top of page

AI in everyday practice: deliberate experiments pay off twice.

Sep 17
4 min read

Experiment deliberately with AI and you learn where its limits lie — while finding exactly those tasks that cost your practice the most time.


In many practices, AI stands at one of two points. Either it was tried once and the matter petered out. Or people are waiting for someone to tell them what it is good for. Both roads end the same way: a year on, nobody in the organization knows any better than before where AI helps and where it does not.


The willingness has long been there. In the Swiss eHealth Barometer 2026, 84% of the healthcare professionals surveyed can imagine automating routine tasks with AI in future. What is missing is not the will. What is missing is a method for getting started.

The good news first: that method requires neither an IT project nor a budget, but four questions put to a single task. How to use them to experiment systematically, build your AI literacy and come across measurable time savings along the way is the subject of this article.


Where AI helps is written in no manual

Harvard Business School, together with Boston Consulting Group, had 758 consultants work with and without AI. On tasks within the capabilities of the system, those using AI completed 12.2% more tasks, worked 25.1% faster and delivered markedly better quality.


On one task outside those capabilities, the picture reversed. There, the AI users were 19 percentage points less likely to arrive at the correct solution than the group without AI. The researchers call this a “jagged technological frontier”: some tasks AI handles effortlessly, others of apparently equal difficulty not at all. And: from the outside, that frontier is invisible.


The limits of AI are written in no handbook. You find them only by probing them.


Every attempt is also a lesson

That is precisely why experimenting is not a pastime but the cheapest form of training. AI literacy — the abilities it takes to live, learn and work in an environment shaped by AI — does not arise in a course alone. It arises where someone tests a task of their own, checks the result and notices the difference. Not knowledge about technology, then, but judgment about its use.


This competence is an obligation besides. Since February 2, 2025, Article 4 of the EU AI Act has required businesses to ensure the AI literacy of their staff; the provision can capture Swiss practices as soon as there is an EU connection. Anyone who has to train their people anyway is best off doing it on a real task.


That leaves the question of where to look. The survey conducted by gfs.bern on behalf of the FMH, the Swiss Medical Association, points the direction: in 2025, doctors working in outpatient practice spent an average of 54 minutes a day on requirements from authorities and insurers alone, most of it on billing queries and on reports for loss-of-earnings insurers. A year earlier, 65% of those working in outpatient practice described this workload as rather or clearly superfluous.


Four questions before you start an experiment

First: demanding, and noticeable. The task costs time or concentration on a regular basis. Something that takes two minutes is no use as a first attempt, even if AI does it well — the gain would stay invisible.

Second: recurring, and regular. The task comes up daily or weekly. That pays twice over: a time saving adds up, and several attempts can be compared directly. Experience grows out of that comparison.

Third: describable, and teachable. The process can be set out in rules or a template — the reply to a recurring billing query, for instance. Anyone who cannot describe a task cannot instruct an AI to do it either. This step sharpens the view of your own process, whether the experiment succeeds or not.

Fourth: verifiable, and defensible. The result can be checked. The FMH recommends using such systems only where you are professionally able to recognize errors and judge the quality of the answer; for patient data it calls for locally hosted or controlled models. For a first attempt that means invented or safely anonymized examples in a freely accessible tool, and never real patient data.


Four times yes, and the task is your testing ground. You learn where the frontier runs. You measure whether the time actually falls. And: you decide from your own experience whether this becomes a fixed step in the work.


It is not the most spectacular tool that moves your practice forward. It is the most unremarkable recurring task, tested deliberately.


Ready for a first conversation? Arrange a no-obligation initial meeting: https://www.felber-advisory.ch/kontakt. Together we will identify the tasks in your practice that lend themselves to a first attempt, and set out how your team can experiment safely and measurably.



Sources

 
 
 

Comments

bottom of page