AI agent examples · by department
AI agent examples: 20 agents doing real work today
Twenty AI agents from practice, by department. For each example you'll read what the agent does, which systems it works with, where a human approves and roughly what it delivers. Six of them already run at our clients, with real numbers. The rest is an honest indication.
- 8 departments
- 6 real client cases
- A human always approves
20
examples across eight departments
192 h
a month back at Van Berkel Professionals
80%
of 6,000 customer questions a month with a prepared answer
132 h
a month saved on product copy at EasyToys
First, this
What is an AI agent, and how do you read this list?
An AI agent is software that gets a goal, plans the steps itself, uses your systems (email, ERP, CRM, documents) and reports back. A chatbot answers a question. An agent looks up the answer in three systems, drafts a reply and puts it in front of you. There's more in what is an AI agent.
The twenty examples below aren't a vision of the future. They're tasks that come back every day in almost any business with office work, and where an agent can already do the heavy lifting today. In none of the examples does the human disappear: the question is always where that person comes in.
Six examples already run at our clients. You'll recognize them by the number on the card and the label 'Client case'. The other fourteen show an indication of the time saved: a realistic estimate based on what we see in similar processes, not a promise. What it delivers for you, we calculate with your own numbers.
01What the agent does
The task in plain language, from the moment it comes in to the result.
02Systems
Where the agent reads and writes. Often what you already have: Outlook, Exact, AFAS, HubSpot, Shopify.
03Human in the loop
The moment a colleague checks, decides or takes spot checks.
04Time saved
A real client figure for a client case, otherwise an indication. Always the expected scenario, never the most favourable one.
At a glance
The AI floor plan of an average office.
Click a department: what AI takes over, what the human does, and what it delivered at clients. Below it, the twenty examples in detail.
Click a department
What AI takes over here
Customer service
- Draft a reply with the order detailsHuman approves
- Label and route ticketsFully AI
- Call the difficult customerAI assists
In practice
30% → 80%
of customer questions get a prepared reply (enterprise client)
01 / Customer service
AI agents for customer service
This is almost always where the biggest volume is, and so the fastest gain. The agent does the digging, the employee keeps the tone and handles the exceptions.
80%
Client case · Enterprise
1. Preparing the answer
Reads every incoming question, looks up the order and customer details and prepares a draft answer before the ticket opens.
- Systems: ticketing system and customer data
- Human in the loop: the employee checks and sends
- Result: at 6,000 questions a month, from 30% to 80% with a prepared answer
Example
2. Sorting and routing tickets
Reads every email and form, works out the topic, language and urgency, and routes the ticket to the right team. Urgent tickets go to the top.
- Systems: shared mailbox, helpdesk
- Human in the loop: the team lead takes spot checks and corrects
- Time saved (indication): 1 to 2 minutes per ticket
Example
3. Handling returns and complaints
Checks the return period and the order history, creates a return label or a draft credit note and drafts the reply.
- Systems: webshop, shipping platform, accounting
- Human in the loop: approves credit notes and anything outside the policy
- Time saved (indication): 5 to 10 minutes per return
02 / Sales
AI agents for sales
Salespeople want to sell, not copy and paste. The agent prepares, the salesperson has the conversation and keeps the price and the relationship in their own hands.
4 → 1 h
Client case · Anonymous
4. Drafting sales proposals
Pulls together customer data, call notes and suitable sections from earlier proposals, and sets up a first version in your own layout.
- Systems: CRM and earlier proposals
- Human in the loop: the salesperson does the final edit and sends it
- Result: a proposal takes 1 hour instead of 4
Example
5. Following up on leads
Picks up every new request from the website, looks up the company information, estimates the fit and drafts a first personal email. A task is ready and waiting in the CRM.
- Systems: website form, CRM (HubSpot or Salesforce), email
- Human in the loop: the salesperson reads the email and sends it
- Time saved (indication): 10 to 15 minutes per lead, and no more leads left lying around
03 / Procurement
AI agents for procurement
Procurement is checking, comparing and chasing. Exactly the kind of work an agent doesn't need a coffee break for.
Example
6. Checking order confirmations
Puts every order confirmation from a supplier next to the purchase order: price, quantities, delivery time. If it matches, the order is updated. If it doesn't, the buyer gets a heads-up with the difference.
- Systems: email, ERP (Exact, AFAS, SAP or Dynamics)
- Human in the loop: the buyer decides on every deviation
- Time saved (indication): 4 to 8 hours a week at a few hundred orders a month
Example
7. Comparing supplier quotes
Reads the quotes from different suppliers, even when each one uses its own format, and puts them in one overview: price, delivery time, terms, and what's missing.
