AI use cases are specific tasks where AI takes real work off your plate: pre-sorting email, drafting quotes, matching receipts. In a small business, the ones that pay off first sit in admin, sales and bookkeeping. The work comes up every week, you can check the result in minutes, and a mistake stays inside the building as a draft. Customer service and marketing come next. HR is where you don't start.
I run my business with fourteen AI employees, and thirteen of them have their own page with real work results on my team page. Wherever I write "running in my business" below, one of those pages backs it up. Everything else is an example, not something I run myself.
How to spot a good AI use case
Most lists of AI use cases tell you what is technically possible. If you run a business with five or fifteen people, the more useful question is what to do first. I check four things, condensed from my delegation questions in Which Tasks You Can Hand Off to AI and the rating step in Process Optimization With AI: The Method:
- The task comes up at least once a week.
- The input is already digital, in an inbox, a folder or an accounting tool. No AI can work with what only lives in your head.
- You can check the result in a few minutes.
- A mistake stays in-house, because the AI produces drafts and a person approves them.
If a use case ticks all four boxes, it goes to the front of the line. If one is missing, it waits. What's new here is the view by department.
AI use cases by department
Admin and office
Running in my business: Frieda sorts my inbox into four piles, Now, Today, Can wait and Trash, plus a fifth for anything that needs my judgement. She saves replies as drafts and has never sent anything herself. On top of that she writes appointment briefings, pre-sorts receipts and puts together a weekly report.
Other examples: filing documents by fixed rules, writing up meeting notes, preparing appointments. At a tax advisory firm I describe on my home page, two AI employees handle the digital tax assessments: one picks them up from the inbox, the second checks them against around 45 rules taken from the front office's daily work and files them.
Why first: The inbox ticks all four boxes. It runs every day, it's digital, and you can judge a draft reply in seconds. The pile system itself is in Sorting Your Inbox: The Pile System.
Sales
Running in my business: Anton turns a first conversation into a quote PDF, and once the client says yes, he creates the contact, quote and invoice as drafts in the accounting tool. Without a price source, he doesn't set a price. Olaf looks for local businesses with outdated websites and delivers a checked shortlist, but never contacts anyone. Sonja writes the email to new clients in my voice and leaves it as a draft in my inbox.
Other examples: preparing follow-ups on open quotes, condensing call notes into a summary for the CRM.
Why early: Quotes repeat, and you can tell at a glance whether a draft fits. The risk lies in prices and commitments, which is why sending stays with you. What a good quote looks like is in How to Write a Quote That Convinces.
Bookkeeping
Running in my business: Peter goes through receipts in my accounting software, sets the receipt type and matches card statements. Anything that isn't clear-cut, he flags as open. He doesn't post entries; sign-off stays with me and my tax advisor.
Other examples: collecting incoming invoices from the inbox, listing open items for the week, pulling together documents for your accountant. What this looks like for payment reminders is in Automate Payment Reminders and Dunning.
Why early, but narrow: Receipts come in every week and follow fixed rules. At the same time, a missed receipt costs more than a misfiled email. So I start with matching and reconciling here, never with posting. What works in this area today is in AI Bookkeeping: What Actually Works Today.
Customer service
Running in my business: Gustav analyses calls with my clients, with consent and pseudonymised, and builds a register of the questions that keep coming up, each with my checked answer. That turns into FAQs and reply snippets, described in Build FAQs From Customer Conversations.
Other examples: analysing sales calls, draft replies for a support inbox, or a phone assistant that picks up calls outside office hours. What such an assistant can do and what the law says is in AI Receptionist: Providers, Pricing, Law.
Why second: Replies to customers leave the building. That works well once the answers rest on checked knowledge. Without it, the AI guesses, and the customer is the first to notice.
Marketing
Running in my business: Conny develops hooks, scripts and captions for my Instagram channel, and Eddi cuts finished reels from my raw footage. Susi builds ad drafts and sets up campaigns that are technically forced to start paused; only I switch them on. Sebastian measures my website and writes articles in two languages, and nothing goes live without my approval.
Other examples: newsletter drafts from existing articles, draft replies to Google reviews. How to handle reviews is in Respond to Google Reviews With AI.
Why second: Marketing involves plenty of legwork, but the result is harder to judge than a quote. You only find out days later whether a reel works. That's no reason to avoid it, but it is a reason not to start there if you haven't handed off a single workflow yet.
HR
In my business: no role with its own page. So only examples here.
Examples: drafting job ads, answering questions about holiday rules from the staff handbook, putting together the paperwork for someone's first day.
Why not as a starting point: Personal data in HR needs special protection, and decisions about people don't belong in the hands of an AI. If you use AI to help select applicants, the EU AI Act puts you in high-risk territory, with considerably more obligations from 2 December 2027. What that means for decisions made by an AI is in What Can an AI Lead Role Decide?. This is not legal advice. If your company has a works council, read AI and the Works Council in Germany first.
Website and IT
Running in my business: Ralf builds and modernises websites from existing material, and Falk looks for security gaps in my own systems, only with a written scope.
For a small business: These are projects, not weekly tasks. They make sense when there's a concrete reason, say an outdated website, but they're not how you get started with AI.
At a glance: what pays off first
The order is my judgement based on the four checks above, not a measurement across many businesses. In your business, customer service may come before sales if that's where most of your time goes.
What use-case lists tend to leave out
Many overviews of AI use cases in business are written for companies with their own IT and large amounts of data. A small business has neither a data team nor years of clean data. It has an inbox, an accounting tool and what the owner knows. That is exactly where the use cases that work first come from.
Two things also matter more than the tool: who approves, and which data the AI is allowed to see. In my business, nothing leaves the building without my approval, and sensitive folders such as job applications or health matters are off-limits for Frieda. Which data an AI should see is in AI and Privacy: What the AI Gets to See.
Frequently asked questions
What is an AI use case?
A specific, recurring task where AI does part of the work and a person checks the result. "AI in sales" isn't a use case. "Draft quote from call notes" is.
Which AI use case should a small business start with?
Usually the inbox or quotes. Both come up every week, are already digital and are quick to check. Pick the one that eats most of your own time.
How many use cases should I start at once?
One. If two run in parallel, you won't know at the end of the week whether either worked. What getting started looks like week by week is in Bringing AI to a Small Business.
Do I need a separate tool for every use case?
No. Most of my roles run on Claude, each with its own personnel file and its own limits. What matters isn't the number of tools but that every task is clearly described and someone signs off.
Is an AI use case the same as an AI agent?
No. The use case is the task; an agent is one way to build it. Many use cases in a small business don't need an agent with tools, just a clearly described role that produces drafts and waits for approval. Examples of agents are in AI Agents: 10 Real-World Examples.
Where to go from here
Take the table above and mark the one row that costs your business the most time. Check it against the four points. Whether your business is ready to start, you can check with the self-assessment in AI Readiness Assessment for Small Businesses. How I map and hand off workflows together with companies is on my process optimization page. Ready-made roles like Frieda, Anton and Gustav are available to load in my community Claude Practitioners.
