Reading time: 6 minutes · For owners of small and medium-sized businesses in the March region, on the Obersee and around Lake Zurich
In a single afternoon, I processed two years of receipts — several hundred slips, invoices and expense records, in several languages, with AI. That sounds like a story about speed. It isn’t. The most important insight of that afternoon was a different one — and it applies to every SMB currently thinking about AI in the back office.
The machine read practically every receipt without error: vendor, date, amount, currency, VAT. And yet it couldn’t do the actual work. It didn’t know what the receipts mean.
The machine read every receipt perfectly. It still had no idea what any of them meant to me.
What AI handles easily today
Document extraction is a largely solved problem. Modern AI models pull the fields — vendor, date, amount, currency, VAT — with accuracy in the high nineties, even from crooked phone photos, faded till receipts and mixed languages. At the same time, the cost per document has dropped from francs to centimes within a few years.
In plain terms: typing in the numbers is done as a job. It has become a commodity. If you still spend hours on manual data entry — or tie up a good employee with it — you are paying for something a machine does faster, cheaper and with fewer slips.
Where the real value lives: in your rules
Because a receipt is not a data record. A receipt is a decision: which company does it belong to? Which account? Is it business or private? Are there tax implications? No extraction, however accurate, answers those questions — they need context that only you and your business have.
One of my rules that afternoon went roughly like this: the addressee always wins — except when the date says otherwise, except when it is clearly personal. A receipt addressed to a former company but dated after the switch to self-employment needs a judgment call. Not for speed, but for defensibility with your fiduciary and the tax office. Rules like that are packed with experience — and they are exactly the part no software ships with.
The discipline of restraint
The second most important decision of that afternoon: don’t auto-categorize everything. Everything unambiguous was waved through. Everything unclear was flagged and checked by hand. That feels like less automation — but it is precisely the point where many SMB AI projects tip over.
A confidently wrong entry in the books does more damage than an honest “please review”. When the tax office asks questions, what counts is not how fast a receipt was booked, but whether the categorization can be cleanly justified. How to anchor this way of thinking in your business is what I described in the post on the AI architect instead of prompt engineers: you need someone who designs the rules — not someone who types faster.
An example from the region: a plumbing business in Galgenen
Picture a plumbing business in Galgenen with eight employees. Every week, wholesale material invoices, fuel receipts, parking slips and small purchases from the DIY store pile up — spread across three service vans and a dozen job sites. Today everything lands in a folder, and at the end of the month the owner spends two evenings on it before the paperwork goes to the fiduciary.
With an AI-supported routine, the team photographs every receipt right after the purchase. The AI reads the fields, and the owner’s rules handle the routing: material purchases to the right job site, fuel receipts to the vehicle account, and anything without a clear project reference or above a defined threshold is flagged for review. Two evenings become half an hour — and quality goes up, because no receipt disappears in a jacket pocket anymore.
How to start — four steps
- Count the volume: how many receipts, delivery notes and invoices come in per month? How much of that is pure reading and retyping?
- Write down your rules: by which criteria do you categorize today? If you can’t write your rules down, that is the real problem — not the software.
- AI for the volume, you for the judgment: extraction, data entry and reading documents go to the machine. Routing, exceptions and edge cases remain your decision layer.
- Define uncertainty: decide when the AI must report “unsure” — and have those cases flagged instead of forcing a categorization.
For owner-led businesses with 3 to 25 employees, this is the most realistic entry into AI: small, measurable, no project budget needed. How I work with businesses of this size is described under Small Business.
AI multiplies expert judgment — it doesn’t replace it
At the end of that afternoon stood no insight about technology, but one about experience: AI doesn’t make expertise worthless. It makes it more valuable, because it lets you apply it at a scale that would be impossible by hand. The advantage is knowing what the numbers mean — not processing them faster.
And the principle reaches far beyond bookkeeping: quotes, customer enquiries, marketing copy, reporting. If you want to know where the biggest levers sit in your own business, you’ll find the framework under Services — working through exactly these questions with owners is what I do.
Frequently asked questions
What tools does an SMB need for this?
Fewer than you might think. Many Swiss accounting packages already have receipt recognition built in, and modern AI assistants read unstructured documents reliably. What matters is not the tool, but that your routing rules are cleanly defined before you automate.
Does this replace my fiduciary or accountant?
No — it improves the collaboration. Your fiduciary receives complete, cleanly categorized, traceable records instead of a folder full of slips. That means fewer queries and often lower fees. Responsibility for the year-end close and taxes stays with the professional.
Is this worth it for a small business in the March region?
Especially there. Whether it’s a plumbing business in Galgenen, a joinery in Lachen or a garage in Siebnen: without a dedicated bookkeeping department, every hour won is felt directly. As a consultant in the region, I see the biggest effects in businesses where the owner still does everything personally.
Where is admin eating your time?
Take the self-check or book a free 30-minute intro call — together we’ll find the places where AI genuinely takes load off your back office.