← All articles

AI for Small Business: Which Processes to Automate First

Four tests for a good first AI process, five kinds of work that often pass them, and what to leave for later.

A printed process map on a wooden desk in a small manufacturing company, one step circled in blue marker, the workshop floor behind

Key takeaways

  • Destatis counts 26 % of German companies (10+ employees) using AI in 2025; Insee counts 18 % in France, against an EU average of 20 %.
  • Lack of know-how is the most common brake: 72 % of German non-users name it, and more than half of French AI users say a lack of expertise holds them back.
  • A good first AI process passes four tests: lots of repetition, rules you can write down, a person who catches mistakes before they cost money, and a result you can measure in hours or euros.
  • Five process types pass those tests most often in small and mid-sized firms: incoming requests, document capture, quote follow-ups, review replies and knowledge capture.

In 2025, 26 % of German companies with ten or more employees used AI, according to the Federal Statistical Office. In France, Insee puts the figure at 18 %. Among firms with 10 to 49 people the numbers drop to 23 % and 15 %. The owners we talk to have stopped asking whether AI is for them. They want to know which process to start with.

Where small and mid-sized firms stand

The size gap is large in both countries. More than half of companies with 250 or more employees use AI (57 % in Germany, 58 % in France). In firms with 10 to 49 people, it's a quarter or less.

In Germany, missing knowledge is the reason companies without AI give most often (72 %). In France, more than half of the firms that already use AI say they lack expertise. A large company sets up a team to test tools. A 40-person business has nobody for that, so the owner ends up choosing between offers they can't compare.

To get started, a good first choice of process is enough. The rest follows from it.

Four tests for your first AI process

Before we talk about technology, we go through these four questions with every client.

1. Does the work come often, in a similar form? AI pays off where the same kind of task comes up dozens or hundreds of times a month. Something that happens four times a year rarely does.

2. Can the rules be written down? If your best person can explain how they decide, a system can learn it. If the answer is "gut feeling", leave it for later.

3. Does someone see the mistake before it gets expensive? A good first process keeps a person at the end who approves or corrects. That keeps the risk small while the system learns.

4. Can you measure the result? Hours per week, turnaround time, missed calls, unanswered quotes. Without a number before, you won't know afterwards whether it worked.

A process that passes all four is a good candidate. To choose between several candidates, see Where to Start With AI: Find the One Bottleneck.

Five processes that often pay off first

This list comes from projects we have built or are building. Your bottleneck may sit somewhere else.

Sorting and answering incoming requests

Emails, contact forms, calls after closing time. In many businesses someone spends the first hour of the day reading, sorting and forwarding. A system can pre-sort, answer the routine questions and pass on only what needs a person. The rules usually already live in the heads of the office team.

Reading and capturing documents

Delivery notes, invoices, forms, damage photos. Volume is high and checking is easy. Our clearest example is TraxAgent, built for a leading truck damage specialist in Germany: drivers report damage around the clock in more than 26 languages, and the team handles up to three times the volume it did before. The full story is in Same Team, 3x the Volume.

Following up on quotes

A quote goes out, and then nothing happens because nobody has time to chase it. A system reminds at the right moment, drafts the follow-up email and puts it in front of the salesperson to approve. The person decides; the machine doesn't forget.

Replying to Google reviews

For trades and shops, every review counts. A system can answer positive reviews in your tone and send negative ones to you first. The effort is small, and it's a good way to get to know AI inside your own business.

Keeping knowledge in the company

In many small firms, a lot rests on two or three people. A two-minute voice note after each customer meeting, turned into a clean record by a system, is a simple start. Whoever takes over the customer finds the history instead of asking for it again.

These patterns are described on our AI automation page. When a process needs a system that prepares decisions and runs several steps on its own, we build custom AI agents.

What to leave for later

Some processes sound tempting but make poor first projects.

Rare, high-stakes decisions are one group: next year's pricing, choosing a new supplier, a hiring decision. There are too few examples for a system to learn from, and a mistake costs a lot.

Negotiation stays with people. AI can help prepare, for example by summarising a file. The meeting itself runs on experience and on reading the person across the table.

Be careful too with work where nobody notices today when something goes wrong. Without a check, there is no way to improve the system. There we build the measurement first and automate after.

And sometimes AI is simply the wrong tool. A plain rule in your existing software or a well-made form solves some problems faster. When that's the case, we say so.

How to decide

Take the candidates that pass all four tests and write down what each one costs today: hours per week times hourly cost, lost jobs, late payments. The most expensive one with clear rules wins.

Before we build anything, we map the process as it really runs, not as the manual describes it. That habit comes from analytical chemistry and manufacturing, where our founder learned to look for the slowest step of a reaction. How a project runs is on our method page.

On cost: a one-time setup fee and a monthly fee for hosting, maintenance and updates, both scoped to your process. We don't bill by the hour. The systems run on servers in the EU, at Hetzner in Germany.

The first process shapes the second

Starting with AI is a choice of one process. Pick one that comes often, has clear rules and can be measured, and your team learns on a real case what AI can do in your business. With that, the second choice gets much easier.

Frequently asked questions

One that happens often, has clear rules, ends with a person checking the result, and whose cost you can already put a number on. Often that's incoming requests or document capture.

No. You need someone who knows the process well and can spare a few hours for conversations. We handle the build, hosting and maintenance.

It depends on the scope. More important than the timeline is writing down a starting number, so you can compare after launch.

The systems we run are hosted on servers in the EU (Hetzner, Germany). For each project we put in writing which data goes where before we start.

Want to see what AI can do for you?

Tell us about your business. We'll get back to you within 24 hours.

Schedule a Strategy Call

More articles