How to use AI in business is easier to see than to explain. Here are six businesses of different kinds and sizes, the tasks each gave to AI, and the person who checks each one.
In this lesson
- Six businesses of different kinds and sizes, and the tasks each gave to AI.
- The kind of tool used for each task, and who checks it.
- What stayed human in each, and one thing that went wrong at first.
- How to open each business in the team planner and see it at all three settings.
These six are illustrative examples, not real businesses. Each is a composite, put together to show common patterns. Tools are described by type, not by product; where we have a page of tools for a role, it is linked. The figures for each business come from the planner’s live data when you open it, not from this page.
1. A dental practice (8 staff)
An illustrative example, not a real business.
Two dentists, two hygienists, two dental assistants and two receptionists. The phones never stop at 8:30 in the morning, and missed calls mean empty chairs.
- Dentist (owner)
- Associate dentist
- Hygienists ×2
- Dental assistants ×2
- Receptionists ×2 — AI phone answerer (checks: lead receptionist)
| Task | Outcome | Tool type | Who checks |
|---|---|---|---|
| Answering calls out of hours and at peak times | Hand it over, with spot checks | AI phone answerer | Lead receptionist reviews call summaries daily |
| Booking and moving appointments | Hand it over, with spot checks | AI phone answerer linked to the booking system | Receptionists |
| Appointment reminders | Hand it over, with spot checks | Automated messages | Receptionists |
| Drafting insurance claim paperwork | AI drafts, a person checks | Assistant inside the practice software | Dentist signs |
| Writing up treatment notes from dictation | AI drafts, a person checks | Voice transcription | The treating dentist or hygienist |
What stayed human: every clinical decision, the patient in the chair, explaining treatment and costs, and anyone anxious or in pain on the phone.
What went wrong at first: the phone answerer told a caller with a swollen face to book “the next routine slot” in two weeks. The practice added a rule: any mention of pain, swelling or bleeding goes straight to a person, and reviewed the call log every day for the first month.
See AI tools for receptionists. Try it in the planner: Low · Medium · Scorched earth
2. An accountancy (5 staff)
An illustrative example, not a real business.
Two accountants (one is the owner), a tax preparer, a bookkeeper and an office administrator, serving about two hundred small-business clients.
- Accountant (owner)
- Accountant
- Tax preparer
- Bookkeeper — AI receipt and invoice capture (checks: Bookkeeper)
- Office administrator — AI drafting assistant (checks: whoever sends)
| Task | Outcome | Tool type | Who checks |
|---|---|---|---|
| Capturing receipts and invoices into clients’ books | Hand it over, with spot checks | Document capture in the accounting software | Bookkeeper samples each client monthly |
| Suggesting bank reconciliation matches | AI drafts, a person checks | Built into the accounting software | Bookkeeper |
| First-draft client emails and meeting summaries | AI drafts, a person checks | Business plan of a chat assistant | The accountant who sends it |
| Summarizing new tax guidance | AI helps when needed | Chat assistant, with the guidance attached | Accountant reads the source |
| Chasing missing documents before deadlines | AI drafts, a person checks | Practice software reminders | Office administrator |
What stayed human: advice, sign-off on every return and set of accounts, and the client relationship.
What went wrong at first: a draft email to a client quoted last year’s filing deadline, taken from an old template the assistant had been given. The firm now keeps a dated “facts sheet” that every draft must use, and the accountant checks dates in every email.
See AI tools for accountants and for bookkeepers. Try it in the planner: Low · Medium · Scorched earth
3. An online shop (3 staff)
An illustrative example, not a real business.
The owner, one customer service person and one person packing orders, selling homeware through their own site and two marketplaces.
- Owner — AI drafting assistant (checks: Owner)
- Customer service — AI chat for order status (checks: Customer service)
- Packing and stock
| Task | Outcome | Tool type | Who checks |
|---|---|---|---|
| Product descriptions for new lines | AI drafts, a person checks | Chat assistant or the shop platform’s built-in AI | Owner |
| “Where is my order?” questions | Hand it over, with spot checks | Website chat bot connected to order tracking | Customer service reads a sample daily |
| Sorting and tagging incoming emails | Hand it over, with spot checks | Help desk software | Customer service |
| Forecasting what to reorder | AI drafts, a person checks | Stock software forecasts | Owner approves every order |
| Replies to reviews | AI drafts, a person checks | Chat assistant | Owner |
What stayed human: supplier relationships, refunds above a set amount, anything a customer is upset about, and decisions about the brand.
What went wrong at first: the chat bot told a customer a sale price applied to an item it did not. The owner honored it, then limited the bot to order status and shop policies, with everything else passed to a person.
See AI tools for customer service. Try it in the planner: Low · Medium · Scorched earth
4. A plumbing and heating firm (10 staff)
An illustrative example, not a real business.
The owner, an office administrator, a dispatcher, four plumbers and three heating engineers, covering a city and its suburbs.
