AI in the middle market: a subscription for everyone is not a transformation.

A lot of AI projects start with a tool and then go looking for a workflow to put it in. I turn that around and ask three questions about every important workflow. Does it depend on a single person? Has nobody questioned it in years? Does it cost the most money? Where more than one applies, you have your first candidate. The list also needs what stays on hold for now, or attention spreads across too many experiments at once. At SUMAX, process knowledge used to be scattered across spreadsheets and individual heads. That knowledge belongs in one place, and that is where the change starts.

Access to a language model doesn't change who does what. We rewrote our role profiles so it's clear who prepares results, who checks them and who signs off. Whoever passes on an AI result has checked it and owns it, even if the AI wrote the draft. Training uses real tasks from each team, so people find the limits of the tools in their own work. AI is there to take load off people. If you plan AI around headcount cuts, you lose exactly the people who have to check the output.

When every department buys its own tool, you end up with separate contracts, data paths and sources of error. So we built the foundation for everyone first: sign-in, rights and access, plus our own layer between the applications and the AI models. That layer lets us switch a model without touching a single tool, and it books every call with its cost. Security is part of it from day one. Tools for individual tasks came after that, and we have replaced several purchased tools with our own. I led the technical foundation and built it with the team, so I know what you can ask of teams and vendors.

A pilot shows what is possible. After that, daily work comes down to data quality, exceptions and whether the team uses the application. So at launch it has to be clear who looks after the application in operation, logs its errors and decides on changes, and by which criteria it gets expanded, reworked or ended. Without someone looking after it, an application falls out of date with the next model change, and the team drifts back to the old way of working. The central cost log shows what each application costs to run. That figure belongs next to the benefit before anyone decides on the next stage.

In a group of companies, a holding company or a buy-and-build platform, one more question comes up. Does each company build its own AI setup, or is there a shared foundation? The order stays the same. Sign-in, rights and access are set up once for everyone, plus a layer that every AI request runs through with its cost and that lets you switch the model. On top of that, each company gets tools for its own workflows, because that is where the knowledge of the work sits. The group sees cost and usage in comparable form, and the companies don't all have to work the same way. An acquisition plugs into the same foundation instead of bringing one more island.