Made for Swiss self-employed & SMEsBookkeeping is rarely the reason anyone starts their own business. And yet it is the task that follows you home: the pile of receipts, the unpaid invoice you keep meaning to chase, the accounting software that suggests bookings you never look at. AI will not make bookkeeping disappear. But it can take over the mechanical parts — sorting receipts, proposing bookings, drafting reminders — so that you and your fiduciary spend your time on the numbers that actually matter.
This article is about three practical jobs where AI genuinely helps today: capturing and classifying receipts, suggesting bookings, and writing reminder emails. Each one is small, each one stays under human control, and each one works with the tools you already use, like Bexio or Abacus. No hype, no promises of a fully automated year-end. Just less manual work.
What AI can do in bookkeeping — and what it cannot
At its core, AI in bookkeeping is pattern recognition. It looks at a receipt and reads the vendor, the date, the amount, and the VAT. It looks at your past bookings and learns which account you usually use for a restaurant bill or a train ticket. It reads your reminder history and learns how firm your tone usually is. None of this is magic. It is the same kind of text and number processing AI has been doing for years, applied to a domain with a lot of repetition.
What AI cannot do is decide. It cannot judge whether a mixed invoice is fifty percent business and fifty percent private. It cannot choose an accounting policy, interpret a new tax ruling, or certify that your books are in order. It cannot tell you whether a reminder should already threaten legal steps — that depends on your contract and on Swiss law, not on a language model. The machine proposes. You, your team, and your fiduciary dispose. Whoever keeps that line clear will get real value from AI without the surprises.
Receipts: from the shoebox to a searchable record
The most common bookkeeping pain is the receipt pile. It collects on the desk, in the bag, in the glove compartment, until someone spends an afternoon sorting it. The first AI step is simple: photograph each receipt with your phone, and let the tool extract the vendor, date, amount, and tax rate. This is not new technology — receipt apps have done OCR for years — but the current generation of AI is noticeably better at messy cases: faded thermal paper, folded corners, mixed currencies, handwritten notes on the edge.
The second step is classification. The AI suggests a category: office material, travel, client meal, IT equipment, continuing education. In tools like Bexio or Abacus, the suggestion lands directly in the receipt list, and you confirm or change it with one tap. After a few weeks, the suggestions become accurate, because most self-employed people buy the same kinds of things every month.
A practical rhythm that works: photograph receipts the day they arrive, or once a week at most, and check the batch in five minutes. A pile of twenty receipts becomes a ten-minute task instead of a lost afternoon. And because the digital copy is stored with the extracted fields, you can find any receipt by vendor, date, or amount in seconds — which is worth a lot at year-end, during a VAT audit, or when a client asks about a charge.
One legal note, kept calm: Swiss law requires you to keep business records for ten years, and the digital copy must faithfully reproduce the original. A clear photo of a receipt is usually enough; the point is that the record stays readable and findable. If you are unsure about your specific setup, ask your fiduciary — it is a two-minute conversation.
Booking suggestions: proposals, not decisions
Once receipts are classified, the next step is the booking itself. Here, AI looks at the extracted fields and your history, then proposes the account assignment: travel expenses, office supplies, client hospitality, and so on. In Bexio and Abacus, this arrives as a suggested booking that you can accept or adjust before it is posted.
This is where the human check matters most. For a normal, unambiguous receipt — a train ticket, a printer cartridge, a business lunch with a client name on it — the suggestion is usually right, and accepting it takes seconds. For anything unusual, trust your hesitation: a mixed invoice with a private and a business share, a large one-off purchase, or a payment that touches several projects at once. These are judgment calls. The AI does not know your situation, and it should not decide for you. Adjust the suggestion, or leave the receipt for the monthly review with your fiduciary.
There is a quieter benefit worth naming: the review itself becomes the learning signal. Every correction you make teaches the system how you book. After a few months, the suggestions fit your habits, not a generic template. That is the difference between a tool that guesses and a tool that has been trained by your own decisions.
Reminders: firm, friendly, and without the guilt
Chasing payments is emotionally draining, especially for self-employed people who know their clients personally. The invoice is three weeks overdue, and every day you postpone the conversation makes it harder. This is where AI is genuinely kind: it drafts the wording, and you keep the judgment.
A good reminder sequence has three steps, and AI can draft all three in your tone. The first reminder is a friendly nudge — a short note that the invoice may have slipped through. The second is firmer: a clear reference to the invoice, a polite request for payment, and an invitation to call if there is an issue. The third states the facts calmly: the amount, the deadline, and the consequences you are entitled to draw. You review each draft, correct the details, and send it from your own mail tool, so the correspondence stays in your record.
Two rules keep this safe. First, the AI writes the words, not the facts: you check the invoice number, the amount, the date, and the client's name before anything goes out. Second, never let a draft invent legal consequences. Reminder law in Switzerland follows the contract and the Code of Obligations — including interest on late payments — and your wording should not promise more than you are entitled to. When in doubt, a sentence from your fiduciary costs less than a wrong threat.
And once the wording is right, you can keep it: a small library of approved reminder templates, in your voice, that you reuse for every client. The AI helps you build it once; you use it for years.
A calm weekly rhythm
None of this requires a big project. A workable rhythm for a self-employed person looks like this: once a week, photograph the week's receipts and approve the classifications. Once a week, go through the booking suggestions in your accounting tool and accept or adjust them. Once a week, review the reminder drafts for invoices that are overdue. Once a month, send your fiduciary the summary and ask the questions that need a human answer.
That is roughly fifteen minutes per week, spread over the week. The ledger stays current, the reminders go out on time, and the year-end is no longer a rescue operation. The point of AI in bookkeeping is not to remove your fiduciary or to automate your entire finance department. It is to remove the repetitive work that makes bookkeeping feel like punishment, so the financial side of your business becomes something you look at calmly instead of avoiding.
Data protection: receipts contain personal data
Receipts are not neutral scraps of paper. They contain names, addresses, sometimes health-related purchases, sometimes private spending that appears on a business card. Once such data moves into an AI tool, the Swiss Data Protection Act applies — the same rules as for any other personal data processing.
In practice, that means three habits. Use business-grade tools with clear terms, and avoid pasting client or employee data into personal consumer accounts. Check whether the provider stores your inputs and whether they may be used for training; for bookkeeping data, a provider that does not train on your files is the safer default. And keep the archive access-controlled: the digital receipt store is part of your business records, not a shared folder.
None of this is complicated, and it does not block the workflow. It simply means choosing your tools deliberately — which, for financial data, you should be doing anyway.
What stays human
Swiss bookkeeping has a quality standard: the books must be orderly, complete, and verifiable. AI helps you reach that standard faster, but it does not take over the responsibility. The final check, the sign-off, the tax declaration, and every decision with legal consequences stay with you and your fiduciary.
The honest picture of the future is not “the fiduciary disappears.” It is “the fiduciary reviews instead of types.” That is a better use of professional time — for them and for you.
Start with one workflow
If you take one thing from this article, make it this: start with the receipts. Photograph them, classify them, and see how the extracted fields look. That single workflow already removes the most annoying part of bookkeeping, and it builds the data that makes booking suggestions useful later.
Once the receipt habit feels normal, add the booking suggestions. Then add the reminder drafts. Small steps, at a calm pace, with a human check at every stage. That is how practical AI works in a Swiss small business — not as a revolution, but as a quieter month.
