Trust is the real AI metric in regulated operations

For decades, pharmaceutical manufacturers have faced the difficult challenge of modernizing production operations while maintaining the rigorous compliance standards required to protect patient safety., Across the industry, a large majority of pharmaceutical production documentation is still paper based today. Every manufacturing recipe, the materials, the equipment settings, the tolerances, the sequence of steps that turn raw ingredients into a sterile, dozed product, has typically been written down, filed, and updated by hand.

That persistence is not a failure to modernize. It is a rational response to what paper offers: it is legible to a regulator, traceable to an inspector, and defensible in an audit. So, when generative AI promises to compress months of manual recipe review into hours, pharmaceutical leaders are right to ask a harder question than "how much faster." The real question is: faster, and still trustworthy?

This is the tension now facing manufacturing and quality leaders across the industry as generative AI moves from back-office pilot to production-adjacent tools. Speed is an easy metric to celebrate. Trust is the harder one to build, and the only one that determines whether an AI capability survives contact with a regulator.

Paper based processes are slow

A single production site can hold hundreds, sometimes thousands, of manufacturing recipes, many spanning thousands of individual process steps. Updating even one of them for a new regulatory requirement is not a copy-edit, it is a forensic exercise.

"You have to read through all those thousands of pages and create the very first stable recipe," says Shaun Tyler, Vice President Artificial Intelligence, Software at Körber Business Area Pharma. "This takes sometimes months, sometimes up to half a year with several people involved."

Regulatory change compounds the problem. When a new Good Manufacturing Practice (GMP) requirement lands, manufacturers must determine exactly where it touches their existing recipe before they can even begin to comply. "The difficult part is that you don't actually know where the new regulation affects your recipes," Tyler explains. "This might be a very long process."

The difference to digitalizing other kind of business documents

What makes this different from digitizing any other kind of business document is what sits on the other side of a mistake. A configuration error in an ordinary workflow tool is an inconvenience. A configuration error in a pharma manufacturing process e.g. a sterilization step, a wrong temperature, or a wrong hold time, can mean contaminated product reaching a patient.

"We can't have the remote possibility that something is incorrect," Tyler says.

Leaders are not being asked to choose between innovation and caution. They are being asked to define, precisely, which categories of information can tolerate AI-assisted judgment, and which must remain numerically exact, no matter how fast the surrounding system moves.

Julia Schmitz, Head of Global Software Integration, Software at Körber Business Area Pharma, frames the ambition beyond raw speed: "Once recipe information becomes structured and trusted, customers can begin interacting with it in entirely new ways." The strategic point is not that AI reads documents faster than people, but that the result is structured, trusted information that becomes an asset an organization can query, challenge, and improve continuously.

Building trust in the architecture, not bolting it on afterward

For AI to be usable in this regulated environment, traceability of the results cannot be a feature added after deployment, it has to be a condition of deployment. "Everything needs to be traceable," Tyler says. "Every version we changed needs to be perfectly documented."

Distinguishing between what AI can interpret flexibly and what it must report with zero tolerance for error is, in practice, the difference between being helpful and a liability. Internal testing from Körber's collaboration with Microsoft on Azure reported precision rates near 99 percent for general pharmaceutical interpretation tasks, alongside 100 percent numerical precision for production-critical values.

"No one would trust a medicine produced with a high hallucination rate when using AI," Tyler notes

First measurable efficiency gains

The earliest returns from this approach are already measurable. In early implementations, manufacturers have reported recipe digitization proceeding roughly 30 percent faster, with manual review work reduced by up to 40 percent.

But Schmitz is careful to frame the goal beyond a one-time efficiency gain: " By creating digitalized reipes our customers are creating a living knowledge base that engineers can query, update, and continuously improve with AI assistance while maintaining the controls required for regulated manufacturing."

A new generation AI native platform for trusted AI results

Körber is partnering with Microsoft to build the next generation AI native platform for trusted AI results from the ground up. Every feature has been designed with generative AI at its core, running entirely on the Microsoft Azure cloud stack.

To create a scalable, compliant AI foundation for pharmaceutical manufacturing, Körber has developed and embedded the Pharma Guardrails technology in it’s AI platform and solutions. Pharma Guardrails are the foundation for trusted AI and pharma and ensure that every AI generated output is accurate, compliant and audit ready. The accuracy Körber has engineered with pharma guardrails on the Azure platform generates a much higher precision rate.

What previously required weeks or even months of manual work per recipe is now completed in minutes with AI generated output ready for human review and qualification.

Oliver Weber, Senior Vice President Software at Körber Business Area Pharma, places the shift in a longer arc: "Our vision is not only to help manufacturers digitalize today's processes. We're building an AI-native platform that enables pharmaceutical organizations to modernize operations responsibly while maintaining the trust, quality, and compliance that patient care demands."

Summary

Generative AI is accelerating pharmaceutical recipe management, but speed alone is not the metric that determines whether an AI capability can be used in GxP-regulated production. What matters most is trust in the AI results and the ability to consistently demonstrate, document, and justify every AI-supported decision to a regulatory authority Early implementations already show measurable gains, roughly 30 percent faster recipe digitization and up to 40 percent less manual work The same standard extends beyond software, into inspection, aseptic processing, packaging, and equipment qualification, wherever AI touches a regulated production environment.

Pharmaceutical leaders weighing how to scale AI responsibly in regulated environments are invited to share notes with Körber Pharma's software experts team.

For a detailed account of how Körber built its AI-native platform on Microsoft Azure, including the governance approach, read the full story: Körber accelerates GxP operations with trusted AI leveraging Microsoft Azure.