DIY AI automation for small businesses: myth or reality?
Nearly four in ten French micro-businesses already use AI, but only eight in a hundred use it to automate tasks. Between “asking the chatbot” and “letting it run on its own” sits automation: an email arrives, AI sorts it, the data lands in a spreadsheet and all you have to do is check. We have gathered real cases from small businesses in France, tool prices as of October 2026 and an honest answer: what you can build yourself in an evening, and what is just a marketing myth.
Prices and terms as of 17 October 2026; AI tool pricing changes often. The cases come from public sources, linked in the text; the time savings are as reported by the business owners themselves.
The short answer
Reality
- AI built into services you already pay for
- A 3–5 step workflow that ends in a draft
- One evening to set up, €0–50 a month
Myth
- “Full automation” with no human involved
- “A €50 agent you set up and forget”
- “23 fewer hours of work a week”
Almost none of the cases we found runs “by itself”. What does work is this: AI takes over the routine in the middle of the chain, while a person handles the start and the end. Owners who work this way save anything from a couple of hours a week to 60–90 minutes a day. Not a revolution, but for a one-person business it makes a difference.
How automation differs from a chatbot
In “Five tasks AI can take off your hands today” we looked at manual work: you open the chatbot, paste in an email, get a draft back. Automation means you don’t need to open the chatbot at all. A new email, an enquiry from your website, a review or a call starts the workflow by itself, and you see the finished result: a filled-in row, a draft reply, a reminder in your calendar.
There are two ways to build these workflows:
- AI built into services you already have: Gmail, accounting apps such as Indy, Pennylane or Qonto, booking tools such as Zenchef or Treatwell, a Shopify store. Nothing to connect; you just switch the feature on.
- Workflow builders such as Make, Zapier or n8n: you connect the steps with your mouse, e.g. “email received → sent to AI → saved to the spreadsheet”. No coding needed, but expect to spend an evening learning the ropes.
Five types of business: what you can build yourself
Trades: plumbers, electricians, roofers
- Quote requests. Website form or email → AI pulls out the address, type of job and urgency → a row in the spreadsheet → an automatic reply to the client: “Got it, we’ll call you back.” Make or Zapier, or Workspace Studio in Google Workspace (from €6.80 per user a month).
- Bank transactions. Your accounting app sorts them into categories itself: Indy includes AI categorisation in all plans; Pennylane offers automatic categories from the “Essentiel” plan upwards.
- Missed calls. An AI assistant takes the enquiry and emails you a summary, for example AirAgent: from €89 a month excluding VAT, plus per-minute charges.
Real caseThe roofing firm Jeannolin (Savoie, 5 people) switched to AI quoting software: pricing, sending and e-signature. They save 1–1.5 hours a day, but a person still checks every quote. The German heating firm Heidbrink hands over to AI any call nobody has answered within 20 seconds: for €179 a month (according to the provider), at weekends only 2–3 calls reach the tradesman instead of nearly twenty. The owner himself calls the system “answering machine 2.0”.
The catchCustomers don’t like being put through to a robot without warning. Another German tradesman, according to the integrator, had callers hanging up until a short greeting from the owner was added before the assistant: “my assistant will take it from here.” The prices in your quote are not AI’s job: you check them.
Cafés and restaurants
- Phone bookings. An AI assistant inside the booking system: Zenchef charges €99 a month, according to the provider.
- Replies to Google reviews. New review → draft reply → you edit and publish. Often already built into your website builder or listing tool.
- Food costs. Photo of a supplier invoice → AI lists the items and prices → a spreadsheet that shows what has gone up.
Real caseThe restaurant Le Feugerolles (Loire, 5 people) uses the AI built into its website platform — copy, social posts, review replies — for €45 a month excluding VAT. The hotel Le Rodin in Arles still works by hand: for around €60 a month, the owner uses AI to draft email replies in several languages following his own instructions; everything is reread and there have been no complaints. A good first step before a workflow.
The catchAI loves to invent offers and prices. The American restaurant Stefanina’s had to ask customers to double-check information because AI answers in Google Search promised discounts that didn’t exist. A booking bot must never promise anything that isn’t on the menu.
Beauty salons and hairdressers
- Front desk. An AI receptionist in the booking software: Treatwell from £69 a month (UK price); other platforms have announced similar features.
- Reminders and reviews. Booking → reminder the day before → review request two days later. Usually already in your booking software: just switch it on.
- Instagram and WhatsApp messages. AI answers routine questions (prices, address, free slots) and passes the rest to you.
Real caseTo be honest, we found no French salon cases with figures, only marketing from the software providers themselves. If you already work this way, get in touch and we will add your story.
The catchDon’t send clients’ health information — allergies, contraindications — to AI services. And not every client likes talking to a robot: always leave a way to reach a real person.
Shops and e-commerce
- Product descriptions and draft replies — Shopify’s built-in seller assistant is included even in the Basic plan (€27 a month, billed annually).
- “Where’s my order?” Email → AI finds the order number → draft reply with the delivery status. Make or Zapier.
- Returns and reviews. Once a week, AI collects the reasons for returns in a spreadsheet, so you can see which product is badly described.
Real caseThe closest case is the magazine L’épicentre in Blois (4 staff plus freelance journalists): a Make workflow costing €100–150 a month sifts through 50–100 press releases a day, picks the relevant ones and prepares drafts. Each article takes 20 minutes instead of an hour, and the editor has the final say. The same logic — “stream → filter → draft → human” — works for a shop’s customer emails.
