Guide / Industrial AI
Digital sovereignty for industrial companies: What should you own?
A provider changes its terms. A model is withdrawn. An integration stops working, and the only person who understands it works for an external supplier. These are practical tests of how much control a company has over its AI.
For an industrial business, digital sovereignty means understanding those dependencies and retaining workable options. It does not require building every component or moving every workload into an on-site data centre. It requires knowing what your business can keep using when something changes.
The IW study on digital sovereignty for SMEs treats the issue as a management responsibility. An individual workflow provides a useful place to begin.
Start with the work that must continue
Consider a hypothetical quotation workflow. Customer requirements arrive in attachments. Product data and pricing rules live in an ERP. An AI application prepares a draft for an employee to review.
The model is one part of that application. The business also depends on document mappings, pricing rules, approvals and the knowledge needed to operate it. Replacing the model achieves little if all those elements remain inaccessible.
| Area | What your company should be able to retain | A practical test |
|---|---|---|
| Data | Usable originals, metadata and clear usage rights | Open a complete export outside the application |
| Business logic | Rules, mappings and approval steps | Explain and change a rule without provider assistance |
| Models | Configuration and an independent evaluation set | Test an alternative on the same task |
| Operations | Access, logs and recovery procedures | Recover a failed workflow using the runbook |
| Skills | Knowledge inside the responsible team | Have an internal operator explain the full process |
The test scope matters. Exporting a sample proves less than rebuilding the agreed workflow from it. Record that distinction rather than turning a partial check into a general assurance.
Buy infrastructure where it makes sense
Your competitive knowledge may sit in how you assess a request, handle an exception or decide that an engineer needs to review a case. Those rules and their evaluation material deserve deliberate control.
That does not make rebuilding identity services or databases a sensible investment. A purchased component can fit the strategy when its interfaces, rights and operating arrangements are suitable. An open-weight model can still require substantial maintenance and careful licence review.
Separate location from control
European data residency answers a location question. Procurement also needs to examine administrative access, support arrangements, further processing and the ability to leave the service.
For cross-border operations, document requirements by workload. A German engineering team and a Swiss business unit may need different contractual or legal analysis. A common architecture should not hide those differences.
Choose one dependency to test
Draw the current workflow and identify the failure that would cause the greatest disruption. Then test a bounded alternative in an isolated environment. It could be an export and rebuild, or a second model operating without production write access.
If the exercise reveals that migration would take weeks, you have useful information for the next investment decision. Keep the findings, limitations and owner of the next action together.
Welf combines this work with application engineering and integration. Explore our private AI approach or discuss a workflow whose continuity matters to your business.