← Back to blog
Automation

The complete guide to business process automation

Everything you need to know before launching an automation project in an SME: where to start, which mistakes to avoid, how to quantify ROI.

ThéoThéo · CTO Lucid-Lab
Updated April 28, 20264 min read

Automation has become the most profitable lever for a French or Belgian SME: fewer manual tasks, fewer errors, more capacity without hiring. But 60% of automation projects fail: not because the tech is bad, but because the right question was never asked: what are we automating, and why?

This guide condenses what we have learned by shipping dozens of automation systems for SMEs. No theory, just concrete advice.

What does it mean to "automate a process"?

Automating a process means replacing a chain of repetitive manual actions with a system that executes them on its own, end to end, with a predictable result.

An important nuance: automating ≠ digitalising. Digitalising means putting a Google Sheet where the paper used to be. Automating means making sure that Google Sheet fills itself, triggers the right actions, and alerts someone when something goes wrong.

Which processes to automate first

Three criteria to identify a good automation candidate:

  1. Volume: at least 20 times a month. Below that, the ROI is rarely there.
  2. Repeatability: same structure every time. If every case is unique, automating is a trap.
  3. Clear rules: you can write what needs to be done in 10 lines. If it starts with "it depends on the context", you first need to clarify the process.

The best candidates in a typical French SME:

  • Invoicing and dunning of unpaid invoices
  • Client onboarding (account creation, documents, access)
  • Weekly reporting
  • Lead qualification and routing
  • Document processing (extracting info from PDFs/emails)
  • Synchronisation between tools (CRM ↔ accounting ↔ project management)

The 4 phases of a successful automation project

Phase 1: Diagnose

We observe the current process. Not as it is documented: as it is actually done. The gap between the two is often where the value hides.

Outputs: map of the current process, list of friction points, real volume measured, manual cost quantified.

Phase 2: Map

We design the target process. What to automate, what to leave manual (sometimes that's smarter), which systems to connect, which rules to encode.

Outputs: target architecture, technical stack choice (Zapier/n8n/custom code), cost and timeline estimate.

Phase 3: Build

We build. In short iterations (1, 2 weeks), we deploy to production quickly, even partially, to observe real behaviour and adjust.

Outputs: operational system, monitoring, technical documentation.

Phase 4: Automate

We industrialise. Logs, alerts, error handling, ownership transfer. At this stage, we should be able to disappear without anything breaking.

Outputs: runbook, team training, optional support.

How much does it cost?

Realistic ranges in 2026:

  • Simple automation (1, 3 connected systems, basic rules): €1,500: €5,000
  • Intermediate automation (conditional logic, AI to process text): €5,000: €15,000
  • Heavy automation (critical systems, high volume, ERP integrations): €15,000: €30,000+

For a detailed breakdown, see our analysis of automation costs.

The 5 mistakes that kill an automation project

  1. Automating a poorly-defined process. If you cannot run it correctly by hand, automating won't save it: it will amplify the problem.
  2. Picking the tool before the need. Zapier is great, except when you need code. n8n is powerful, except when you need an SLA. The right tool depends on the case.
  3. Underestimating the data. 70% of an automation project is cleaning and structuring the input data.
  4. No monitoring. An automation without alerts is a ticking time bomb.
  5. No documentation. If the person who built it leaves, you'll pay twice.

And what about AI?

Generative AI (GPT-4, Claude) opens up cases that were impossible to automate 18 months ago: extracting info from unstructured PDFs, classifying emails, generating meeting notes, qualifying leads from free-form text.

But AI is not a magic wand. It plugs into a classic automation, with safeguards (human validation on critical cases, fallback when hallucinations happen, traceability of decisions).

To go deeper: our AI for SMEs guide.

How to start, concretely

If you're reading this thinking "OK, we should automate something here", here is the order of action:

  1. List your 5 most time-consuming manual tasks
  2. For each one, estimate: monthly volume × hours × hourly rate = annual cost
  3. Keep the 2 where the annual cost exceeds €10,000
  4. Book an Audit Flash: we look at whether it's feasible, at what cost, in what order

No PowerPoint roadmap. No consulting firm handing you an 80-page report. We look, we quantify, you decide.

One article a month, nothing else.

Real automation and AI cases in SMEs. Zero spam.

Read next