- Five proven flows: sorting incoming email, answering quote requests, meeting minutes, competitive watch, invoice reminders.
- Each is built in half a day with n8n or Make, a language model and your existing tools.
- Typical cost: 20 to 60 euros a month per flow, all included.
- The golden rule: the automation prepares, a human validates. For at least the first three months.
Before starting: three rules
First rule: automate what you already do by hand at least ten times a week. Otherwise the gain does not cover the setup. Second rule: keep a human in the loop at first. The flow prepares a draft, a person validates in one click. Third rule: one person owns each flow, with their name in the title. A flow without an owner dies within three months.
On tools, n8n (self-hostable, very flexible) and Make (simpler, fully online) do the same job: connect your applications and call a language model at the right moment. Pick the one the owner prefers. That is the only criterion that matters.
1. Sorting incoming email
The flow reads every email arriving at the company's generic address, asks the model to classify it (sales enquiry, support, invoice, application, spam) and extract the key information, then forwards it to the right person with a two-line summary. What it brings: an hour a day for the person who used to sort, and sales enquiries handled the same day instead of two days later.
The classification prompt must contain five real emails from your company, one per category. Without examples the model classifies "roughly". With them, it exceeds 95% correct routing.
2. Answering quote requests
When a quote request arrives through a form, the flow extracts the need, compares it with your price grid (a simple shared table) and writes a personalised draft reply, with a price range and two qualifying questions. The draft waits in the salesperson's inbox, who adjusts and sends it. What it brings: a response time that drops from two days to two hours, which, in our measurements, clearly raises the conversion rate.
3. Meeting minutes
The meeting is recorded with participants' consent, transcribed, then the model produces structured minutes: decisions, actions with owner and deadline, open points. The minutes are filed in the team space and the actions created in the task tool. What it brings: thirty minutes per meeting for whoever wrote them, and actions that no longer get lost.
4. Competitive watch
Every morning the flow collects the new pages of three to five competitor sites, press releases and published job offers, asks the model for a summary of what changed, and sends a ten-line note to management. What it brings: a watch that is actually read, because it is short and regular, and weak signals (a hire, a new offer) spotted weeks before competitors.
5. Invoice reminders
Each week the flow reads the list of overdue invoices in your billing tool, writes for each a reminder adapted to the delay and the client's history (a first cordial reminder, a firmer second, a third announcing next steps), and submits it for validation. What it brings: payment delays shortened by several days on average, and a thankless task that stops being postponed.
How not to fail
The failures we have seen always have the same cause: a flow installed by a contractor, that nobody in-house understands, and that breaks at the first tool change. Build with the owner, document in one page what the flow does and how to stop it, and plan a ten-minute monthly review. Sensitive data follows the same rule as for any AI use: nothing personal or contractual in a consumer model without a business account.
"An automation that runs without anyone understanding it is not a gain. It is a debt."
Conclusion
Five flows, five half-days, a few dozen euros a month: the ratio between effort and gain is rarely this favourable. Start with the one that hurts most today, measure the time saved for a month, then move to the next. That is how the SMEs we support build, flow after flow, real AI-assisted productivity.
FAQ
n8n or Make, which to choose?
Make to start fast without installing anything; n8n if you want to self-host, keep data in-house and build more complex flows. Both are equally good for these five cases.
Which language model to use?
A mid-range model is enough for these tasks. Keep frontier models for cases where writing quality really matters, such as quote replies.
What about email confidentiality?
Use a business account with the model provider, which guarantees your data is not used for training, and exclude from the flow the mailboxes that receive sensitive data.
How long before a return on investment?
Usually the first month for email sorting and invoice reminders, two to three months for the other flows.