The catchNever leave discounts, prices or return policies to AI. Air Canada’s chatbot made up a refund rule, and a small claims tribunal ordered the airline to pay. According to a British shop owner on Reddit (this cannot be verified), his overnight bot invented an 80% discount on an order of over £8,000, and the order had to be cancelled.
Consultants, freelancers and the liberal professions
- New client. Form → folder in Google Drive → contract from a template → reminders about missing documents.
- Meetings. Call recording → transcript and summary → note on the client record.
- Email. Labels and draft replies in Gmail via Workspace Studio, included in Google Workspace plans.
Real caseThe pension advisory firm SAM (2 people) worked with an outside developer to build a multi-agent system on n8n: collecting documents, writing to pension funds, keeping a log. Each case now takes 2.5 hours instead of 6, and the conversion rate rose from 28% to 41%. But at a price: €25,000–40,000 for the first version, three quarters of it covered by grants, plus about €650 a month for services. The French tech site JDN tested Workspace Studio: the AI labelled emails correctly, but the draft replies needed editing.
The catchProfessional secrecy (secret professionnel): France’s data protection authority, the CNIL, advises against sending clients’ personal data to public AI services, recommends checking your contract with the provider and, for sensitive data, preferring solutions that run on your own hardware.
What it really costs
France Num, the French government’s digital programme for small businesses, gives some benchmarks, and the cases we found fit within them:
| Level | What it is | Cost |
|---|---|---|
| 1. AI subscription | ChatGPT, Mistral and similar services, used by hand | €0–25 a month per person |
| 2. Workflows | AI connected to your tools, including via Make, Zapier, n8n | €100–300 a month for a small team |
| 3. Custom project | agents built around your processes, as at SAM | €15,000–50,000 to launch |
Workflow builder prices as of 17 October 2026 (all billed annually): Make has a free plan with 1,000 credits, paid plans from $9 a month; Zapier gives you 100 tasks free, the Pro plan starts at $19.99; cloud-hosted n8n starts at €20 a month, and its free version can be installed on your own server. AI built into your email, accounting or booking software often costs nothing on top of your plan. AI steps use up credits faster than ordinary ones, so keep an eye on the counter in the first few weeks.
The real cost is not money but time. Rouergue Saveurs, a butcher’s shop with a restaurant in Rodez, puts it plainly: what holds them back is not the budget but the time it takes to learn, experiment and fit AI into the work. The training fund you already contribute to may pay for that learning — see our article “Learning AI at the state’s expense”.
Where it breaks
Silent failureA workflow can stop working without warning. For example, if your Gmail connection is set up in Google’s “testing” mode, access expires after 7 days, and enquiries quietly stop arriving in the spreadsheet. Once a week, check that the latest entry is recent.
Made-up answersAI confidently states things that aren’t true: prices, deadlines, rules. Anything involving money or commitments to a client must go through your review.
Self-hostingThe free self-hosted version of n8n needs updating. In January 2026, a vulnerability with the highest severity rating was found in it: if forms or webhooks were exposed to the internet, an outsider could reach the server without a password. If you’re not ready to keep up with updates, use the cloud version.
Someone else’s voiceThe flip side: scammers call customers with an AI voice in the name of real companies. Nova Énergie, a company in Brittany, received up to 30 angry calls a day from people who had been called in its name. Tell your customers how you contact them.
A word on big promises. In a video published by France Num, the roofing firm Tri-Énergie says its n8n workflow saves “about 23 hours a week”, but in a note to the transcript France Num points out that 2–3 hours may have been meant. For comparison, a 2024 survey of tradespeople found savings of about 2 hours a week among those already using AI. Figures like that are a realistic benchmark.
Rules you need to know
- AI must introduce itself. Since 2 August 2026, the EU AI Act (Article 50) has required that people know they are dealing with AI unless it is obvious. The obligation falls on the system’s provider; if you built the bot yourself and run it under your own name, it is safer to assume it applies to you too. “Hello, I’m the salon’s virtual assistant…” and you’re covered.
- Call recording. If an AI assistant records the conversation, CNIL rules require you to say so at the start, explain why and, as a rule, not keep recordings for longer than six months.
- Customer data. Turn off the use of your data for training in the service settings, choose plans that come with a data processing agreement and, ideally, with storage in the EU. More on this in “What never to feed an AI”.
Where to start this week
- Write down three repetitive tasks that take you more than an hour a week.
- Check whether the services you already pay for have an AI feature for them: email, accounting, bookings, website.
- If not, build a single workflow on the free plan of Make or Zapier. The output should be a draft or a row in a spreadsheet, not something sent to a customer.
- Check every result for two weeks. No mistakes? Keep it. Mistakes? Refine the instructions you give the AI.
- Only then move on to the next workflow.
If you’d like help building your first workflow from someone who knows both AI and small business in France, it is one of the topics covered in mentoring.
Information correct as of 17 October 2026. Sources: France Num case studies — Jeannolin, SAM, L’épicentre, Le Rodin, Le Feugerolles, Rouergue Saveurs, Tri-Énergie; France Num on AI budgets; France Num Barometer 2026 (pp. 53–54); Deutsches Handwerksblatt; JDN; Air Canada tribunal decision; CNIL — generative AI, call recording; EU AI Act. Prices taken from the services’ official pages on 17 October 2026. Webinkub is not affiliated with any of the services mentioned.
Want to build your first workflow?
Tell me which task eats up most of your time. I’ll let you know whether it can be automated and which tool to start with.
Want your whole situation looked at?
Mentoring covers the whole path: from registration to reporting and choosing your tools.