Launching a governed GenAI platform in a conservative group
Context
A diversified industrial and pharma group wanted to explore Generative AI, but executives were rightly concerned about privacy, IP leakage and regulatory constraints. At the same time, employees were already using public tools without guardrails.
What we did
I led a cross-functional initiative (IT, security, legal, HR, communications) to design and deploy a private, governed GenAI assistant. We selected and integrated the underlying model stack, implemented a secure data and access architecture, and designed an evaluation and guardrail layer including policy, prompts and usage monitoring.
In parallel we created an adoption program: internal champions, training sessions, office hours, clear “do and don’t” guidelines and a backlog of high-value use cases agreed with business functions.
Impact
Within a few months the assistant reached 70% adoption in the target population, with heavy users self-reporting 3–4 hours saved per week. The platform generated a pipeline of new AI products (including a support assistant and knowledge search) and became the foundation for the group’s wider AI platform strategy.
Replacing naive analytics with causal thinking in retail
Context
A leading grocery retailer faced a worrying drop in sales after a store layout change and multiple marketing initiatives. Different stakeholders were drawing conflicting conclusions from simple before/after reports, undermining trust in analytics.
What we did
I introduced a causal-inference framework and worked with business, finance and category management to redesign how initiatives were evaluated. We created a library of synthetic control and A/B test designs, standardized KPIs and built self-service templates so teams could plan and read experiments consistently.
Impact
The approach reframed several “failed” initiatives as neutral or positive once confounders were accounted for and led to better allocation of a multi-hundred-million-euro promotional budget. It also increased confidence in the analytics team and provided a common language for decisions.
Creating a “Factory DNA” for pharma manufacturing
Context
A pharma manufacturing network wanted to systematically improve OEE on filling and packaging lines. Data existed, but it was fragmented and not connected to how operators actually discussed line behaviour.
What we did
Together with operations leaders and engineers, we developed the concept of a “Factory DNA”: a way to encode all production steps, combining historian data, MES events and human annotations. Data scientists and engineers then built analytics that surfaced recurring patterns and “motifs” driving the most downtime.
Impact
Pilots delivered a 4% OEE uplift on a critical line, with a clear financial case to scale the approach to other assets. Equally important, the shared DNA metaphor created alignment between operations and analytics teams and accelerated the culture shift toward data-informed decisions on the shop floor.
More stories
I’m happy to share further anonymized examples – including forecasting, predictive maintenance, customer analytics and AI-enabled support – during a conversation, depending on your industry and needs.