AI in your team: three in four use it, many quietly. How a small business makes the time saved visible

Your team is probably using AI already: for emails, summaries, translations, first drafts. In Swiss surveys, about three in four employees say they use AI in their work. Many don't mention it at the company. That leaves management seeing little of what is already happening: who uses AI for what, how much time it saves, where mistakes creep in, and which data ends up in which programs.

The silence has an understandable reason. Anyone who shows that a program does part of their work puts their image at risk: in experiments, people who use AI are seen as less capable and less hardworking. A policy alone won't bring the benefit into the open. That happens when leaders show how they use AI themselves, say honestly what happens with the time saved, make mistakes safe to discuss and give the team a fixed place to share. The end of this article shows what that can look like in 30 minutes a month.

What is happening in Swiss companies right now

In French-speaking Switzerland, 72 percent of working people use AI in their jobs (survey by Qualinsight with a provider of transcription software, March 2026, 1'030 respondents). In the Swiss results of a global study by KPMG and the University of Melbourne, the figure is 77 percent, collected around the turn of 2024/2025. Among companies, three in four small and midsize businesses use AI, 35 percent deliberately in their processes and 39 percent on a trial basis (AXA study, March 2026, 336 businesses). A survey by EY puts the figure at 89 percent of respondents, though more than half of its 604 respondents work at companies with 1'000 or more employees.

The rules are lagging. In French-speaking Switzerland, 49 percent work at organizations without clear rules on AI, and 79 percent have never had any training on it. In German-speaking Switzerland, 48 percent of office workers report clear rules and 45 percent a training offer from their employer (survey by the market research firm Marketagent for HP Switzerland, December 2025, 1'367 respondents, as reported by the trade journal Netzwoche). Much of it bypasses the company: 72 percent of AI users in French-speaking Switzerland use at least one free version, often through a personal account, and 27 percent use AI tools without their employer's approval. This is known as shadow AI.

Why many keep their AI use to themselves

Researchers at Duke University in the US ran four online experiments with 4'439 participants to see how people judge those who use AI (published in the journal PNAS in May 2025). People who imagined completing a task with AI expected to be seen as lazier, less competent, less diligent and more replaceable, and said they would be less likely to tell their manager. That expectation is justified: observers did rate people who got help from AI as lazier, less competent and less diligent than people who got other kinds of help.

Psychologists know this pattern. When someone delivers good work with help, others tend to put the success down to the help and the need for help down to the person: apparently they couldn't or wouldn't do it alone. Anyone who senses this keeps their tools out of sight. The term for it is impression management, the effort to protect the image others have of us. Using AI quietly is a sensible response to what people expect from those around them.

The Swiss figures fit. In 2024, half of Swiss office workers said they felt uncomfortable telling their manager they used AI (Slack Workforce Index, as reported by the trade magazine Organisator). Worldwide, the respondents who felt that discomfort mostly cited a sense that using AI is cheating and the worry of being seen as less competent or lazy; company policy was the least common reason (21 percent). In KPMG's Swiss results, 69 percent say they have presented AI-generated content as their own work. And in June 2026, the Swiss public broadcaster RTS reported that many people in French-speaking Switzerland use AI in secret, for fear of being judged or of losing their job.

Then there is the job market. According to the Swiss HR Barometer 2026 by the University of Lucerne and the Lucerne University of Applied Sciences and Arts (1'597 employees surveyed from March to June 2026), job insecurity has risen in almost every sector since 2024 and is at the level of its previous record highs. Only 11 percent, however, fear losing their job because of new technology or AI. Against that backdrop, people naturally wonder what will happen to those who show how much faster they have become.

What invisible AI costs a business

As long as nobody talks about it, a business pays in three places.

Data. 70 percent of AI users in French-speaking Switzerland have never checked where their inputs go. Anyone who pastes customer, payroll or health data into a free personal account hands it to a provider the company has neither chosen nor vetted.

Quality. In KPMG's Swiss results, 74 percent say they have relied on AI output at work without checking it, and 63 percent have made mistakes in their work because of AI. In the US, 40 percent of office workers received AI-generated work in the past month that looked good but had little substance, and it took them about two hours per case on average to fix it (survey by the coaching company BetterUp with the Stanford Social Media Lab, 2025). A team catches errors more easily when it knows where AI was involved.

Time. Management only knows where the time saved goes if people talk about it. A worked example shows how much time is at stake.

Worked example: where does the time saved go?

Take a business with 20 employees, 15 of whom use AI, in line with the surveys. All figures on time saved come from self-reports, so there are two assumptions. The low one comes from a Danish study of around 25'000 employees: users there report saving 2.8 percent of their working time on average, or about 1.2 hours in a 42-hour week. The high one is 3 hours a week; in French-speaking Switzerland, 82 percent report saving less than that.

