AI and Jobs in Small and Mid-Sized Firms: What Studies Show So Far

Will AI replace jobs in small and mid-sized firms?

Few topics generate as many headlines as the question of whether artificial intelligence destroys jobs. For owners of small and mid-sized businesses the question matters twice over. They decide whether and where to use AI, and they are responsible for people who want to know whether their role will still be needed in five years.

Put simply, the question is: what do the studies so far actually show about the effect of AI on jobs, and what follows for the workforce planning of a mid-sized business? To answer it, we first clarify a model from labor economics, then the situation in Germany, then two readings of the findings, and finally a tool for your own business. Where AI saves time in sales today is described in the comparison AI in sales.

AI rarely replaces jobs, but it often replaces tasks. If you confuse the two, you plan either too fearfully or too carelessly.

Jobs are made of tasks: the model from labor economics

Labor economics has long examined the effect of technology not on whole occupations but on individual tasks. The foundation is the task-based approach of Autor, Levy, and Murnane (2003). In this view, every job is a bundle of tasks. Technology takes over some of them, complements others, and creates new ones. The consequences for a job depend on how this bundle changes.

In the first wave of computerization it was mainly routine tasks, activities that can be captured in clear rules, such as entering data, checking receipts, or simple calculations. Generative AI reaches further. It can draft text, answer questions, and summarize information, tasks previously considered non-routine. That is why it affects office and knowledge work more than earlier technology did.

An analogy from the trades makes this tangible. The cordless screwdriver did not abolish the carpenter's trade. It made one task, driving screws by hand, almost disappear and shifted the time gained to other work. Whether fewer carpenters were needed as a result depended on whether demand for furniture grew alongside.

How widespread is AI in German companies?

Usage figures diverge widely depending on who asks and how AI is defined. The ifo Institute asks in its business survey whether companies use AI in their business processes. In May 2026, 54.5 percent said yes, after 40.9 percent a year earlier. The Federal Statistical Office, using a narrower definition, arrived at 26 percent of companies with ten or more employees for 2025. Why the surveys differ so much is explained in Digitalization in the Mittelstand in numbers.

The ifo Institute published results on employment expectations in June 2025. 27.1 percent of companies expected AI to lead to job cuts over the next five years, 37.3 percent in manufacturing. Only 5.2 percent expected additional jobs, two thirds no change. Those expecting cuts anticipated about eight percent fewer jobs on average. These are expectations, not measured effects. According to the institute, companies are still exploring in which areas AI brings productivity gains.

What a large field study shows about AI at work

The most thorough study so far of the effect of generative AI in real workplaces comes from Brynjolfsson, Li, and Raymond (2025). They observed how a Fortune 500 software company gradually introduced an AI assistant in its chat-based customer service. The data cover 5,172 support agents, most of them in offices in the Philippines. Because the rollout was staggered, the authors could compare how the same employees performed before and after gaining access.

• Productivity: With AI support, employees resolved on average 15 percent more issues per hour.
• Unevenly distributed: Less experienced and less skilled agents became faster and better. The most experienced gained little speed and lost slightly in quality.
• Learning: The AI in effect passed the working methods of the best agents on to newer ones. New employees stayed with the company more often.
• Work climate: Customers wrote more politely and asked less often to speak to a manager.

Equally important are the limits the authors themselves name. The results describe the medium term in a single company. In their words, the study is not designed to show economy-wide effects on employment or wages, and the data do not reveal whether the company then hired fewer or different people. In the longer run, the authors note, companies could respond to the higher productivity of new agents by hiring more of them or by building systems that replace labor altogether.

Job killer or relief? Two readings of the findings

Two conclusions can be drawn from the same data. Both find support in research.

The complement reading is supported mainly by looking back. Gregory, Salomons, and Zierahn (2022) studied the effect of routine-replacing technology in 238 regions across 27 European countries between 1999 and 2010. The technology displaced a great deal of labor from routine tasks. At the same time, prices of the goods produced fell, demand rose, and sectors that do not compete across regions benefited too. On balance, employment grew. The authors stress, however, that this result is not inevitable. If demand does not rise enough, the net effect can also be negative.

I consider the complement reading the more accurate one for the Mittelstand at present, but not an all-clear. It holds for the total, not for every single job. The field study in particular shows that AI changes the most where newcomers used to be trained. If you currently give simple tasks to career starters so they can learn, you have to think about how they will learn in the future. And the European data come from a time before generative AI. Whether the pattern repeats cannot be derived from them.

The task inventory: reviewing roles in your own business soberly

For a small business, the economy-wide question matters less than its own. The task-based approach leads to a tool I call the task inventory. It breaks each role down into its activities and sorts them into three groups.

1. Record the activities

   Have each person write down a typical week: which activities, how many hours. Eight to twelve activities per role are enough.

2. Form three groups

   Assign each activity: AI can largely take it over, AI can support it, or it stays with people because it requires judgment, responsibility, relationships, or manual work.

3. Plan the time gained

   Decide what happens with freed-up hours: more customer contact, neglected tasks, training, or a position that is not refilled.

4. Talk with the people affected

   Discuss the result openly with the team before tools are introduced. People who help build the inventory also see limits you cannot see from outside.

5. Measure after three months

   Check whether the expected time savings materialized and what they were actually used for. Only then decide on further steps.

In this worked example no role disappears, but a quarter of the working time is freed up. Whether that turns into more revenue, better service, or a position not refilled is a business decision, not a technical one. For tools, the principle from Process before tool applies: only once the process is clear is it worth asking which software supports it.

What follows for workforce planning and leadership?

For the Mittelstand there is one more factor often missing from the general debate. Many businesses cannot find applicants for open positions anyway. For them, AI is less a question of cutting jobs than a way to achieve more with the existing team. How hard the skills shortage hits sales is shown in Skills shortage in sales.

• Say openly what AI is used for and what it is not used for, before rumors arise.
• Explicitly assign the time gained to new tasks instead of leaving it undefined.
• Keep giving newcomers tasks they learn from, even if AI could do them faster.
• Set rules for handling data and for checking AI output, as described in the article on the EU AI Act.
• Treat training in working with AI as part of the job, not as a spare-time topic.

Transparency here is more than a question of style. If you introduce AI quietly and then don't refill positions, you lose the team's trust, which you need for every further change. What obligations the EU AI regulation imposes on use in a business is described in EU AI Act. How to keep good people in times of change is covered in Keeping good people.

The next step

• Choose two roles with a lot of text, email, or data entry.
• Write down a typical week by activity and hours together with the people in those roles.
• Assign each activity to one of the three groups and roughly estimate the possible saving.
• Decide in writing what the time gained should be used for.
• Set a date in three months to review savings and their use.

Conclusion: what studies show so far about AI and jobs

The guiding question was what the studies actually show and what follows for a mid-sized business. It is established that AI speeds up individual tasks considerably, by 15 percent on average in the largest field study, and that less experienced employees benefit most. It is also established that earlier waves of technology in Europe shifted tasks substantially but created jobs on balance. It is not established how generative AI affects the total number of jobs. Companies' expectations are muted, and a majority expects no change.

For the individual business, the answer is therefore less a forecast than a decision. AI changes tasks, and what happens with the time gained is up to the owner. If you know the tasks in your roles and talk about them openly, you turn a vague worry into a change you can plan.