In short
The five processes to automate as a priority in an SME: sorting and answering incoming emails, generating documents (quotes, contracts, invoices), follow-ups (quotes, unpaid invoices, missing documents), internal information search, and meeting summaries and reporting.
- An office worker spends about 28% of their week managing emails and 19% searching for information internally (McKinsey Global Institute).
- Each of these five processes, well automated, saves 1 to 5 hours per week per person concerned; combined, that's often more than a full day.
- On payment delays: 13.6 days of average delay in France at the end of 2024, and 15 billion euros of missing cash for SMEs (Banque de France). Automatic follow-ups attack this problem directly.
- Realistic budget: from 440 € for a simple automation to 1 760 € for a complete business agent, generally paying for itself within a few months.
- Honest limit: automating a shaky process makes it worse. You clean up the process first, then automate.
The problem, concretely
In an SME of 5 to 50 people, no one holds the title of "head of repetitive tasks". Yet they exist: sorting the morning's emails, redoing a quote that looks like the previous one, chasing the customer who hasn't paid, searching for "the 2023 contract" across three tools, writing the minutes that no one will read. The reference study from the McKinsey Global Institute estimates that an office worker spends 28% of their time on emails and 19% searching for information: nearly half the week before even producing anything. The study dates from 2012, but anyone who has opened an SME inbox in 2026 knows it hasn't aged a day.
The good news: these tasks are precisely the ones that AI and automation do well today, because they are frequent, repetitive, and governed by rules. I'll review all five, with for each the before/after walkthrough, the hours saved, the difficulty, and above all the criterion for knowing whether it can be automated at your company.
1. Sorting and answering incoming emails
Before: everything lands in a contact inbox or the owner's inbox. Someone reads, dispatches, answers the mundane questions, forwards the rest. Urgent requests wait behind newsletters.
After: an AI agent reads each incoming email, categorizes it (quote request, customer question, supplier invoice, job application, spam), routes it to the right person, and drafts a reply for questions whose answer is known. The human validates or corrects: they no longer start from a blank page.
Hours saved: 2 to 5 h per week per person handling incoming flow. Difficulty: medium, sorting alone is simple, assisted replies require a clean knowledge base. Automatable if: you receive at least 20 to 30 "actionable" emails per day, 70% of requests fall into recurring categories, and the standard replies exist somewhere (even in someone's head, we write them down first). If every email is unique and committing, it's not a good first candidate. For the specific case of support, I detailed the approach in automating your customer service with AI.
2. Document generation: quotes, contracts, invoices
Before: you duplicate the previous customer's quote, replace the name, forget a line, recalculate the totals by hand, and the document goes out with the old customer's logo one time in twenty. Count on 30 to 60 minutes per somewhat structured document.
After: the data (customer, services, prices) comes from your CRM or a form; the document is generated in a few seconds, compliant with your branding, accurate, and goes out for electronic signature right after. The version, the date, and the status are tracked.
Hours saved: 2 to 4 h per week from a dozen weekly documents onward, more in construction or services where the quote is the lifeblood (see the concrete case quotes and signature in construction). Difficulty: low to medium, it's the most deterministic of the five processes. Automatable if: your documents follow stable templates and the data exists in structured form (spreadsheet, CRM, ERP). If every contract is negotiated line by line by a lawyer, we automate the skeleton, not the negotiation.
3. Follow-ups: quotes, unpaid invoices, missing documents
Before: follow-ups rely on memory and guilt. You follow up too late, or never, because it's a thankless task. National result: 13.6 days of average payment delay at the end of 2024, and 15 billion euros of cash missing from French SMEs according to the Observatoire des délais de paiement.
After: every quote sent, invoice issued, or document requested enters a follow-up cycle: D+7, D+15, D+30, with a tone that adapts (cordial, firm, formal notice prepared for human validation). The agent knows to stop when the customer replies and alerts the human on cases that turn sour.