- Systems: email, PDFs, Excel or the purchasing system
- Human in the loop: the buyer chooses and negotiates
- Time saved (indication): 1 to 2 hours per request for quotation
04 / Finance
AI agents for finance
High volume, fixed formats, and every error shows up in the check. Finance is often where a task can move fastest to 'fully AI, with spot checks'.
12 h/wk
Client case · Anonymous
8. Processing invoices
Pulls invoices from the mailbox, reads the supplier, amounts, VAT and PO number, matches them with the purchase order and prepares them for booking.
- Systems: mailbox, accounting software
- Human in the loop: accounts payable approves, deviations are listed separately
- Result: 12 hours a week saved
Example
9. Chasing unpaid invoices
Drafts reminders in the right tone for each customer, processes payment promises from the replies and keeps track of who owes what.
- Systems: accounting, email, CRM
- Human in the loop: approves reminders to large or sensitive customers
- Time saved (indication): 3 to 5 hours a week
Example
10. Preparing the monthly report
Gathers the figures, compares them with the budget and the previous month, and writes a first explanation for every notable deviation.
- Systems: accounting, Excel or Power BI
- Human in the loop: the controller interprets and signs off
- Time saved (indication): half a day to a full day a month
05 / HR
AI agents for HR
HR is about people, so here we look extra carefully at where AI does and doesn't belong. AI that filters applications or assesses candidates falls under the European AI Act (Annex III) in the high-risk category, with strict requirements for human oversight and documentation. That's why the two examples below stick to questions and administration.
Example
11. HR help desk
Answers questions about leave, expense claims, pensions and policies from your own staff handbook and the collective labour agreement, with the source attached. Personal or sensitive questions go straight to HR.
- Systems: staff handbook, intranet, Teams or Slack
- Human in the loop: HR takes spot checks and handles anything personal
- Time saved (indication): 2 to 4 hours a week for an HR team of a few people
Example
12. Arranging onboarding
As soon as a contract is signed: requesting accounts, ordering a laptop, scheduling intro meetings and sending the new colleague a welcome email with a checklist.
- Systems: HR system (AFAS, for example), Microsoft 365, calendar
- Human in the loop: HR checks the plan before it goes out
- Time saved (indication): 1 to 2 hours per new employee
How a human approves
The agent makes a plan. You approve it.
Every example on this page has a human in it somewhere. This is how that works in practice.
- 01
The agent shows its plan
Before it does anything, the agent shows what it intends to do.
- 02
You approve or adjust
You give the go-ahead, or steer the plan in a different direction.
- 03
Only then does it act
The agent carries out the plan once you've approved it.
- 04
More room over time
Once it has run without errors for a while, it's allowed to do more by itself.
06 / Operations and planning
AI agents for operations and planning
Orders, requests and planning: the heart of many businesses, and often where the most retyping happens. Also where the planner hasn't dared to take a holiday in three years.
192 h
Client case · Van Berkel Professionals
13. Processing requests
Reads every request that comes in by email, pulls out the important details, labels it and prepares everything in the system.
- Systems: mailbox and their own system
- Human in the loop: an employee checks and passes it on
- Result: 192 hours a month back, about 2 FTEs
Example
14. Entering orders
Orders by email, PDF or Excel: read, item numbers and prices checked, and entered as a draft order in the ERP. Unknown items and deviating prices are listed separately.
- Systems: email, ERP (Exact, AFAS, SAP)
- Human in the loop: inside sales confirms with one click
- Time saved (indication): 5 to 10 minutes per order
Example
15. Making a planning proposal
Puts orders, capacity, delivery times and absences side by side and proposes a weekly schedule, explaining where things get tight.
- Systems: ERP, WMS or TMS, rostering software
- Human in the loop: the planner moves things around, decides and publishes
- Time saved (indication): 4 to 8 hours a week, and the knowledge no longer sits in one person's head
See how it works
How an agent processes a request, from email to system.
Example 13, step by step. The agent at Van Berkel Professionals works this way: reading, extracting the details, assessing and preparing. An employee approves.
- 01
Read
The agent reads every request that comes in by email.
- 02
Extract
It pulls out the important details.
- 03
Assess
It assesses the request and labels it.
- 04
Prepare
It prepares everything in the system.
- 05
Approve
An employee checks it and passes it on.
