Claims about AI and productivity range from modest to extravagant. For an SME owner deciding whether to spend time and money on it, the useful question is narrower: what does the better evidence show, and how would you know whether it applies to your business?

What controlled experiments show

The strongest evidence comes from experiments in which some workers use generative AI and others do not. An OECD review of this research, summarised in an OECD blog in July 2025, found average productivity gains ranging from 5% to more than 25% for people in customer support, software development and consulting.

Three findings from that review matter for small firms:

  • Less-experienced staff tend to gain the most. AI gives them quick access to information and feedback, narrowing the gap with senior colleagues.
  • Experienced staff can benefit too, but only where the tool complements what they already know, and only if they trust it enough to use it.
  • Using AI on the wrong task makes things worse. When the task is beyond what the tool can do, it introduces errors and lowers quality. The review also notes signs of reduced independent thinking among heavy users.

In other words, gains depend on matching the tool to the task and on people being able to judge the output.

What workers report

Experiments are precise but narrow. Survey data show what happens at scale. The European Central Bank asks roughly 20,000 people across eleven euro area countries about their use of AI at work. The share of workers using AI on the job rose from 26% in 2024 to 41% in 2025 and 52% in 2026.

The median user reports saving about three hours a week, or 7.7% of working time. The ECB authors are careful about what that means. Only around half of workers reported using AI and saving time, so across the whole workforce the saving is closer to 3.8%. And saved hours only become productivity if they are turned into more output, which depends on the employer having useful work to fill them with.

The ECB data also show where savings are largest. Coding and debugging produce the biggest reported gains, close to eight hours a week, but only about 8% of workers use AI for that. Writing, editing and research are far more common uses, yet they produce smaller savings. A third of non-users said AI was simply not relevant to their tasks.

What SMEs themselves say

The OECD's survey of more than 5,000 SMEs found that 65% of those using generative AI said it had improved employee performance. Far fewer said it had helped them scale up (35%), compete with larger companies (29%) or raise revenue (26%). In Switzerland, the Sotomo survey for AXA reported by Swissinfo found 57% of AI-using SMEs reporting time savings.

These are self-reported figures, so they should be read as direction rather than measurement. But the pattern is consistent: most firms see people working faster, and fewer can yet point to a change in the business results.

Reading the evidence as an SME owner

Several practical conclusions follow.

First, the gains are real but moderate for most office work. Three hours a week per person is meaningful in a small team. It is not a reason to restructure the business.

Second, the gap between time saved and value created is where most firms lose the benefit. If a drafted reply saves twenty minutes but nobody decides what to do with those twenty minutes, the gain disappears into the day.

Third, the larger gains come from more structured, repetitive tasks rather than from general writing help. That is an argument for automating specific workflows rather than simply handing out chatbot licences.

How to measure it in your own firm

Before and after any AI change, record the same few numbers:

  • Time per unit of work. Minutes per quote, per enquiry, per report.
  • Volume handled. How many of those units the team completes per week.
  • Error or rework rate. How often output needs correcting.
  • Where the saved time went. Agree in advance what the freed hours are for.
  • Who is using it. Note whether gains come from junior or senior staff, since the evidence suggests they differ.

Run the comparison for at least four weeks. A short trial with one enthusiastic user tells you little about the team.

The bottom line

The best available evidence supports a measured view. AI tools save real time on suitable tasks, most of all for less-experienced staff and for repetitive, well-defined work. They can lower quality when misapplied, and the time saved only matters if the business puts it to use. For a Swiss SME, the sensible approach is to choose tasks carefully, measure honestly and decide in advance what the recovered hours are for.

Want a clear baseline before you start? The free AI Readiness Audit maps your recurring tasks and estimates where time savings are realistic.

Sources

  • OECD, "Unlocking productivity with generative AI: Evidence from experimental studies", 8 July 2025. link
  • European Central Bank, "AI adoption and the productivity promise: what workers report", 26 August 2026. link
  • OECD, "Generative AI and the SME Workforce", 5 November 2025. link
  • Swissinfo, "Swiss SMEs ramp up adoption of AI in workplaces", 8 October 2025. link