10 automations for a 50-person company

In a 50-person company, nobody is hired to copy an order into the invoicing system, and yet someone does it every week. The same actions come back from one department to the next: an enquiry to retype, a quote to chase, an invoice to match, an access to cut off on someone’s last day.

Here are the ten flows we recommend looking at first, team by team. For each one: what it prevents, what stays with a person, and the step where AI helps, when it helps. Most of them do without it, which is the position of our guide Automation without AI: when a rule is enough.

How did we choose these ten flows?

Four criteria, no fewer. The task comes back every week. It crosses at least two tools, so someone copies. A mistake there is costly: a lost enquiry, an invoice paid twice, an access left open. And its rule fits on one page, so a new colleague could apply it.

Marketing has its own article, marketing automation for small businesses, with or without AI, so we leave it aside here. Here is how the two families split:

Sales: no lost enquiries, no forgotten quotes

1. The incoming enquiry assigned to the right salesperson

The website form creates the contact and the deal in the CRM, then a rule assigns the salesperson by region, language or rotation, and notifies them in the tool they already have open. With no response after the delay you set, their manager gets an alert. No model is involved: the form fields are already structured.

One rule is not negotiable: the enquiry is stored before any notification, so a failed email cannot make it disappear. What stays human: the first conversation, which often makes the sale. The full flow is described in the incoming enquiry flow, from form to salesperson.

2. The reminder for an unanswered quote

A quote goes out, then nothing. After the number of days you set, the salesperson receives a task with the quote and the history of exchanges, and the reminder stops on its own as soon as the quote is signed or declined in the CRM. What stays human: the reminder message itself, which depends on the relationship. Above all, the flow prevents a quote being forgotten in an outbox while the customer signs elsewhere.

3. The customer created everywhere on signature

When a deal moves to “won”, the customer already exists in the CRM. Now they need to exist in invoicing, in the project tool and in the calendar of the team that is starting.

What stays human: checking any special terms (a discount, a payment schedule) before the first invoice. What comes next, from signed quote to payment, is the subject of our article on the flow from signed quote to paid invoice.

Customer service: every request in the right place

4. The shared inbox, sorted

Shared addresses (contact, support, accounts) mix quotes, complaints, invoices and emergencies. Here AI has its place, at one step: reading a freely written email and filing it into a closed list of categories, with a confidence level. Above the threshold, the message goes to the right department; below it, a person decides.

This sorting does not need a large model. A small model, or a decision model that picks from a list instead of writing, is enough and costs little; our guide on what AI costs a business does the maths for two thousand emails a month. What stays human: ambiguous cases, and any reply to a complaint.

5. The reply drafted from your documents

For recurring questions (lead times, terms, procedures), AI finds the answer in your documents and drafts a reply that cites its source. The person handling the request reviews, corrects and sends. The gain comes from the search avoided, not from automatic sending, and quality depends on your documents: an up-to-date knowledge base does more for the answers than the choice of model. The split of roles is detailed in automating customer support: AI sorts, humans review. What stays human: sending, and any commercial gesture.

Finance: less data entry, fewer oversights

6. The supplier invoice, read and matched

The PDF arrives by email, with a layout that changes from one supplier to the next. A model extracts the references and amounts, then the calculation goes back to code: the amount is compared with the purchase order, and a discrepancy goes to review instead of being paid. Our guide on automation without AI details this split, step by step. What stays human: approving a discrepancy or a new supplier.

7. The expense claim sent as a photo

The person photographs their receipt from their messaging app, and the record prepares itself.

What stays human: that approval, which we never remove from a flow that touches money.

8. Reminders for unpaid invoices

No AI here: a missed due date is a fact, not an interpretation. The flow sends a first polite reminder a few days after the due date, a second with the invoice attached, then creates a task for a person before any formal notice. It stops as soon as the payment is recorded in invoicing.

In France, reminders can mention what the law provides between businesses: according to the Service Public page on payment terms, checked on 7 August 2026, a flat €40 fee for recovery costs applies to every invoice paid late (Article D441-5 of the French Commercial Code). What stays human: the formal notice, and any payment plan granted to a customer in difficulty.

Team and leadership: access and figures

9. An employee’s arrival and departure

An arrival triggers a list: the email account, tool accounts, equipment, rights on shared folders. A departure triggers the reverse list, and that one matters most. France’s national cybersecurity agency, ANSSI, makes it one of the measures in its IT hygiene guide:

The flow starts from a record in the HR tool, creates each department’s tasks and checks they are done. What stays human: deciding everyone’s rights, and checking the list on the last day.

10. The Monday morning report

On Monday at 8 am, a message gathers the week’s figures in the leadership team’s channel: enquiries received, deals signed, revenue invoiced, unpaid invoices, open tickets. They come straight from the tools, with no copying and no intermediate spreadsheet. An online plush toy shop we work with receives its summaries this way every day.

AI can then comment on the variations in plain language, but it calculates nothing: the figures come out of the tools, and the commentary gets reviewed. Our article on automated marketing reporting covers which figures to track and which traps to avoid. What stays human: the decisions the report is meant to inform.

What these ten flows have in common

One source of truth per piece of data. The customer lives in the CRM, the invoice in invoicing, the employee in the HR tool. The other tools read a copy, they do not correct it.

An alert when a flow goes quiet. An automation that stops without warning is worse than a manual task: nobody knows the work was not done. n8n lets you attach to each workflow an error workflow, triggered on every failure, and our guide to monitoring your n8n workflows describes the rest of the setup.

A known cost. A self-hosted automation costs only the server that runs it; an AI step is paid on every pass, and can be measured and capped. The choice of tool matters too, since Zapier, Make and n8n do not bill the same unit: our comparison n8n, Make or Zapier for a small business goes into detail.

Where to start?

1. 01Count.

   Ask each team what it copies from one tool to another, and how much time that takes each week.

2. 02Choose.

   Pick the flow that costs the most time and whose rule fits on one page.

3. 03Measure first.

   Note the time spent and the current errors: without a starting point, no gain can be proven.

4. 04Run in parallel.

   The old way of working stays available until the new one has proven itself.

5. 05Move to the next.

   One flow that runs and is monitored beats five half-done.

Our n8n automation offer, with or without AI starts from this method, from €2,400: we watch the task with the person who does it, then deliver the flow with its alerts and a meter showing its cost. If everything seems to need automating at once, our digital and AI maturity audit first reviews your processes and tools to say in which order to move. And if your teams live in their messaging app, a Telegram or WhatsApp business bot becomes the front door for flows 7 and 9.

Frequently asked questions

Sources

1. The Service Public page on payment terms (new tab)entreprendre.service-public.gouv.fr
2. ANSSI IT hygiene guide (new tab)messervices.cyber.gouv.fr
3. N8n documentation on error handling (new tab)docs.n8n.io