- Owner
- Office administrator — AI quoting assistant (checks: Owner)
- Dispatcher — AI scheduling and route planning (approves: Dispatcher)
- Plumbers ×4
- Heating engineers ×3
| Task | Outcome | Tool type | Who checks |
|---|---|---|---|
| Turning site notes and photos into written quotes | AI drafts, a person checks | Job management software or a chat assistant | Owner checks price and scope |
| Scheduling jobs and planning routes | AI drafts, a person checks | Scheduling software | Dispatcher approves the day’s plan |
| Follow-up messages and service reminders | Hand it over, with spot checks | Automated messages | Office administrator |
| Answering calls when everyone is on site | AI drafts, a person checks | AI phone answerer taking messages | Dispatcher calls back |
| Writing up job sheets from voice notes | AI drafts, a person checks | Voice transcription | The engineer who did the job |
What stayed human: all work on site, diagnosis, gas and safety sign-off, and judging what a customer really needs.
What went wrong at first: the scheduling tool packed the day so tightly that every overrun made the next three jobs late. The dispatcher now adds buffer time and approves the plan each morning instead of letting it send automatically.
Read our guide to AI and trades careers. Try it in the planner: Low · Medium · Scorched earth
5. A restaurant (20 staff)
An illustrative example, not a real business.
A manager, a shift supervisor, five cooks, three kitchen prep staff, eight servers and two hosts, open six days a week.
- Manager — AI drafting assistant for rotas, orders and review replies (checks: Manager)
- Shift supervisor
- Cooks ×5 and prep staff ×3
- Servers ×8 and hosts ×2
| Task | Outcome | Tool type | Who checks |
|---|---|---|---|
| Drafting the weekly rota from availability | AI drafts, a person checks | Scheduling software | Manager |
| Suggesting supplier orders from sales and stock | AI drafts, a person checks | Point-of-sale or stock software | Manager or head cook |
| Replies to online reviews | AI drafts, a person checks | Chat assistant | Manager approves every reply |
| Taking bookings by phone and online | Hand it over, with spot checks | Booking system with an AI phone answerer | Hosts |
| Translating menus and allergen notes | AI drafts, a person checks | Chat assistant | Manager checks allergens against recipes |
What stayed human: cooking, service, hospitality, allergy questions at the table, and dealing with any complaint in person.
What went wrong at first: a drafted reply to a bad review was polite but generic, and the reviewer posted it as proof nobody had read their complaint. Replies now start from the manager’s own notes on what happened.
Try it in the planner: Low · Medium · Scorched earth
6. A marketing agency (15 staff)
An illustrative example, not a real business.
Two directors, four marketing specialists, three writers, three designers, two PR specialists and a web designer, working for about thirty clients.
- Directors ×2
- Marketing specialists ×4 — AI research and reporting assistants (checks: account lead)
- Writers ×3 — AI drafting assistant (checks: writer, then editor)
- Designers ×3 — AI image tools (checks: designer)
- PR specialists ×2 and web designer
| Task | Outcome | Tool type | Who checks |
|---|---|---|---|
| Research summaries on a client’s market | AI drafts, a person checks | Chat assistant with web search | Specialist checks every source |
| First drafts of blog posts and emails | AI drafts, a person checks | Business plan of a chat assistant | Writer, then a second reader |
| Monthly performance reports | AI drafts, a person checks | Built into analytics and reporting tools | Account lead |
| Image variations and resizing | Hand it over, with spot checks | Image tools in design software | Designer |
| Meeting notes and action lists | Hand it over, with spot checks | Meeting transcription | Attendees |
What stayed human: strategy, client judgment, final creative work, and the relationships that win and keep clients.
What went wrong at first: a research summary cited a statistic that did not exist. The agency’s rule is now that every figure in client work has a link a person has opened.
See AI tools for marketing. Try it in the planner: Low · Medium · Scorched earth
How to use AI in business: what the six have in common
- They started with frequent, checkable tasks: reminders, capture, first drafts. None started with a decision.
- Every AI task has a named checker, and the checker is usually the person who did the task before.
- The first snag was always the same kind: the AI answered something it should have passed on, or used out-of-date information. Each fix was a rule about what goes to a person, plus a source of current facts.
- Hands-on work, judgment and relationships stayed human in every one, from the dental chair to the client pitch.
Key terms
- Human in the loopA named person who checks or approves AI’s work before it takes effect, and answers for the result.Lesson 10: Risks and rules · More in the glossary
- Retrieval (RAG)Giving a model your own documents to work from, so it answers from them rather than from memory. RAG stands for retrieval-augmented generation.More in the glossary
- HallucinationWhen AI produces something that sounds right but is false, such as an invented fact, policy or source.Giving it your documents, asking for sources and checking its work make it much rarer and easier to catch.Why AI makes things up · More in the glossary
Every term is explained in the glossary.
Try this
Pick the example closest to your business and write your own version of its table: five tasks, the outcome for each, the kind of tool, and who checks. Then add a sixth row for the thing most likely to go wrong in your business, and the rule that would catch it. Open the AI team planner with your own roles and compare its suggestions with your table.
Sources
This lesson uses no outside figures. All six businesses are illustrative composites. Task ratings in the planner and on the job cards are our own, printed live from the current release.