Novartis’s Playbook for Moving AI from Pilot to Production

In the AI hype cycle, the gap between a impressive demo and a reliable production system is often a chasm. Novartis is attempting to bridge this gap not with flashy new models, but with unglamorous, foundational data work. Christian Diehl, the company’s chief data and digital officer for biomedical research, recently outlined how years of platform investment are finally yielding tangible returns in biomedical research.

The core of Novartis’s strategy revolves around three specific application areas: AI-driven safety prediction, generative chemistry, and faster clinical translation. Diehl emphasizes that the value does not come from the AI models themselves, but from the quality and accessibility of the underlying data. By standardizing and centralizing disparate datasets, Novartis has created a substrate that allows AI agents and tools to operate with greater accuracy and consistency.

For the AI ecosystem, this case study offers a critical lesson: infrastructure is the moat. Many startups focus on building novel architectures, but Novartis’s approach suggests that the real competitive advantage lies in data hygiene. When data is clean, structured, and accessible, even standard AI models can outperform complex systems fed by dirty data. This is particularly evident in their safety prediction models, where reducing false positives is a matter of data quality, not just model complexity.

Furthermore, the focus on generative chemistry highlights a shift from exploratory AI to utilitarian AI. Instead of generating random molecular structures for the sake of novelty, Novartis uses AI to propose compounds that are synthetically feasible and align with specific therapeutic targets. This practical constraint ensures that the AI’s output is immediately actionable by chemists, rather than requiring significant manual post-processing.

For organizations looking to scale AI beyond the pilot phase, the takeaway is clear: do not skip the data engineering phase. Investing in robust data platforms may slow down initial prototyping, but it accelerates the journey to production. Novartis’s experience suggests that the most significant ROI in AI comes not from the algorithm, but from the disciplined integration of AI into existing workflows, supported by high-quality data foundations. This is a pragmatic, no-nonsense approach that prioritizes operational reliability over technological spectacle.

Photo: National Cancer Institute / Unsplash (https://unsplash.com/@nci)