Hours saved: 1 to 3 h per week, but most of the gain is elsewhere: quotes signed that would have gone nowhere, and cash coming in sooner. Difficulty: low, it's often the best first project. Automatable if: your deadlines are known (quote date, invoice due date) and your follow-up rules can be stated in one sentence. Accounting firms, champions of the missing document, get an immediate benefit: see automating an accounting firm.
4. Internal information search (RAG over your documentation)
Before: "does anyone know where the procedure for... is?" asked in the hallway or on the WhatsApp group. The answer is in a 2022 PDF, on the drive, in a folder only the former manager knew about. New arrivals ask the same questions ten times.
After: an internal assistant answers questions relying solely on your documents (procedures, contracts, product sheets), with the source cited in every answer. This is the RAG technique: the AI is bounded to your documentation and says "I don't know" rather than making things up (I explain the safeguards in how to stop AI from making things up).
Hours saved: 1 to 3 h per week per person, harder to measure but real, especially during onboarding. This is exactly what runs on beforbuild.com, my B2B SaaS: the assistant answers from the business documentation, with sources to back it up. Difficulty: medium to high, the technique is mature but the quality depends on your documentation. Automatable if: the information exists in writing and is roughly up to date. If everything is verbal, the prerequisite work is documentary, not technical. Real estate agencies, where information sleeps in mandates and inspection reports, are a good example: see automating a real estate agency.
5. Summaries and reporting
Before: the meeting minutes get written in the evening (or never), the weekly review takes an hour of copy-pasting between the CRM, the spreadsheet, and the emails, and the owner discovers the figures two weeks late.
After: the AI transcribes and summarizes meetings with decisions and actions, and aggregates the key figures each week (sales, outstanding amounts, tickets) into a summary sent automatically. The human reviews and keeps the analysis: the AI prepares, it does not decide.
Hours saved: 1 to 3 h per week. Difficulty: low for minutes, medium for reporting (you need clean access to the data). Automatable if: the source data is accessible via API or export. If your reporting consists of interpreting weak signals and political context, keep the human at the center: the AI provides the raw material.
How much it costs (and what it brings in)
On the French market, an automation developed custom by an agency commonly bills between 5,000 and 20,000 €, plus maintenance. No-code tools (Make, n8n, Zapier) bring the entry ticket down to a few dozen euros per month, but with a ceiling quickly reached as soon as the logic gets complex or reliability becomes critical.
My positioning is in between: AI agents and automations already built and proven on beforbuild.com, which I adapt to your business. A simple automation (follow-ups, email sorting, document generation on a template) starts at 440 €; a complete business agent (multi-step, connected to your tools, with a supervision interface) at 1 760 €. Fixed prices, announced before we begin.
The return calculation is deliberately simple: a process that saves 3 h per week for a person whose loaded hour costs 30 €, that's about 4,500 € per year. An automation at 440 € pays for itself in less than 5 months, and it keeps running afterward. I invite you to redo this calculation with your own figures rather than take my word for it: that's the purpose of the audit below.
Where to start
Definitely not with all five at once. A single process, the most painful one, and three steps:
- The free 30-minute audit. We list your repetitive tasks, run them through the three criteria (sufficient volume, clear rules, accessible data), and put figures on the hours at stake. If nothing can be automated profitably, I tell you, and we leave it there.
- Clean up the chosen process. This is the honest limit of everything above: automating a shaky process makes it worse, it will produce its errors faster and at a larger scale. Before automating, we write the ideal walkthrough in a few lines; if we can't, the problem isn't technological.
- Automate, measure, extend. Going live on a restricted scope, measuring the hours actually saved and the error rate, then extending to the next process. The first visible success creates momentum; the first quiet failure avoids disaster.
Sources
Frequently asked questions
Have a feature in mind? Let's talk.
30 minutes to scope your need and quote the fixed price. Reply within 24h.
Book my free audit · 30 min