Time saved	low	high
per person per week	about 1.2 hours	3 hours
for 15 people per week	about 18 hours	45 hours
in full-time positions of 42 hours	about 0.4	about 1.1
per year, at 46 working weeks	about 810 hours	2'070 hours
salary cost of those hours	about CHF 38'000	about CHF 96'000

Assumptions for the salary cost: a monthly salary of CHF 6'500 paid 13 times a year (a 13th monthly salary is common in Switzerland), plus 6.4 percent employer contributions to AHV, IV, EO and ALV, the Swiss old-age and survivors', disability, income compensation and unemployment insurance schemes, for a total of CHF 89'908 a year. Divided by 1'932 working hours (42 hours times 46 weeks), that comes to about CHF 46.50 per hour, before pension fund, accident insurance and family allowances. It is the same monthly salary as in our article on sick leave.

The calculation shows an order of magnitude, not a measurement, and part of the time goes into checking and fixing. In the Danish study, employees often put the time saved into other tasks; where employers encouraged AI use, into new tasks rather than leisure. So the time is already being freed up. The open question is whether anyone in the business decides what it is used for. AXA put it this way when presenting its study of small businesses (our translation): "The crucial question is how companies bundle the many small efficiency gains of each individual and turn them into real added value. That will be one of the most demanding leadership tasks of the coming years."

What a policy does, and what it doesn't

Any business that allows or tolerates AI needs a written policy. One page is enough to start: which tools are allowed, which data must never go in, who answers questions, and that every result is checked before it leaves the business. It protects data and quality. It does little to bring AI use into the open, and three findings explain why.

First, the silence is mostly about people's image; in the Slack survey, company policy was the least common reason. A policy does not ease that worry.

Second, the signal from leadership matters. At the end of 2024, the economists Anders Humlum and Emilie Vestergaard surveyed around 25'000 employees in Denmark across eleven occupations whose tasks are especially suited to AI chatbots, from accounting to customer service (working paper of the US research network NBER, 2025). Where employers explicitly encouraged AI use, the share who had used it for work was 35 percentage points higher and the share of daily users 11 points higher, and employees reported more time saved and better quality. Training and company-provided tools alone did little to raise daily use; without encouragement, the authors suspect, they mainly serve to prevent misuse. This is a correlation from a survey, not proof of cause.

Third, the Swiss government says so itself. At the end of October 2026, the Federal Office of Communications (OFCOM) published guidance for voluntary AI rules in companies and associations. It states (our translation): "For the AI code to have a real effect, it has to become part of the organization's culture." As building blocks it names communication, training and formats for exchange.

The legal position (as of November 2026)

• Instructions: Employers may issue general rules and specific instructions on how work is done and how people behave at work, including on AI. Employees must follow them in good faith (Art. 321d of the Swiss Code of Obligations).
• Data protection: According to the Federal Data Protection and Information Commissioner, the Swiss Federal Act on Data Protection applies directly to AI. Anyone who enters personal data of customers or employees into an AI tool is processing that data within the meaning of the Act.
• Monitoring: Systems intended to monitor employees' behavior in the workplace are prohibited (Art. 26(1) of Ordinance 3 to the Swiss Employment Act). Control systems needed for other reasons, such as security, must not impair employees' health or freedom of movement. If you want to know who uses AI for what, the simplest way is to ask.
• EU: Anyone covered by the EU AI Act who provides or deploys AI systems must take measures to support the AI literacy of their staff (Art. 4); since the amendment of July 2026, the regulation no longer prescribes a specific level. Our article on AI in hiring explains when a Swiss company is covered.
• Switzerland: There is no Swiss AI act yet. The consultation on a federal act implementing the Council of Europe's AI convention is planned for spring 2027. In this consultation, known as Vernehmlassung, cantons, parties and associations comment on the draft before it goes to parliament.

Four signals a team watches for

1. Leaders use AI themselves and say so. In the third experiment of the Duke study, 1'668 participants acting as managers decided on job candidates. Those who rarely used AI themselves preferred candidates who don't use it; those who used AI often preferred candidates who use it daily. So the judgment depends on one's own use. That suggests leaders set the tone: if they show openly what they use AI for and what they check, it becomes easier for everyone to show theirs.

2. Leaders say what happens with the time saved, and only what is true. According to AXA, just under half of small businesses with AI experience expect time savings with no effect on headcount, while 12 percent expect to need fewer staff in future. Your team doesn't know which group your business belongs to. Tell them as concretely as you can, and if you don't know yet, tell them when you will decide. The HR Barometer points to a related pattern: fear of automation is lower where employees are more involved, managers are supportive and the employer helps people build skills.

