AI for SMEs: two processes that really run on their own in 2026, and what it took
If you read about AI for SMEs, you read a lot about possibilities and very little about processes that run every day. Here we show the opposite: a real case. At a tax advisory firm, two processes run on their own. They start by themselves and report back with a completion email. We describe what it took, what surprisingly wasn't needed, and what you can take from it for your own business. This isn't a market overview, and you won't find survey numbers here either.
What "AI for SMEs" often means, and what we mean by it
When owners talk about AI in their business, it's usually about one of two things. Either someone on the team uses a chat tool to write texts, draft emails or sort ideas. Or there was an attempt: a pilot project, a demo that impressed and was then left sitting. That's our view, not a statistic.
Both have their value. But they change little about the fact that the work still crosses your desk and your team's desks. Our home page puts it like this: maybe you already use ChatGPT. But nothing runs as a system yet.
That's why we distinguish between two kinds of AI employees. The first kind are thinking partners: you talk to them, they help you think, and you steer every step. The second kind are process employees: they take over a fixed workflow, start on their own and send a report at the end. Only the second kind climbs the autonomy levels with us.
When we talk about AI for SMEs in this article, we mean the second kind. For us, a process only counts as running when it starts without a nudge, runs through to the end and sends a report.
The case in brief: a tax advisory firm, two processes
At a tax advisory firm, two AI employees have taken over two processes that used to depend on the office team. Process no. 1 is filing the digital assessment notices: up to 30 to 40 a day, previously about 2 minutes per notice. Process no. 2 covers the notices that arrive on paper, which used to take the office team up to 2 hours, depending on volume. One AI employee files, and a second one checks the inbox every 30 minutes and hands the paper notices over to the first.
Both processes are at Autonomy Level 2, Autopilot, as of September 2026. That means they start on their own on a schedule or trigger, run through to the final report and hand anything uncertain to a person. How it went in detail, step by step, is described in What a German tax advisory firm lets run on its own today. Kevin also writes about the same firm on his blog, there from the perspective of building it: AI Agents for Tax Firms: A Real Case.
This article is about the question behind it: what did the firm have to do to make it run?
What it took
None of it is a secret. There are six things, and each of them takes time and attention, especially at the start.
Time for someone to show how it works today
Before anything was built, Kevin watched the office team at work.
For the business, that means the person who does the process today needs time to show it and to answer questions. In our assessment, many attempts fail right here. Nobody has written down how the work really runs, and the knowledge sits in the head of someone who has been doing it for years.
Patience with a long list
What looked like one task turned out to be around 45 rules at the firm: many types of assessment, different types of tax and types of letter, and mail from the city with different layouts.
For the owner, this is the moment when a process suddenly looks bigger than expected. For processes with many variants, that's normal. The list of variants is the core of the process. Every variant that is written down beforehand is one nobody has to come and ask you about later.
Someone who sits next to it during training
Training happened one type of assessment at a time, with someone sitting next to the AI employee, and every variant was demonstrated once.
That's the same onboarding a new colleague would get. The difference: at the end, every variant is written down as a rule. It doesn't have to be passed on by word of mouth to every new person who joins the team.
A sign-off with conditions
It was only signed off after four clean runs. After that, it was monitored side by side for a while, and only then did the process run on its own.
For the owner, that means you decide beforehand how you'll recognize a clean run, and you stick to it. At this firm, it took 2 weeks from the first round of watching to sign-off. That's what happened in this one case, not a time promise for your business.
A person for what the AI employee doesn't know
At the firm, anything the filing employee can't match with certainty goes to the office team by email. That's followed by a feedback session in which the new variant is trained in.
For the business, that means you need a name these cases go to, and someone who covers for that person. Without this escalation path, we don't sign off a process. An AI employee that guesses when it's unsure still needs someone checking its work. One that asks can run on its own.
A report someone reads
After every run, filing reports back with a completion email.
For the business, that means deciding beforehand who gets the report and what goes in it. Without a report, you end up following up yourself, and then you're back to managing every step.
What it didn't take
Just as telling is what wasn't needed.
No new software for filing. The notices end up in the document management system the firm already had. The AI employee works with what's already there.
No large-scale project. It was about one process, then the second. There was no plan for the whole business and no switch of every workflow at once. With us, that approach has a name: one process at a time.
No AI that thinks along. The two AI employees don't advise clients and don't check notices for substance. They file and hand over. The professional work and the final check stay with people.