07 / Marketing
AI agents for marketing
Marketing always has more ideas than hands. The agent makes the first version, the marketer makes it right.
132 h
Client case · EasyToys
16. Product copy in six languages
Reads the suppliers' PDFs and emails, writes product copy in six languages in the brand's own tone and puts it straight into the ERP.
- Systems: email, supplier files, ERP
- Human in the loop: the team checks, as with every workflow we build
- Result: 132 hours and € 6,600 a month saved
Example
17. Content from your own knowledge
Turns a client case, webinar or internal presentation into a first version of a blog, LinkedIn post or newsletter, in your tone.
- Systems: shared drive, CMS, content calendar
- Human in the loop: the marketer edits and publishes
- Time saved (indication): 3 to 6 hours a week
08 / Knowledge and legal
AI agents for knowledge and legal work
Knowledge that lives in two people's heads, contracts nobody reads in full, files of a hundred pages. Here the agent is a fast reader, and the expert stays the expert.
Example
18. Knowledge base assistant
Answers questions from your own handbooks, procedures, quotes and project files, always with a reference to the source. 'Ask Pete' becomes 'see what the assistant says, and ask Pete if in doubt'.
- Systems: SharePoint or Google Drive, intranet
- Human in the loop: when in doubt, the assistant refers you to the expert
- Time saved (indication): 10 to 20 minutes per search
Example
19. Checking contracts
Puts an incoming contract next to your standard terms and checklist, and flags what deviates: liability, term, payment terms, penalties.
- Systems: document management, email
- Human in the loop: the lawyer or managing director decides on every deviation
- Time saved (indication): 30 to 60 minutes per contract
700+
Client case · KplusV
20. Assessing pitch decks
Reads every pitch deck, assesses it against the consultancy's own canvas and finds suitable investors from a database of more than 700.
- Systems: their own platform, documents
- Human in the loop: a consultant always checks
- Result: every deck assessed against their own canvas and matched with 700+ investors
All 20 side by side
Twenty AI agent examples in one table.
For each agent: the department, where the human sits according to the four roles from our APO method, and the time saved. Client case means a measured result at a client; the rest is an indication.
| Department | Role of the human | Time saved | |
|---|---|---|---|
| 1. Preparing the answer | Customer service | Human approves | Client case: 30% → 80% prepared |
| 2. Sorting tickets | Customer service | Fully AI, with spot checks | Indication: 1 to 2 min per ticket |
| 3. Returns and complaints | Customer service | Human approves | Indication: 5 to 10 min per return |
| 4. Sales proposals | Sales | AI assists | Client case: 4 hours → 1 hour |
| 5. Following up on leads | Sales | Human approves | Indication: 10 to 15 min per lead |
| 6. Order confirmations | Procurement | Human approves deviations | Indication: 4 to 8 hours a week |
| 7. Comparing quotes | Procurement | AI assists | Indication: 1 to 2 hours per request |
| 8. Processing invoices | Finance | Human approves | Client case: 12 hours a week |
| 9. Chasing unpaid invoices | Finance | Human approves | Indication: 3 to 5 hours a week |
| 10. Monthly report | Finance | AI assists | Indication: 0.5 to 1 day a month |
| 11. HR help desk | HR | Fully AI, with spot checks | Indication: 2 to 4 hours a week |
| 12. Onboarding | HR | Human approves | Indication: 1 to 2 hours per employee |
| 13. Processing requests | Operations | Human approves | Client case: 192 hours a month |
| 14. Entering orders | Operations | Human approves | Indication: 5 to 10 min per order |
| 15. Planning proposal | Planning | AI assists | Indication: 4 to 8 hours a week |
| 16. Product copy | Marketing | Human approves | Client case: 132 hours a month |
| 17. Content from your own knowledge | Marketing | AI assists | Indication: 3 to 6 hours a week |
| 18. Knowledge base assistant | Knowledge | AI assists | Indication: 10 to 20 min per question |
| 19. Checking contracts | Legal | AI assists | Indication: 30 to 60 min per contract |
| 20. Assessing pitch decks | Advisory | Human approves | Client case: 700+ investors |
Swipe sideways to see all columns →
Indications are estimates based on similar processes, not a guarantee. What it delivers for you depends on volume, data and how the process runs now.
The pattern
Twenty agents, one pattern: AI does the heavy lifting, a human signs off.
Not one example on this page is an agent making unseen decisions about customers, money or people. Start with 'human approves'. Only when a task has run without errors for months does it move on to spot checks.
Where do you start?
