AI in the business: where it saves time today, and where it does not

Since ChatGPT no topic has entered businesses as quickly as artificial intelligence. After two years of practice it can be said soberly what works.

What reliably saves time

What does not hold up yet

Anything that goes out without control. An assistant that quotes prices without knowing the current price list does damage. Likewise applications that pass personal data to public services. That is not only delicate but also hard to justify under Swiss data protection law.

The difference between toy and tool

Successful projects have three features: a single clearly defined use case, in-house checked content as the basis and a person who is responsible. Projects without these three points fall asleep after three weeks.

How we proceed

We start with the question of which task in the business costs the most time. From that comes a prototype in a few weeks, tested with real data in everyday use. If it holds up, we develop it into a platform, with user management, logs and operation in Switzerland or the EU. If it does not, it has cost little.

Four applications that have proved themselves in SMEs

First, the assistant on the website that answers from in-house content and pre-qualifies enquiries before they land in the inbox. Second, the preparation of product data: technical specifications become saleable descriptions in a consistent structure. Third, translations into the national languages that a person checks. Fourth, the analysis of free text, such as feedback from forms, sorted automatically.

What it costs

A clearly defined prototype is manageable in effort. Ongoing operation depends on usage: an assistant with a few hundred enquiries a month costs less in model fees than a lunch a day. It is not the technology that gets expensive, but unclear requirements.

Data protection in concrete terms

The revised Swiss Data Protection Act demands transparency and purpose limitation. For AI that means: inform users that they are talking to an assistant. Do not put particularly sensitive data into public models. Choose processing in Switzerland or the EU where personal data is involved. Keep logs and set deletion periods.

How to recognise a serious provider

They ask about the process first, not the model. They name limits. They show where the answers come from. And they build so that you can leave the system again.

More on this: AI platforms and software and apps.