What this means for your business
You can't copy the case one to one. Your business has different workflows, different inputs and a different team. Whether it's AI in small business or in a larger SME, what carries over is the sequence.
- Draw up the map. Write down every recurring process in your business together with your team: who does what, how often, and how long it takes.
- Separate thinking from doing. Where someone has to think, the work stays with a person. What simply has to get done goes on the levels.
- Pick process no. 1. The one with the biggest lever on time, money or revenue. In our program, we rate every process by time, money, risk and feasibility.
- Pin the process down. By five points: trigger, start, work, end, report. Plus the escalation path and the hard limits.
- Train and sign off. Only after clean runs and a period in which the process is monitored side by side.
You don't need to know how to program for this. It's built together in the 1:1, and one person on your team looks after the AI employees with you.
What these steps look like in detail, and which ways there are to automate in the first place, is covered in Process automation in a business: what works without developers and where a person stays involved. What the four levels mean, which we use to measure how far a process runs on its own, is explained in The four autonomy levels: how a process in your business runs on its own.
One piece of advice for any business starting out with AI: check who's guiding you. A provider should be able to tell you which process comes first, how they measure that it's running, and what they already have running themselves. We've written down ten such questions in Hiring an AI consultant for a small business: ten questions to ask every provider, including us, along with our own answers.
What this looks like for us
The tax advisory firm is our client case. The proof of the method is Kevin's own business. Kevin Welter runs his business with 14 AI employees: content, SEO, video editing, bookkeeping prep, leads and email. Each one has a personnel file, a fixed role and clear limits.
One example from it: the website kevinwelter.com is looked after by an AI SEO employee. In 28 days (Sep 2 to Sep 29, 2026), it got 1,942 clicks from Google Search. In the 28 days before that, it was 88.
For the topic of this article, one rule matters most: Kevin tries new levels in his own business first, before they come to a client's business. That's how it is with Level 3 right now. In Kevin's business, an AI manager reviews the work of the other AI employees, and that role is still on trial.
And we also say what isn't running yet. The two processes at the tax advisory firm are at Level 2. Level 4, Autonomous, the test of a week without the owner, hasn't been passed by any of our processes yet.
Runs without me
Every two weeks, one process that runs without the owner: what came in, what the AI did, where a person decided. Two minutes to read.
Worksheet
AI Rulebook on One Page
What your AI employee may do, what it must never do, who keeps an eye on it and what is checked before release: eight fields to fill in and an acceptance checklist. Comes as a PDF by email once you confirm.
Strategy session
In a 20-minute strategy session, you tell us where things get stuck. We'll tell you honestly whether an AI employee is a good fit for you.
Book a strategy sessionFrequently asked questions
Does the example carry over to my industry?
The sequence does, the rules don't. A tax advisory firm has assessment notices, a trade business has orders and delivery notes, an agency has inquiries and approvals. Which processes come into question for you, the map will show. Workflows that come up often, follow rules and arrive digitally or can be made digital are a good fit.
How big does a business need to be for this?
Our program, Owner Autonomy, is meant for businesses with a team of roughly 3 to 15 employees, where the business is up and running and a lot still depends on the owner. It isn't meant for large corporations.
Who on the team needs to have time for this?
Two people above all: you as the owner, and one person on your team who looks after the AI employees with you. The two of you come to the 1:1 every week for 60 minutes and complete the agreed tasks before the next session. At the start, there's also the person who does the process today. They show how it works and sit next to the AI employee during training.
Why do so many AI attempts stall at the trial stage?
We don't cite a study on this, because we haven't checked one. In our view, what's usually missing is a building block from leadership, rarely one from technology. What's often missing then:
- a description of where the process starts and where it ends,
- an escalation path, so someone still has to check every run,
- a report, so the owner ends up following up again,
- a person on the team who is responsible.
Then the demo stays a demo. That's why, in the program, we work with you and one person on your team, every week, on the real process.
What if it isn't running after 90 days?
That's what our guarantee is for. If your process isn't running on Autopilot (Autonomy Level 2) after 90 days, Kevin will keep working with you at no extra cost, as long as you come to the 1:1 every week, complete the agreed actions and name one person on your team.
Your next step
Two processes at a tax advisory firm don't prove everything. But they show what a business has to bring: time to show the work, a list of variants, someone sitting next to the AI employee, a sign-off with conditions, an escalation path and a report. If you want to know which workflow in your business meets these conditions, we'll look at it together on a call, along with the question of whether Owner Autonomy is a fit for you. That's also where we go over how it works and what it costs.