What Can a Business Automate With AI? 25 Concrete Examples

Most repetitive work does not live inside one application. It lives in the handoffs between them. Here are 25 workflows we see in small and mid-sized businesses, each described as a trigger, an action and an output, followed by a method for deciding which one to build first.

How to read these examples

A trigger is the event that starts the workflow. An action is what the system does, which may include a bounded AI step such as classifying, extracting or drafting. The output is what the business receives. Where the action writes to a customer or a system of record, assume a human approval step sits before it. None of these are client case studies; they are patterns.

Sales

1. Enquiry to CRM. Trigger: web form or inbound email. Action: classify request type, enrich company data, de-duplicate, assign owner. Output: clean CRM record and an owner notification within minutes.
2. Lead qualification draft. Trigger: new CRM lead. Action: score against your fit criteria, draft a tailored first reply. Output: reply for the owner to approve and send.
3. Meeting booking. Trigger: approved reply. Action: send the right calendar link, confirm, create the calendar event with context. Output: booked call with notes attached.
4. Follow-up sequences. Trigger: proposal sent, no reply after N days. Action: draft a follow-up referencing the proposal. Output: queued message for approval; deal stage updated.
5. Call and meeting notes. Trigger: recording or transcript available. Action: summarise, extract next steps and objections, update the deal. Output: CRM note and task list.

Customer support

6. Ticket triage. Trigger: new ticket or support email. Action: classify intent and urgency, detect language, tag and route. Output: prioritised queue; nothing sits unread.
7. Order-status answers. Trigger: “where is my order” ticket. Action: look up the order in the shop system, draft a reply with the real status. Output: agent-approved answer in one click.
8. Draft responses from a knowledge base. Trigger: classified ticket. Action: retrieve relevant articles, draft an answer with sources. Output: draft for the agent; escalation when confidence is low.
9. Return and refund intake. Trigger: return request. Action: validate against policy, prepare the case, request approval above a threshold. Output: pre-checked case for a person to decide.

Finance and admin

10. Invoice intake. Trigger: PDF arrives in a mailbox or folder. Action: extract supplier, amounts, dates and line items; match the supplier; suggest coding. Output: booking-ready record with the original attached.
11. Approval routing. Trigger: invoice above a limit. Action: route to the right approver, remind, escalate on timeout. Output: approved or rejected with an audit trail.
12. Expense receipts. Trigger: photo or PDF submitted. Action: extract, categorise, check policy. Output: expense line in the accounting tool.
13. Payment reminders. Trigger: invoice overdue. Action: draft a reminder in the right tone for the stage. Output: reminder sent after approval; status logged.
14. Contract data capture. Trigger: signed contract received. Action: extract parties, dates, notice periods and values. Output: contract register entry and renewal reminder.

Operations

15. Client onboarding. Trigger: deal marked won. Action: create project, folders and channels; send the welcome pack; schedule the kick-off. Output: onboarding checklist started the same day.
16. Shared inbox sorting. Trigger: email in a team mailbox. Action: classify, forward to the owner, create a task where needed. Output: empty shared inbox; tasks in the right place.
17. Approval workflows. Trigger: request form (purchase, leave, access). Action: route by rules, remind, record the decision. Output: decision logged and the requester informed.
18. Appointment reminders. Trigger: appointment in 24 hours. Action: send confirmation, process replies, update the calendar. Output: fewer no-shows and a current schedule.

Marketing

19. Content approval. Trigger: draft ready. Action: check against a style guide, route to reviewers, collect comments. Output: approved version with a change log.
20. Campaign reporting. Trigger: Monday morning. Action: collect ad, email and web data; summarise changes; flag anomalies. Output: short written brief and refreshed dashboard.
21. Review monitoring. Trigger: new public review. Action: classify sentiment and topic, draft a response. Output: response for approval; recurring issues tagged.

HR administration (without decisioning)

We automate administrative steps in HR. We do not build workflows that make or recommend hiring, firing, promotion or performance decisions; those are high-risk uses under the EU AI Act and require a different kind of project.

22. New-starter setup. Trigger: start date confirmed. Action: create accounts, request equipment, schedule inductions, send documents. Output: everything ready on day one.
23. Leave and absence admin. Trigger: approved leave request. Action: update calendar, notify the team, adjust the rota. Output: consistent records without re-typing.

Management reporting

24. Weekly operations brief. Trigger: schedule. Action: pull pipeline, support, finance and delivery figures; compare with last week; write three paragraphs. Output: a brief a manager can read in two minutes.
25. Anomaly alerts. Trigger: metric outside its normal range. Action: check related data, summarise the likely cause. Output: alert with context, not just a number.

Start here: how to prioritise

Twenty-five options is too many. Score each candidate on four factors, then build the one with the best score first.

Factor	Question	Score
Frequency	How often does this happen per month?	1 = rarely, 5 = daily or more
Hours	How many person-hours does it consume per month?	1 = under 5 h, 5 = over 40 h
Rule clarity	Can someone write down how it is done today?	1 = it depends, 5 = documented steps
Risk (inverted)	What happens if it goes wrong?	1 = customer or legal harm, 5 = easily reversible

Multiply the four scores. A high-frequency, high-hours, clearly ruled, low-risk process scores 625 and is your pilot. A process that happens rarely, depends on judgement and carries legal risk scores in single digits and should wait, however interesting it sounds. The highest scores usually land in sales intake, invoice intake, ticket triage or reporting, which is why our example workflows live there.

Risks worth naming

Automating a process nobody has measured means you cannot show it worked. Automating a process that changes every month means constant rework. And automating a decision about a person, rather than a task around a person, moves you into regulated territory that needs specialist review before anything is built.

Next actions

• List the five handoffs your team complains about most, and score them with the table above.
• For the top one, record a baseline: monthly volume, hours, error rate and response time.
• Run the free Opportunity Scan to check fit and get a package suggestion, or estimate payback with the ROI calculator.