3. Mistakes can be put on the table. AI makes mistakes, and people miss them. A team learns from them when it can show mistakes without embarrassment. The organizational researcher Amy Edmondson studied this in 51 teams at a manufacturing company, in a paper published in 1999. Where members trusted that nobody in the team would be embarrassed, rejected or punished for speaking up, the teams showed more learning behavior: they sought feedback, asked for help, talked about errors and tried new things. Edmondson called that trust psychological safety. For AI, this means: whoever reports an AI mistake gets help, and the mistake goes on a shared list.

4. Sharing is recognized, and targets don't quietly go up. Anyone who shows how a task can be done in half the time, and then gets twice the workload without a conversation, learns that showing doesn't pay. In KPMG's Swiss results, at least 31 percent say AI has increased their workload, stress and pressure. So the time saved belongs on the table: how much goes into tasks that have piled up, into quality, into training, into relief?

A suggestion: the monthly 30-minute AI round

The routine below is a suggestion and has not been tested as a method. It turns the findings above into a fixed place in the working month and fits into an existing team meeting of up to about ten people. Taking part is voluntary, there is no ranking, and no customer or personal data appears in the examples.

1. 5 minutes: the leader goes first. One example of their own: what they used AI for, roughly how much time it saved, what they had to check or correct.
2. 15 minutes: one example per person, for anyone who wants to. What did I use AI for? Roughly how many minutes did it save? What went wrong, or what did I have to fix?
3. 5 minutes: agree on what to keep. One use that everyone adopts. One thing everyone stops doing or regulates more closely, such as a data rule.
4. 5 minutes: where the time goes. The team makes suggestions, the leader decides and says so: tasks that have piled up, quality, customer contact, training or relief.

Before the first round, the leader explains what it is about. Four sentences help, as long as they are true. Only say what you can stand by; if you can't honestly say a sentence, leave it out and say what you do know instead.

• "I use AI myself, for example for …, and I check …"
• "I'd like to know what you use AI for, so we can all learn from it."
• "Showing how to save time won't count against anyone. What we do with the time saved, we discuss here."
• "If you report a mistake involving AI, you'll get help and no blame."

After 90 days, three numbers count: how many people in the team have shown at least one example, how many uses are on the shared list, and how many mistakes and near misses have been reported. If the third number rises at first, that is more likely a good sign, because what was hidden is becoming visible. One anonymous question to everyone completes the picture: "Can I say here what I use AI for without fearing any downside?"

The bottom line

AI has arrived in most teams, often quietly, and the silence is a sensible response to a setting where needing help can look like weakness. A policy protects data and quality. Making the benefits visible is a leadership job: show your own use, talk honestly about the time saved, make mistakes safe to discuss and give sharing a fixed place. It takes 30 minutes a month.

If you'd rather not set this up alone: in our AI in the workplace team program, we work with a team of up to ten people for six weeks on its own tasks and agree how it handles AI output. For leadership teams there is the Leadership Lab, a workshop that covers leading with AI and psychological safety, among other topics. Our article AI in HR shows where AI helps in HR work itself, and AI-written job applications how a small business selects candidates when applications are written with AI. A first conversation takes 30 minutes: book an intro call.

Note: This article reflects the position in November 2026 and does not replace legal advice in an individual case. The surveys are based on self-reports; the experiments on how AI users are judged were run in the US, the study on encouragement in Denmark.

Sources (20)

Sources: AXA, SME labor market study 2026: artificial intelligence (Sep 15, 2026, in German) · University of Lucerne and Lucerne University of Applied Sciences and Arts, Swiss HR Barometer 2026, executive summary (in German) · Qualinsight, AI in the workplace in French-speaking Switzerland (May 2026, in French) · KPMG and University of Melbourne, Trust, attitudes and use of AI 2025, Switzerland · EY Switzerland, AI survey (May 27, 2026, in German) · Netzwoche on the HP Switzerland study (Jan 29, 2026, in German) · Reif, Larrick, Soll, Evidence of a social evaluation penalty for using AI, PNAS 2025 · Slack, Fall 2024 Workforce Index · Organisator on the Swiss figures (Dec 10, 2024, in German) · RTS (Jun 14, 2026, in French) · BetterUp Labs and Stanford Social Media Lab, Workslop · Humlum, Vestergaard, Large Language Models, Small Labor Market Effects, NBER 33777 · Edmondson, Psychological Safety and Learning Behavior in Work Teams, 1999 · Federal Office of Communications, Artificial intelligence (Oct 28, 2026, in German) · Federal Office of Communications, Guidance for an AI code (Oct 26, 2026, in German, PDF) · FDPIC, Data protection law applies directly to AI (May 8, 2025, in German) · SECO, Technical monitoring at work (in German) · SECO, guidance on Art. 26 of Ordinance 3 to the Employment Act (in German, PDF) · Swiss Code of Obligations, Art. 321d · European Commission, AI Literacy, Questions & Answers.