How do you choose your first AI agent?
Choose the task that comes back often, involves a lot of retyping and has a result you can measure. Not the most exciting task, but the most boring one with the most volume. That's where an agent pays for itself fastest, and where your team learns to trust it fastest.
First, do the maths with three numbers: how long does the task take, with how many people, and how often a month? Forty minutes, two people, eighty times a month comes to € 57,600 a year at € 45 per hour. Then you also know what the agent is allowed to cost.
Then check the data. An agent that reads invoices needs access to the mailbox and the accounting software, and a few hundred examples of correctly processed invoices. If that's missing, the first step is getting the data in order, not building.
In the AI Quick Scan we do this together in one day on site: which tasks, in which order, what they deliver. Within a week you have a business case for each opportunity.
01High volume
Dozens to thousands of times a month. Once a quarter rarely pays off.
02Digital input
Email, PDF, form or system. Handwritten notes can work too, but they're harder.
03Clear outcome
You know when it's right. Then a human can approve quickly.
04Data and examples
Without examples of good work, we calculate with 40 percent confidence at most.
From example to working agent
How an agent goes into production at your company.
We build agents within the AI Partnership, so that after delivery there's also someone who monitors, improves and extends them.
- 01
Intro call
Fifteen minutes
Which of these twenty do you recognize? And where's the most volume at your company?
- 02
AI Quick Scan
One day plus a week
Processes mapped, a business case per opportunity, a twelve-month roadmap. € 6,000 one-off.
- 03
Build and connect
Quarter 1
The first agent, connected to your systems and tested with real examples.
- 04
Live with a human in the loop
Then
At first a human approves everything. Every month you see what it delivers, in hours and euros.
Questions
Frequently asked questions about AI agent examples.
What are examples of AI agents in business?
Common AI agents answer customer questions with a draft reply, read invoices and match them with the purchase order, put orders from email into the ERP, process requests, draft sales proposals, write and translate product copy, check contracts and answer questions from your own knowledge base. At NinA AI, a human approves the output of every one of these agents.
What's the difference between an AI agent and a chatbot?
A chatbot answers questions in a conversation. An AI agent gets a goal, plans the steps itself, uses systems such as email, ERP and CRM, carries out tasks and reports back. A chatbot tells you where your invoice is; an agent looks up the invoice, checks it and prepares it for booking.
Which AI agent delivers results fastest?
Usually an agent on a task with high volume and a lot of retyping: customer questions, orders, invoices or requests. At Van Berkel Professionals, an agent for request processing saved 192 hours a month. What pays off fastest for you, we calculate with three numbers per task: duration, number of people and frequency.
Are the time savings on this page guaranteed?
No. Six examples are real client cases with measured results. The other fourteen have an indication based on similar processes. What it delivers for you depends on volume, data and the current process. In the AI Quick Scan we calculate that with your own figures, and we only start when it pays back at least twice the investment.
Does an AI agent make decisions on its own?
Only within the limits you set. At NinA AI, every agent starts with a human who approves everything. Only when a task has run without errors for a long time can it move to spot checks. We never leave decisions about people, such as selecting applicants, to an agent.
Which systems can an AI agent work with?
With almost any system that has an API or an export: Outlook and Gmail, Exact, AFAS, SAP, Microsoft Dynamics, Salesforce, HubSpot, Shopify, SharePoint, Google Drive and most WMS and TMS packages. We build a lot in n8n, and custom where needed.
What does it cost to have an AI agent built by NinA AI?
We build agents within the AI Partnership: after the € 6,000 AI Quick Scan, you choose package S (€ 4,500 per month), M (€ 7,500) or L (€ 12,000), for a minimum of twelve months. Hosting and maintenance are covered by a maintenance contract from € 245 per month, including AI usage. All amounts excluding VAT.
Is it safe to give an AI agent access to our systems?
Yes, if the permissions are set up properly. An agent only gets access to what it needs for its task, everything is logged and a human approves. At NinA AI, data stays in the EU and we work in line with ISO 27001 and NEN 7510; certification is in preparation.
Which agent will you hire first?
Book an intro call. Bring the example you recognized most, and we'll work out together what it would deliver for you.
01
Intro call
Fifteen minutes, no strings. Where does work get stuck and what do you want to achieve?
02
AI Quick Scan
One day on site. Within a week, a business case and a roadmap.
03
AI Partnership
A dedicated team that builds, trains and shows what it delivers every month.
160+ organizations · Sessions rated 9.3 · Based in Amsterdam