BYTEC Guides · Article 7 of 8

AI for SMEs: where it delivers measurable value today, and where it does not yet

AI for SMEs is no longer an experiment: almost everyone who uses it feels a benefit. Far fewer can put a number on it. This article separates what studies have measured from what respondents perceive, and shows where AI reliably helps today.

AS OF OCTOBER 2026 · 5 MIN. READ

Perceived and measured value

80% of respondents say AI has improved their personal productivity. But only 37% see a contribution to operating profit (EBIT), and only about 6% a significant one of at least 5% (McKinsey, August 2026). In Germany, AI users report effects more often: 48% have significantly accelerated internal processes, 45% see a measurable contribution to business success (Bitkom, September 2026). Both are self-reports.

Studies that measure before and after or compare groups are more robust. They show a clear pattern: large effects in narrowly defined tasks, small effects on average across all working time.

AI in customer service: the best-documented field

The best-known field study comes from customer service: 5,179 support agents were gradually given an AI assistant. With it they resolved on average 14% more issues per hour, and new and less experienced agents as much as 34% more. For the most experienced, the effect was small (Brynjolfsson, Li and Raymond, published in the Quarterly Journal of Economics in 2025).

In Germany, customer contact is the most common area of use: 72% of AI users use AI to handle enquiries (Bitkom 2026). Swiss SMEs are more cautious. 17% use AI in customer care and support, no more than in 2024; AXA suspects reputational risk and concern about wrong answers in direct contact as the reasons (AXA, September 2026). What an AI phone assistant can do and where its limits are is explained in What an AI phone assistant can do.

Time savings from AI in texts, translations and correspondence

In the everyday work of Swiss SMEs, language dominates: 47% use AI for translations, 42% for correspondence, 35% for advertising copy. The fastest growth is in optimising work steps, from 23% in 2024 to 35% (AXA 2026).

How much time this saves is well researched. In a representative US survey, users of generative AI saved on average 5.4% of their working hours, about 2.2 hours in a 40-hour week. Across all workers it was 1.4% (Federal Reserve Bank of St. Louis, February 2025, self-reported). In Denmark, the average time saving was about 3% (Humlum and Vestergaard 2025).

That sounds small, but it adds up across a whole team and a whole year. What matters is where the time goes. How to calculate this honestly is shown in Capacity is not profit.

Where companies already see money

Those who can attribute a financial effect to individual functions most often name supply chain, service operations and manufacturing for cost reductions. Revenue gains are most often attributed to marketing and sales, followed by product and software development (McKinsey 2026).

The successful 6% differ less in budget than in approach: nearly three quarters of them have fundamentally redesigned workflows because of AI, compared with one quarter of the others. They are twice as likely to have defined processes for measuring impact (McKinsey 2026). How to get there is described in Introducing AI in your company.

Generative AI in business: where it does not help yet

Measured results sometimes contradict perception. In a controlled trial with 16 experienced software developers, they took 19% longer with AI tools. Beforehand they had expected to save 24% of the time, and afterwards they believed they had been 20% faster (METR, July 2025). METR itself considers these figures out of date for today's tools. The lesson remains: perceived value does not replace measurement.

AI agents that work independently on a computer have also improved, but are not reliable. On the OSWorld test their success rate rose from about 12% to about 66%. That means they still fail at roughly one task in three (Stanford AI Index 2026). Agents therefore need clear permissions, human approvals and measurement before they are expanded. This is how we deploy AI agents with defined permissions.

How to measure the value in your SME

Choose a workflow that occurs often and is clearly defined, such as pre-sorting enquiries or preparing quotes.
Measure beforehand: number of cases, minutes per case, errors, waiting time.
Set the target figure before the AI starts.
Measure for about 30 days and compare with the baseline.

The steps in detail are in The first AI pilot. Whether your workflow is suitable can be checked with the checklist for your first AI pilot.

Frequently asked questions

How much time does AI save in office work?

Measured across all working time, not much: users in the US report about 5%, in Denmark it was about 3%. In individual tasks it is considerably more, in customer service for example 14% more issues resolved per hour.

Why do many companies see no effect on the bottom line?

Because time saved does not automatically turn into money. Only 37% of McKinsey respondents see a contribution to operating profit. Those who see an effect have redesigned their workflows and measure the result.

Where should an SME start?

Where many similar cases involving text or documents occur: customer enquiries, correspondence, quotes or checking documents. You will find examples by industry in our use cases.

Sources

All figures checked against the original sources on 23 October 2026.

Would you like to know where AI helps measurably in your company? In an initial conversation we identify with you the three workflows that cost the most time. Examples by industry: AI use cases.

Book an initial conversation