Merck KGaA's AI Chief: Models Aren't the Advantage—Context and Trust Are

Walid Mehanna joined Merck KGaA as group data officer after a five-year tenure as chief data officer at Mercedes-Benz. When he arrived, the company had already begun rebuilding its data and analytics systems for its 62,000-person workforce; about two years into that work, ChatGPT set off a global corporate scramble in late 2022. Eight months into the generative A.I. era, Mehanna's title was elevated to chief data and A.I. officer—a deliberate evolution, he said, "not a reaction to a development, hype or trend."

Merck KGaA, founded in the 17th century by pharmacist Friedrich Jacob Merck in Darmstadt, Germany, spans pharmaceuticals, medical equipment manufacturing, and electronics. Leading A.I. strategy inside a 350-year-old company is a unique challenge; Mehanna compares it to building a pyramid. At the base are everyday A.I. tools that improve personal productivity across the workforce, building digital fluency and saving time to reinvest in growth. The middle layer—where the company currently focuses—embeds A.I. into core workflows such as research and development, supply chains, and commercial operations over the coming years. At the pyramid's top sits product A.I., where machine learning becomes part of what the company sells, helping accelerate drug discovery and innovation.

In June 2023, Merck KGaA deployed MyGPT, an internal platform built in-house to give employees a safe space to experiment with generative A.I. But within a year, Mehanna saw the limits of relying on a single setup. The company partnered with LangDock, a Berlin-based startup that at the time of their initial conversations had four customers and roughly $50,000 in annual recurring revenue. The decision to work with a young European company was not sentimental, Mehanna emphasized. "The thinking behind the collaboration was about optionality, speed and sovereignty, not the sentimentality of working with a German startup," he said. LangDock provided the ability to build something fully GDPR-compliant that could be hosted in Merck KGaA's own environment, achieving enterprise-grade security while retaining startup agility. The arrangement also created a buffer between employees and major A.I. providers. "It was important for us to establish a flexible, model-agnostic user layer that gave us access to the world's best models, regardless of who made them, without creating vendor lock-in," Mehanna said.

Governance was built into the strategy from the start. Merck KGaA internally generated over 12 million prompts—a rich dataset containing insights into how employees work, what concerns them, and where they need A.I. support. The company analyzes these prompts only at an aggregate level, never monitoring individual employee activity. "We learn from patterns in prompts, never from individual employees or their usage," Mehanna said. "All usage is privacy-protected, and we do not monitor individuals." The company established a Digital Ethics Advisory Panel guided by five principles: autonomy, beneficence, non-maleficence, justice, and transparency. Rather than simply approving or rejecting projects, the panel pushes teams to identify risks early and design safeguards before launch.

Mehanna's thinking about A.I.'s competitive landscape has evolved. Two years ago, he believed the race for enterprise A.I. would be decided at the foundational model level—that the company with the best model would win. "I have definitely changed my mind about that," he admitted. "Models are still crucial, but they leapfrog one another every few months. They are therefore not a durable advantage." Instead, he now sees durable advantage in context, data, processes, workforce fluency, and the trust an organization builds with employees and customers. That shift changed how Merck KGaA builds its systems: a top-down semantic data layer proved too slow, so teams now integrate data incrementally—process by process and use case by use case.

On A.I.'s impact on jobs, Mehanna rejects the notion of simple elimination. "We do not see roles disappearing. We see roles evolving," he said. "A.I. becomes a dynamic tool for accelerating and extending work. The leader of the future will also have to lead a hybrid workforce composed of people and agents." Drawing on unexpected sources—he references Arnold Schwarzenegger's management guide, Be Useful: Seven Tools for Life—Mehanna emphasizes the constancy of foundational principles across diverse careers. "The underlying foundational principle is to always be useful, and that deeply resonated with me."

Merck KGaA's approach to A.I. reflects a maturing enterprise perspective that moves beyond the model-centric thinking of the early generative A.I. era. When ChatGPT sparked a global corporate rush in late 2022, Mehanna had already spent two years rebuilding the company's data and analytics backbone; his role was elevated to chief data and A.I. officer roughly eight months into the GPT era. That grounding in data infrastructure gave him a realistic view of what actually drives value at scale. His "pyramid" model—everyday tools at the base, embedded workflows in the middle, and product A.I. at the top—reflects a deliberate, phased approach rather than a reactive scramble. The shift from MyGPT to partnering with LangDock (a startup with $50,000 in annual recurring revenue at the time) underscores a calculated trade-off: rather than lock the company into a single vendor, Merck KGaA chose speed, sovereignty, and optionality. That choice is particularly significant for a 350-year-old company operating under strict European regulatory frameworks and strong labor-relations customs.

Mehanna's governance and ethics strategy further illustrates the gap between A.I. hype and enterprise reality. His comparison of guardrails to high-performance car brakes—essential to enable speed, not to slow it—reframes governance as enabler rather than brake. The Digital Ethics Advisory Panel's five-principle framework (autonomy, beneficence, non-maleficence, justice, transparency) and its emphasis on early risk identification before launch rather than post-hoc approval reflects hard-won experience managing large organizations through technological transitions. His earlier role at Mercedes-Benz (five years as chief data officer) clearly informed this philosophy. For Merck KGaA's 62,000 employees, the privacy protection (aggregate-level analysis only, no individual monitoring of the 12 million internally-generated prompts) demonstrates how to scale A.I. adoption while meeting both regulatory requirements and workforce expectations.

No comments yet. Be the first to share your thoughts!