Which processes to automate first in an SME

"Let's automate everything." It is the phrase most projects start with, and it is also the reason so many of them go nowhere. The vast majority of AI projects leave no measurable mark on the company's accounts: not because the technology does not work, but because they start from the wrong process.

The right question is not "what can we automate?". It is "what is worth automating first?". In an SME, where time and budget are limited, choosing the first process well is the decision that determines whether automation becomes an investment that pays for itself or yet another tool abandoned after three months.

This is the guide we use with our clients to make that choice.

The golden rule: start from the back office, not the storefront

The temptation is to start from what people see: marketing, the website, social media. But the most solid returns come from the back office: administration, internal operations, document management. It is there, in the internal and barely visible processes, that the biggest savings hide, even if attention tends to focus elsewhere.

The reason is simple, and it is three things together: internal processes have clearer rules, they use data you already own and they produce a measurable error. A payment reminder sent late has a cost you can quantify; the effect of a post published automatically, far less so.

The five criteria for recognizing the right process

Before even looking at the tools, assess each candidate process against five criteria. The more "yes" answers it collects, the better a starting point it is.

• Frequency. Does it repeat often? Every day, every week? The more frequent it is, the more the time saved accumulates.
• Time volume. How many person-hours does it consume in a month? A 30-second task done 200 times is worth more than a half-day task done once a year.
• Standardization. Does it follow clear, stable rules, or is it full of exceptions and gut-feel decisions? Automation loves rules.
• Data quality. Is the information already digital and tidy (in a management system, a clean Excel, a CRM)? Or is it scattered across paper and email? Dirty data means fragile automation.
• Cost of error. When it goes wrong, how much does it cost? A frequent and costly error is an excellent reason to automate, with the right controls.

The effort / impact matrix

With the five criteria in hand, place each process on two axes: how much impact automating it would have and how much effort it requires. Four quadrants come out.

• High impact, low effort → quick win. This is where you have to start: results visible fast and internal trust that builds.
• High impact, high effort → real projects. Important, but they need planning. They are not the first step.
• Low impact, low effort → fillers. Do them if there is time left over.
• Low impact, high effort → traps. To be avoided, especially at the start.

The first project is there to prove value fast. Choose a quick win, not Everest.

Where SMEs really start: examples by department

In concrete terms, here are the processes that most often become the first "yes" in the companies we work with, broken down by department.

Administration and finance

• Automatic payment reminders when an invoice goes past its due date.
• Recording and routing the incoming invoices that arrive by email.
• Deadline reminders (VAT, contributions, renewals) generated from an Excel sheet.

Before: someone manually checks deadline schedules and inboxes. After: the system alerts the right person at the right time.

Human resources

• Onboarding: checklists, documents and access set up automatically when a new colleague arrives.
• Collecting leave requests and timesheets without having to chase anyone.

Sales

• Automatic follow-up of leads who have not yet replied.
• Updating the CRM from emails, without manual copy-paste.

Operations

• Periodic reports (weekly, monthly) generated and distributed on their own.
• Status updates (orders, cases, tickets) communicated automatically to the customer or the team.

Marketing

• Distributing and republishing content across several channels from a single source.

Notice the common thread: these are all repetitive activities, with clear rules and data that is already digital. That is no coincidence.

Three processes not to automate first

• Unstable processes or ones in constant exception. If the rules change every week, automating them means chasing. Stabilize first, then automate.
• Processes with dirty data. If the information is inconsistent or scattered, automation will only spread the errors, faster. Fix the data first.
• Processes that require judgment or empathy. A delicate negotiation, a strategic decision, an angry customer: here AI assists, it does not replace. More than one company has learned this the hard way, fully automating customer service only to put people back on the most complex cases.

And the tool? It comes after, not before

Once the process is chosen, the tool almost chooses itself. In order of increasing complexity:

• Sometimes it is enough to tidy up an Excel or the management system you already have. Not everything requires AI.
• To connect systems and create flows ("when X arrives, do Y"), tools like Microsoft Power Automate do most of the work.
• For text, summaries, drafts and replies, assistants like Copilot or Claude help people work faster.
• When you need to interpret a lot of data or handle cases that are not perfectly regular, a custom AI agent comes into play (what an AI agent is, and when you really need one).

The most common mistake is to fall in love with the tool and look for a problem for it to solve. Do the opposite: start from the process, the tool is a consequence.

How much it pays, and why it is worth starting small

The opposite temptation to "let's automate everything" is "let's wait for the perfect project". That is wrong too. The right way is to start small and measure:

• Measure the before: how many hours, how many errors, how much that process costs today.
• Automate a single process, done well.
• Measure the after. The delta is your ROI, in numbers, not in promises.

That first result funds and legitimizes the second. It is the principle of "build once, reuse many times": every automation becomes a reusable building block for the next one.

One last thing. Those who tackle these projects with a specialized partner are far more likely to succeed than those who try to do everything in-house. Not because the internal capability is missing, but because choosing the right process and connecting it well to the systems is a craft of its own.

---

The first step is not choosing a tool. It is mapping your processes and understanding, with the numbers in hand, where it is worth starting. It is exactly what we start from with every client: a consultation that identifies the first quick win and estimates its return, before writing a single line of automation.

Frequently asked questions

Which process is the best one to start with overall?

Usually a repetitive administrative task, high-volume and with clear rules. Payment reminders and managing incoming invoices are the classic first "yes".

Do I absolutely have to use artificial intelligence?

No. Many first processes are solved with classic automation, made of flows and rules. AI is needed where there is language to interpret or judgment to assist.

Can an Excel sheet be enough as a starting point?

Often yes. A tidy Excel connected to some automation is an excellent first step, much better than a complex project that never gets off the ground.

How much does it cost to automate a process?

It depends on the complexity and the tools, but a well-chosen first quick win usually has a contained cost and a return measurable in weeks. The serious way to find out is to start from a process mapping.

How long before you see results?

If you start from a quick win, weeks and not months. That is exactly the point of starting small: proving value fast.

← Back to the Magazine