer |
| Onboarding tasks | creates tasks for account, device and training | onboarding process, role profiles | HR confirms | 2 · with approval |
| Check documents | checks submitted documents for completeness | checklist per case | HR decides | 1 · Suggestion |
| Draft a job ad | writes an ad from the requirements profile | profile, templates, benefits | before publishing | 1 · Suggestion |
| Training records | reminds of due training and collects records | training plan | spot checks | 3 · within limits |
AI agents in IT.
IT has many standard cases with clear solutions. Agents suggest fixes and handle routine work; changes to systems and permissions stay behind an approval.
| Use case | What the agent does | Knowledge | Approval | Level |
|---|---|---|---|---|
| Classify tickets | sets category and priority and assigns the team | categories, responsibilities | spot checks | 3 · within limits |
| Suggest a fix | finds matching guides in the IT wiki and suggests them | IT wiki, solved tickets | technician checks | 1 · Suggestion |
| Software request | checks the request against approved software and prepares the order | software list, licences | IT lead | 2 · with approval |
| Incident report | combines alerts and logs into a report | logs, monitoring | before sending | 1 · Suggestion |
| Access request | checks the request against the role concept and prepares it | role concept | owners confirm | 2 · with approval |
AI agents in sales and marketing.
In sales, agents save time on preparation and follow-up. Anything that goes to customers is checked by a person before sending.
| Use case | What the agent does | Knowledge | Approval | Level |
|---|---|---|---|---|
| Draft a quote | creates a draft from product data and pricing rules | product data, pricing rules | before sending | 1 · Suggestion |
| Meeting prep | summarises customer history and open points | CRM, past quotes | none, internal only | 1 · Suggestion |
| Answer objections | suggests answers to typical objections, with sources | sales playbook, product documents | sales checks | 1 · Suggestion |
| Follow-up email | writes a follow-up email after the meeting | meeting note | before sending | 1 · Suggestion |
| Adapt content | adapts texts to audience and channel | style and brand guide | before publishing | 1 · Suggestion |
AI agents in finance and purchasing.
Where money is involved, the agent checks and prepares, and a person approves. Payments and postings should never run without approval.
| Use case | What the agent does | Knowledge | Approval | Level |
|---|---|---|---|---|
| Check an invoice | matches the invoice with the order and delivery note | orders, delivery notes | accounting confirms | 2 · with approval |
| Check travel expenses | checks receipts against the travel policy and flags deviations | travel policy | managers confirm | 2 · with approval |
| Compare suppliers | compares quotes by price, delivery time and terms | quotes, purchasing policy | purchasing decides | 1 · Suggestion |
| Watch deadlines | reminds of termination and renewal deadlines | contracts | none, reminder only | 3 · within limits |
| Monthly report | summarises key figures with explanations | accounting reports | controlling checks | 1 · Suggestion |
AI agents in legal and compliance.
In legal and compliance, every reference counts. Agents find and compare; the assessment stays with experts.
| Use case | What the agent does | Knowledge | Approval | Level |
|---|---|---|---|---|
| Find contract clauses | finds clauses such as liability, term or termination with references | contract archive | none, answer only | 0 · Answer |
| Compare a contract with a template | flags deviations from the standard template | templates, position paper | legal team assesses | 1 · Suggestion |
| Policy questions | answers questions on internal policies with sources | policies, code of conduct | none, answer only | 0 · Answer |
| Collect evidence | gathers evidence for an audit | documentation, logs | compliance checks | 1 · Suggestion |
| Track changes | summarises new requirements from subscribed sources | official sources | experts assess | 1 · Suggestion |
How to choose your first use case.
Choose a case that comes up often, where the knowledge is already in documents and where a mistake costs little. Start at level 0 or 1 and measure how often the result fits without correction. Only then does a higher level pay off.
Level 0 examples and many level 1 examples can be built directly as a RAGSuite assistant, as long as the result is an answer or draft from your documents. For actions, a workflow tool such as n8n queries RAGSuite and carries out the action.
Does the use case suit an agent?
Tick what is done. Nothing is stored.
Questions about the examples.
Which AI agent examples suit a first project?
Tasks that come up often, follow clear rules and have an easily checked result are a good fit. Typical examples are draft replies with sources, internal policy questions and finding contract clauses. They work with read-only access and need no actions that are hard to undo.
What do the levels in the table mean?
The levels describe how autonomously the agent acts. Level 0 only answers questions, level 1 makes suggestions, level 2 carries out actions after approval, and level 3 acts on its own in clearly limited cases, with spot checks. The classification is a working aid from RAGSuite and is explained in the agentic AI guide.
Can RAGSuite run these AI agents?
RAGSuite provides the knowledge: answers and passages with sources from your documents, via chat, search, REST API or MCP server. The actions themselves are handled by a workflow tool such as n8n or by an agent that queries RAGSuite. Level 0 examples and many level 1 examples can be built directly as a RAGSuite assistant, as long as the result is an answer or draft from your documents.
Does the works council have co-determination rights over AI agents in HR?
If you have a works council, often yes. Section 87(1) No. 6 of the Works Constitution Act (BetrVG) gives it co-determination rights over technical systems intended to monitor employees’ behaviour or performance; the Federal Labour Court holds that being capable of it is enough. Logs that tie actions to individuals often qualify. Settle purpose and evaluation early in a works agreement. This text is not legal advice.
Your first agent, matched to your department.
Jürgen helps you pick the example that pays off fastest for your team.