“We want to automate, but we don’t know where to start.” It is the sentence we hear most often in SMEs. There is no shortage of ideas, quite the opposite: every department has its list of tedious tasks. The real risk is starting with the wrong process, one that looks impressive but turns out too vague, too rare or too dependent on special cases to deliver quick results.

This article offers a simple method to choose, then reviews five processes found in almost every SME that usually make excellent first projects.

The four criteria of a good first process

  • Volume: the task comes up often, every day or every week. A yearly task, however tedious, rarely pays back the effort.
  • Known rules: you can explain to a newcomer how to do it. If each case needs an expert decision, automation will be hard.
  • Stable inputs: information arrives through identified channels and in roughly predictable forms, even if not perfectly standardized.
  • A measurable outcome: you can measure the time spent today, the lead time, the number of errors. Without a baseline, there is no way to prove the gain.

AI has widened what is possible: inputs that could not be automated yesterday, such as freely written emails or documents with varied layouts, can be today. But these four criteria remain the best indicators of success.

A simple scoring method

For each candidate process, give a score from 1 to 3 on each of the four criteria, then add two questions: what is the impact of an error (the lower, the better for a first project), and are the people who do the task ready to get involved? Rank the processes by total. Those at the top are your best candidates; those with a low score on rules or measurement are for later.

This exercise takes half a day with the managers concerned. It avoids weeks of work on a poorly chosen project.

The best first automation project is not the most ambitious one. It is the one whose success you can measure within a few weeks.

1. Supplier invoice processing

Regular volume, known accounting rules, easy-to-measure outcome: it is an almost ideal candidate. AI extracts data from invoices whatever their layout, rules check amounts, duplicates and bank details, and only anomalies go to a person. We describe this chain in a dedicated article.

2. Sorting incoming emails

Shared mailboxes (contact, support, orders) account for a large share of time lost to sorting. Automatically classifying messages, extracting key information and routing them to the right person frees up time from the first weeks, without touching the customer relationship: replies are still approved by the teams.

3. Recurring reporting

The weekly or monthly report someone builds by hand, copying figures from several tools, is an excellent candidate. Data is collected automatically, indicators are computed by code, and a language model can draft a first version of the commentary for the manager to review. The gain is twofold: time saved, and more reliable figures.

4. Onboarding a new employee

Creating accounts, ordering equipment, sending documents, scheduling first meetings, granting tool access: onboarding involves several departments and often relies on a mental checklist. A workflow triggered by the hire coordinates each step, follows up on anything forgotten and gives the newcomer a clear path. Volume is lower than for invoices, but the impact on employee experience is strong.

5. Entering quotes and orders

When quote requests or purchase orders arrive by email or as PDFs, someone retypes them into the ERP or sales tool. Automatically extracting the information (customer, items, quantities, deadlines) and pre-filling documents shortens response times, which directly affects revenue, and cuts data entry errors.

A concrete example

An industrial SME of around sixty people lists about fifteen automation ideas with us. The most cited internally is an assistant able to answer every technical customer question. On the scoring grid, however, it gets an average score: the rules are fuzzy, answers depend heavily on the engineers’ expertise, and the impact of an error is high.

At the top of the ranking come entering orders received by email and the weekly sales report. Two more modest projects, but quick to deliver, easy to measure and appreciated by the teams. Their success builds the trust and budget needed to tackle the technical assistant next, better prepared.

Mistakes to avoid

  • Starting with the most visible process rather than the best-suited one.
  • Automating a process that is not clear: if nobody can describe the steps, they need clarifying first.
  • Launching five projects at once: one clear first success beats a series of unfinished projects.
  • Forgetting to measure beforehand: without a starting point, even a good result will remain debatable.

Our recommendation

List your ideas, score them with the four criteria and the two extra questions, and pick a single process to start. Measure the starting situation, set a clear goal, and deliver a first version within a few weeks. Once this first project succeeds, the next one will be faster: the tools, connections and method will already be in place.