AI in finance: a practical guide for mid-sized companies

26 July 2026 · 8 min read · Manfred Schönberger

The share of Swiss SMEs deliberately integrating artificial intelligence into their working processes rose from 22 to 34 per cent within a single year. The question is therefore rarely whether AI reaches the finance function, but where to begin. This article sorts the available applications by maturity and sets out an order of work for Swiss conditions.

Where Swiss SMEs stand today

The SME Labour Market Study published by AXA Switzerland in October 2025 provides the most robust snapshot. The share of companies deliberately integrating AI into working processes rose from 22 to 34 per cent. A further 37 per cent are piloting the technology, up from 33 per cent. The share of companies with no use at all fell from 45 to 29 per cent.

The distribution of applications is revealing. Communication tasks lead: translation at 52 per cent, correspondence at 47 per cent. Optimising work steps reaches 34 per cent, up from 23 per cent, and data analysis 32 per cent, up from 22 per cent. Perception follows usage: 45 per cent rate AI positively, up from 35 per cent, and among users 60 per cent see an opportunity against 8 per cent who see a threat.

Kathrin Braunwarth, Head of Data, Technology and Innovation at AXA Switzerland, reads the development as many companies having left the piloting phase behind and now integrating AI deliberately into their working processes.

Applications by maturity

For the finance function the reviewed contributions produce the following classification. The maturity column indicates how dependable the use is in an SME's day-to-day operations today.

Document capture heads the list for good reason. In December 2025 the vendor bimetrics demonstrated an agent approach that interprets documents by their meaning rather than by rigid rules, citing the correct classification of hardware purchases and split postings for hospitality receipts as examples. A January 2026 lab session by the TAXPUNK channel compares DATEV's AI bookkeeping with an alternative approach and explicitly covers the pitfalls of introduction.

The order that works

The Swiss firm TreuVision AG captures a stance in a short contribution that I share: AI as a co-pilot, combined with human strengths. From that an order of work follows:

1. Start with the highest volume. The process with the most repetitions per year delivers the largest absolute gain. In most SMEs that is accounts payable capture.
2. Keep a human in the approval. The machine may propose, a human approves. That separation protects the books and preserves almost all of the time saved.
3. Measure before and after. Capture time per document, number of queries, days to close. Without baseline figures any impact remains an assertion.
4. Settle the data flow before rolling out. Which data may leave the company and which stays in house belongs in writing.

Three questions to settle first

• Where does the data sit? Payroll and personal data fall under the Swiss Federal Act on Data Protection. The question of processing location and the processor relationship belongs before the rollout, not after it.
• Who reviews the results? A coding proposal remains a proposal. As long as the hit rate is unknown, every case gets reviewed.
• What happens on an outage? A process that stops without the service needs a documented fallback.

The author's assessment