In the media, Frankwatching, November 3, 2026

On AI, small businesses are falling behind.

For Frankwatching I wrote about why AI gets off the ground at large companies and stalls at small and midsize businesses. I think that has less to do with knowledge than with the layer underneath, because there's often nothing for an AI application to connect to.

Frankwatching

Waarom AI bij grote bedrijven wél van de grond komt en het mkb achterblijft

Article on Frankwatching, a platform about online marketing and digital trends

read the article on Frankwatching

What the piece is about.

According to Statistics Netherlands (CBS), 66.2 percent of large companies used AI in 2025, compared with 13.8 percent of micro businesses, and that gap is growing. A company of ten to fifty people easily works with eight to fifteen separate systems, each with its own version of the truth. A large organization has an integration layer and people who maintain the integrations. At a small business, nobody owns that. So at the swimming pool in Barendrecht, no AI went in during the first months. The work was connecting systems.

I wrote this piece myself, and Frankwatching published it on November 3, 2026. The link to the original is above. What you read here is my own summary.

  1. For one week, write down which work consists of retyping, how often it happens and how many minutes it takes each time.
  2. For each line, weigh what's at stake if it goes wrong, and cross off anything that has no export or API on either side.
  3. Check whether both systems share the same key, such as a customer or order number. If they don't, there's a matching job first.
  4. Build only the integration with the highest frequency times minutes times risk, then measure again. Only then do the smart things come in.

Where things stand now.

The Inge de Bruijn Zwembad platform keeps running every day. These are the recounted figures as of September 30, 2026, counted from the platform itself.

12+systems connected, without replacing any of them
5 minthe platform checks itself every five minutes
338signals from building management, 463,400 measurements on an ordinary day
24/7the AI receptionist answers customer questions, live since May 2026

The article includes a detail about the backup that I've since phrased more precisely. The figures above are the current ones.

Where does AI get stuck for you?

Usually the snag is in the systems underneath, rather than in the AI itself. I connect what's already there, starting with the integration that saves the most retyping.

Walk me through one process