Executive Summary
A multinational logistics group, headquartered in the EU with operations across several Member States, had AI initiatives running across the business: in operations and planning, in document processing and customer service, and in internal productivity tools. An inventory of AI use found 25 initiatives, 20 of which had never been formally approved. There was no common way to decide which of them to fund, how to build them safely, or who owned them once they were live.
The group engaged Epitechnic to design an AI factory and lead its establishment. Over 12 months, Epitechnic worked with the group's teams to set up the four parts of the model: a secure shared foundation, a delivery process, operations, and governance. The work started from a full inventory of the AI use already in place, so the design reflected what the business was actually doing.
The result is one governed route from idea to production. Every AI initiative is recorded with a business owner and a risk classification, every new use case passes through the same approval route, and leadership funds use cases against defined investment criteria. Five use cases have reached production through the AI factory, each building on the same foundation and controls.
Business Challenge
The group's AI initiatives had started in different parts of the business, each making its own choices about data, tools, suppliers and oversight. Where AI use grows this way, an organisation can lose sight of which data reaches which services, who is accountable for what an AI system produces, and what each initiative costs to run. The inventory showed the scale of the gap: 20 of the 25 initiatives had no formal approval.
Leadership had no consistent basis for comparing initiatives or deciding which to fund. Promising pilots had no repeatable route to production, so each team faced the same questions about security, data access and monitoring on its own. With the EU AI Act's obligations taking effect, the group also needed an early view of which uses could fall into regulated categories.
Success Criteria
- A complete inventory of AI use across the group, including use nobody had formally approved
- One approval route, with risk classification and human oversight proportionate to each use
- Investment criteria that let leadership fund use cases with a measurable benefit and stop the rest
- A secure shared foundation that every new use case builds on
- A predictable, repeatable route from idea to production
- A named business owner for every use case in production
Epitechnic Approach
Epitechnic began with the AI use already in place. Each initiative was recorded with its business owner, the data it used, the suppliers involved and how far it had progressed, which gave the group a factual starting point for every later decision.
Epitechnic then designed the AI factory around the group's priorities, covering the foundation, the delivery process, operations and governance, and led its establishment alongside the group's own teams. The design is independent of any one model or cloud provider. A first wave of priority use cases went through the new route as it was set up, which tested each stage in practice and gave later use cases a proven path to follow.
Solution
A full view of AI use
The group had no single record of where AI was used. Epitechnic built an inventory of all 25 initiatives across operations and planning, document processing and customer service, and internal productivity tools, with an owner, data use, supplier and stage recorded for each. Leadership can now see every AI initiative in the business and who is accountable for it.
One approval route
Each initiative had been approved, or not, in its own way. Epitechnic designed a single intake and approval route: every use case is classified by what it does, the data it touches and the consequences if it goes wrong, and that classification sets the human oversight and controls it needs before approval. The classification takes account of the EU AI Act's risk categories, so uses that could be regulated are identified at intake.
Funding the use cases that matter
Investment decisions had no common criteria. Each use case now needs a named business owner, a measurable benefit, the full cost of running it in production, and agreed grounds to continue or stop. Benefit is tracked after go-live, so leadership can see which use cases deliver and can stop those that do not.
A reusable route to production
Each team had been solving the same delivery problems separately. The AI factory gives every use case a shared foundation, covering identity and access, controlled access to data, logging and monitoring, and approved patterns for models and tools. A standard delivery process takes each use case through defined stages from idea to production, and operations covers monitoring, incident handling and change control once it is live. Five use cases have reached production on this route, each reusing the same components.
Outcomes
Visibility: 25 AI initiatives inventoried, 20 of which had no formal approval, each now recorded with a business owner, its data use and a risk classification.
Delivery: Five use cases taken to production through the AI factory during the 12-month engagement.
Governance: One approval route for every AI use case, with risk classification and human oversight requirements set by what each use does and the data it touches.
Investment: Funding decided against defined criteria, with benefit tracked after go-live and agreed grounds to stop.
Reuse: Every new use case builds on the same foundation, delivery process and controls, giving the group a predictable route from idea to production.
Why It Worked
The design started from the AI use the business already had, so the governance fitted the work people were doing and covered the 20 initiatives that had grown up outside any approval. Setting up the route with a first wave of real use cases tested every stage in practice, and gave each later team a working path to production with the security, data and monitoring questions already answered.
Key Takeaways
Challenge: 25 AI initiatives across a multinational logistics group, 20 of them without formal approval, with no common way to fund, build or own them.
Approach: Epitechnic inventoried existing AI use, designed an AI factory covering foundation, delivery, operations and governance, and led its establishment with the group's teams over 12 months.
Outcomes: Every initiative visible with an owner and risk classification, one approval route, investment criteria with benefit tracking, and five use cases in production on a reusable route.
Lessons: A predictable route to production depends on governance designed around the AI use already in the business, and on shared components that each new use case can reuse.