Manufacturing company, family-run
We brought AI pilots into a defined approval process
A family-run manufacturer wanted to make its ongoing AI pilots visible and create a manageable path for review, approval and support.
| Industry | Manufacturing |
|---|---|
| Services | AI Governance |
Business teams tested AI outside shared rules
AI pilots for document search, text drafts and internal information were emerging across the company. Procurement and IT knew some of the tools; business teams knew others. Existing software lists could not reliably show which data people entered or who used the results.
Management wanted an overview. A general ban would have interrupted applications that were already useful. Blanket approval would have overlooked open questions about confidentiality, purpose and responsibility. Teams lacked a procedure for describing their specific use and receiving a traceable decision.
Approvals needed to assess the specific use
Our task was to build an AI inventory and a clear approval process. Reviews would follow the actual intended purpose. IT, data protection and legal owners needed to be able to document their decisions. Users needed clear rules for daily work, supported by practical AI literacy training in the context of Article 4 of the EU AI Act.
Every AI use received a purpose and named owners
The inventory captured usage
For each application, we recorded its purpose, data used, people affected, provider and business owner. Informal tests were included too.
Risk reviews included their reasoning
We prepared information for assessment under the EU AI Act. Responsible specialists reviewed risk classes and possible obligations against each use case; unresolved points remained documented as open.
Approvals had specific conditions
Permitted data, human review, access rights and usage boundaries were recorded. Changes in purpose or data access triggered another review.
Training related to work tasks
Teams practiced checking sources, handling confidential inputs and reporting incorrect results. Content and attendance were documented for internal tracking.
IT and business teams reviewed existing pilots together
We began with conversations in the business teams and asked them to show us real workflows. We deliberately kept the inventory process simple. Users needed to be able to report uncertain or previously unapproved experiments too. IT, data protection staff and business owners then examined the uses in recurring review sessions.
Decisions depended on the specific purpose. A tool name alone said little about whether confidential data was processed or consequential decisions were prepared. For unclear uses, owners first agreed on limited test conditions. Legal classification remained with the responsible specialists; our task was to prepare technical facts and process boundaries for review.
In training, teams worked with relevant examples from daily work. They had to recognize a plausible but unsupported answer and use the designated reporting route. Managers also practiced identifying when a planned extension needed resubmission. We simplified templates based on feedback while preserving the information needed for decisions.
New AI uses followed a familiar review process
Management received an overview of recorded applications and their open conditions. Business teams knew their contacts and could submit new uses in a structured way. Approved pilots had documented boundaries and support. When changes occurred, the team could trace which assessment needed revisiting. Training provided concrete exercises for handling errors later.
Approvals stay current only if changes are reported
Inventory the use case
The same tool can serve different tasks. Purpose and data access therefore belong in every entry.
Unresolved classifications need an owner
Record who is handling an open question and which uses are permitted until a decision is made.
Practice reporting in the context of work
Users must be able to route errors and changes of purpose to the right place for rules to work in practice.
A first conversation takes 30 minutes.
We discuss how your ongoing AI pilots can receive traceable approvals and clear ownership.