FOLYOO
Log inGet in touch
All articles
Automation4 min read

What businesses actually do with AI

One in five mid-sized German companies uses AI. KfW surveyed what for, and the answer is less spectacular and more useful than any trade fair suggests.

Christoph Beuter · Developer and founder of Folyoo

One in five mid-sized companies in Germany now uses at least one AI technology. That figure does not come from a vendor's brochure but from the KfW Mittelstandspanel, one of the largest company surveys in the country. What they use it for is more interesting than the number.

The simple things win#

KfW measured four kinds of AI separately. The result is unambiguous:

Generating language14 %
Reading documents10 %
Data analytics3 %
Controlling machines1 %
Share of mid-sized companies using this AI technology
Zahlen als Tabelle
BezeichnungWert
Generating language14 %
Reading documents10 %
Data analytics3 %
Controlling machines1 %

Fourteen percent have AI generate language, ten percent read documents. Data analytics sits at three percent, autonomously moving machines at one.

That is the opposite of what gets demonstrated at trade fairs. The technologies that stick are the ones you can switch on Monday morning and use on Tuesday. Anything that needs a project stays on the shelf.

"We are too small for that" does not hold#

The most common argument against AI in a small business is size. KfW looked at exactly that and reached a different conclusion: what decides adoption is less the size of the company than its level of digitalisation, the knowledge already in the building, and whether a business is used to changing things at all.

The figures for the smallest companies bear this out. Businesses with fewer than five employees trail the average by a single percentage point on both of the widespread technologies:

  • all mid-sized
  • under 5 employees
Generating language
14 %
13 %
Reading documents
10 %
9 %
Adoption by company size
Zahlen als Tabelle
Bezeichnungall mid-sizedunder 5 employees
Generating language14 %13 %
Reading documents10 %9 %

One point. A three-person business is level with companies ten times its size on the applications that actually get used.

The real obstacle is somewhere else#

KfW reports a correlation that looks unremarkable at first: companies using AI are strikingly likely to inform themselves through simple channels. Trade magazines, fairs, websites, advice. The correlation with universities, specialist providers or exchanges with customers is markedly weaker.

The study draws a conclusion I see confirmed in conversation after conversation:

Identifying a first, practically usable AI application is a central obstacle.

The problem is neither the technology nor the money. The problem is working out what to do with it in your own business. Whoever answers that once gets somewhere. Whoever waits for it to become obvious keeps waiting.

What this means for a business with five people#

The numbers suggest a practical order of operations. Generating language and reading documents are what works. Not because they are the most exciting applications, but because they fit into existing routines without anybody changing how they work.

Translated into the day of a small business, that is three things:

  • Something that answers the phone when nobody can. That is language generation in the narrow sense.
  • Something that turns an enquiry into a quote instead of somebody typing the same data twice. That is document reading plus automation.
  • Something that turns paper into data: delivery notes, invoices, handwritten notes from a site visit.

What is missing from that list says as much as what is on it: no forecasting models, no image recognition, no autonomous machinery. All of it exists; none of it pays for a five-person business, and KfW's figures show those businesses worked that out long ago.

Where I stand#

I build this sort of thing, so my opinion is not neutral. What I take from the figures anyway: a share of twenty percent also means four out of five businesses use none of it. That is not a backlog to catch up on, it is a question everyone gets to answer for themselves.

The right order is not to decide in favour of AI and then look for an application. It is to name the task that eats time every week, and then check whether something exists for it. In three out of four cases I come across, the answer is a simple tool rather than artificial intelligence. In the rest, it is worth it.

Sources#