Insight
Workflow automation examples: 12 patterns and how they fail
Twelve concrete workflow automations across sales, operations, finance, support and staff access, each with its trigger, its steps and how it can go wrong.
Most lists of automation examples stop at “when this happens, do that.” The part that decides whether an automation saves time or quietly causes trouble is what happens when a step fails. So each example below has three parts: what starts it, what it does, and what to watch for.
These are patterns, not client case studies. They are the shapes that come up again and again when a team describes the work it copies from one tool to another.
What is workflow automation?
Workflow automation is software that carries out a sequence of steps between your tools when something happens, following rules you set. A form is submitted, a deal is won, a payment arrives: the automation creates the records, sends the messages and assigns the tasks a person would otherwise do by hand.
It works best where the next step follows from the input by a rule that is right every time. Where something has to be interpreted first, see AI agent or automation: which does your task need?
Sales and enquiries
1. Website enquiry to the right person
Starts when: someone submits the contact form.
What it does: finds or creates the contact in your CRM, assigns an owner by a rule such as the service they chose, posts the enquiry to the team channel and sends the enquirer an acknowledgement with a realistic reply time.
Watch for: duplicate contacts when the same person enquires twice with different capitalisation in their email, and enquiries assigned to someone who is on leave. Match on a cleaned email address, and give every rule a fallback owner.
2. Won deal to project set-up
Starts when: a deal is marked won in the CRM.
What it does: creates the project from a template in your project tool, sets up a shared folder, adds the client to your invoicing tool and books a kick-off task for the project lead.
Watch for: a run that fails halfway, leaving a project with no folder or a client with no invoicing record. Each step should check whether its work already exists, so the whole thing can be run again safely. Decide too what happens if a won deal is reopened.
Operations
3. Approval routing
Starts when: someone submits a request, such as a purchase, a discount or time off.
What it does: routes it by rule. Under a set amount it is approved automatically; above it, it goes to the manager; above a higher amount, to finance as well. Once approved, it becomes a task in your project tool with the request attached.
Watch for: approvers who are away. A request that waits silently is worse than a manual process. Remind after a set time, then escalate to a named deputy.
4. Paid order to fulfilment
Starts when: an order is paid.
What it does: sends the order to the kitchen, warehouse or team that fulfils it, updates stock and tells the customer what happens next.
Watch for: the payment succeeding while the hand-off fails. That is the most expensive silent failure on this list, because a customer has paid for something nobody is preparing. Failed hand-offs need a retry, an alert to a person and a daily check that every paid order reached fulfilment.
5. Bookings kept in step across systems
Starts when: a booking is made, changed or cancelled in any system that takes bookings.
What it does: updates availability everywhere else: your own site, partner channels and the calendar your staff work from.
Watch for: double bookings in the seconds between one system changing and the others catching up, and two systems each believing they hold the true record. Decide which system is the source of truth, and have the others ask it before confirming.
6. The weekly numbers, compiled
Starts when: a set time each week.
What it does: pulls the figures that matter from your CRM, accounting and support tools and posts one short summary to the team channel.
Watch for: a source that fails and reports as zero. A report that says “no new enquiries” when the CRM connection broke is worse than no report. Show missing data as missing.
Finance
7. Invoices drafted from approved work
Starts when: a milestone is approved, or at the end of each month for time-based work.
What it does: drafts the invoice in your accounting tool with the right client, amounts and references, and asks a person to review it.
Watch for: sending without that review. Drafting is safe to automate. Sending money requests to clients is where a person should look first, at least until the rule has proved itself.
8. Payments matched to invoices
Starts when: a payment arrives, through a card processor payout or a bank feed.
What it does: matches it to the open invoice by its reference, marks the invoice paid and flags anything it cannot match for a person.
Watch for: partial payments, several invoices paid at once, and processor fees taken out of a payout before it lands. The rule should handle the cases it understands and hand everything else to a person, not guess.
9. Overdue invoice reminders
Starts when: an invoice passes its due date by a set number of days.
What it does: sends a polite reminder, then a firmer one later, and stops the moment the invoice is paid.
Watch for: chasing a client who has already raised a dispute or agreed a payment plan with your team. Give people a simple way to pause reminders for one invoice.
Support and customers
10. Support requests routed by rule
Starts when: a new support request arrives through a form.
What it does: tags it by the product and topic the customer chose, routes it to the right queue, raises the priority for customers on a support plan and replies with the expected response time.
Watch for: requests that arrive as free-text email with no fields to route on. That is the point where a rule runs out, and where reading the message first, with AI or a person, earns its place.
11. New customer onboarding
Starts when: a contract is signed or an account is created.
What it does: sends the welcome message, creates the customer’s accounts in the tools they will use, gives the account manager a checklist and schedules a check-in two weeks later.
Watch for: a customer who gets a welcome email but no working login. Check that each account was created before sending the message that tells them to use it.
People and access
12. Starters and leavers
Starts when: a start date or leaving date is recorded.
What it does: creates accounts in email, chat and project tools ready for the first day, and removes them on the last, with a checklist for the steps a tool can’t do, such as returning equipment.
Watch for: leavers more than starters. A missed account on day one is an inconvenience; a former employee who keeps access is a security problem. Confirm each removal and keep a record of it.
All twelve at a glance
| # | Example | Starts when | Main risk to design for |
|---|---|---|---|
| 1 | Enquiry to the right person | Form submitted | Duplicates and absent owners |
| 2 | Won deal to project set-up | Deal marked won | A half-finished run |
| 3 | Approval routing | Request submitted | Approvers who are away |
| 4 | Paid order to fulfilment | Order paid | Paid but never fulfilled |
| 5 | Bookings kept in step | Booking changed | Double bookings |
| 6 | Weekly numbers | Set time | Missing data shown as zero |
| 7 | Invoices drafted | Work approved | Sending without review |
| 8 | Payments matched | Payment received | Partial and combined payments |
| 9 | Overdue reminders | Invoice overdue | Chasing a disputed invoice |
| 10 | Support routing | Request received | Free text with nothing to route on |
| 11 | Customer onboarding | Contract signed | Welcome sent before access works |
| 12 | Starters and leavers | Date recorded | Leavers who keep access |
Which of these need AI?
Most of them need none. Each follows a rule that someone on your team could write down today.
AI helps at the edges, where something arrives unstructured: support emails with no form fields (10), payments with no usable reference (8), or supplier invoices that arrive as PDFs rather than data. In each case the useful shape is the same. AI reads the message and turns it into fields, and the rule takes over from there.
What makes an automation reliable?
Five things separate an automation you can leave running from one someone has to keep an eye on:
- Safe to run twice. If a step runs again, it finds its earlier work rather than creating a second record.
- Failures reach a named person. Not a log nobody reads, but a message to someone who knows what to do.
- A record of what ran. When a client asks why they received something, you can answer.
- One source of truth for each piece of data. Two systems that both believe they own a record will eventually disagree.
- An off switch. Anyone responsible for it can pause it without a developer.
These can be built in a no-code tool such as Zapier, Make or n8n, or in code when the volume, the running cost or the checks call for it. The five points apply either way.
Where do you start?
Find the copy-and-paste
Pick the task someone on your team does by hand every week, moving the same information from one tool to another.
Write the rule down
If you can write what should happen next, and it is right every time, it can be automated. If you can't, that step needs a person or an agent.
Decide what happens when it fails
Who is told, what is retried, and what a person checks. Settle this before anything is built.
Our workflow automation service connects the tools you already use and builds in the checks and recovery above. If you have a task in mind, send a brief with one real example of it, and we will tell you whether it needs automation, an agent or neither.