Successfully Moving AI Prototypes into Production Environments

What’s it about?

Transforming AI experiments into production-ready solutions presents companies with significant challenges. While numerous initiatives get stuck in the pilot phase, two major companies demonstrate different but successful approaches: Ernst & Young relies on risk-based governance and compliance structures, while telecoms provider Lumen has established a democratic AI culture that gives every employee access to intelligent tools.

Background & Context

At consulting firm Ernst & Young, the perspective of responsibility is at the center. Particularly in heavily regulated areas such as financial services and tax advisory, responsible AI frameworks are being implemented to identify and minimize potential risks early. Data quality plays a decisive role here: inadequate data quality can doom projects to failure even at the initial stage. The ongoing technological evolution from generative to agentic AI, and in future to physical AI and quantum computing, further amplifies these requirements.

Lumen follows a fundamentally different approach and promotes a corporate culture that integrates AI use into daily work from the outset. New employees get access to AI tools already during onboarding. An established governance model strengthens personal responsibility and problem-solving capabilities in teams. Practical training such as Copilot Studio enables employees to experiment directly with the tools and develop their own solutions. Concrete applications such as the Migration Buddy for automating sales-relevant processes or AI-powered customer service tools demonstrate the efficiency gains of this strategy.

Both approaches underline that context-related training is far more effective than traditional training formats. While EY qualifies employees in their specific work context, Lumen relies on learning by doing with immediate practical application. The technological acceleration also requires flexible structures that can keep pace with the speed of AI development.

What does this mean?

The choice between a compliance-oriented and a cultural approach depends heavily on industry, regulatory environment, and company structure
Data quality and availability must be ensured before prototype development to avoid later production hurdles
Employee training should be hands-on and conducted in the direct work context, rather than relying on abstract training formats
Governance models must guarantee both security and room for experimentation so as not to stifle innovation
The rapid technological evolution requires adaptable frameworks that can integrate future AI generations as well