While AI is already revolutionizing patient care, research, and medical education, the path from promise to practice has its challenges. How do you ensure AI delivers real value rather than just adding complexity — and stands in absolute integrity when it comes to patient care and protection of personal information?
Stanford Medicine has developed a comprehensive approach to AI implementation that balances innovation with responsibility. In a recent conversation on the Box AI-First Podcast, Dr. Mike Pfeffer (Chief Information and Digital Officer) and Dr. Todd Ferris (Deputy CIO) shared how Stanford Medicine is deploying AI at scale, from ambient scribing in patient visits to automating workflows in clinical trials.
Read on to see how one of healthcare’s most innovative institutions is transforming patient care with Intelligent Content Management. Stanford’s approach offers valuable lessons for other life sciences organizations looking to harness the power of Content + AI.
Key takeaways:
- Democratize AI across the organization: Rather than creating an isolated AI team, Stanford Medicine made AI tools accessible to everyone to unleash innovation from unexpected quarters
- Prioritize workflow integration over technology sophistication: The most advanced AI model will fail if it doesn’t fit seamlessly into clinical workflows, and Stanford’s year-long rollout of AI scribes demonstrates the value of patience and careful implementation
- Build governance for continuous learning, not just initial deployment: Stanford’s frameworks for ongoing assessment ensure AI projects continue delivering value over time as environments and AI systems evolve
AI is the responsibility of everyone (not just IT)
Stanford Medicine rejects the “elite AI team” concept. Instead of creating a specialized group to own all AI initiatives, the team embeds AI capabilities throughout the organization. This is a philosophical shift from a more traditional technology approach.“Technology is not under the ownership of the CIO per se,” says Dr. Pfeffer. “It’s becoming more democratized than ever before. We created a secure platform for everyone at Stanford Medicine to be able to use the models in a safe way.”
Technology is not under the ownership of the CIO per se, it’s becoming more democratized than ever before.
Beyond just providing AI access to all teams, Stanford Medicine has invested heavily in education, requiring all technology team members to complete AI training. As Dr. Ferris explains, this comprehensive approach ensures informed decision-making: “One of the strategies that we employed was enabling all of our Technology & Digital Solutions (TDS) team members to level up on AI. That way, when they’re evaluating working with vendors, they understand the pros and the cons.”
As a result, rather than waiting for top-down directives, staff across the organization can identify AI opportunities and proactively develop solutions. This grassroots innovation has led to unexpected applications, from network teams using AI to automatically configure routers to clinicians developing tools for patient care.
The critical distinction between automation and augmentation
Stanford Medicine’s framework for evaluating AI applications divides potential use cases into two categories: automation and augmentation. This distinction helps the team prioritize implementations and set appropriate expectations. Dr. Pfeffer provides clear examples of each approach.
“Automation is taking a task that humans can do today and making it much easier,” he says, giving the example of ambient scribes. These tools listen to physician-patient conversations and automatically generate clinical notes. The impact on patient care has been profound — physicians in the room with a patient can maintain eye contact and focus on the human connection rather than being distracted by typing on a device.
“Augmentation,” Dr. Pfeffer says, “is doing things humans can’t do by themselves — or the computer can’t really do by itself.”
Automation is taking a task that humans can do and making it much easier. Augmentation is doing things humans can't do by themselves.
These projects tend to tackle more complex challenges. Dr. Ferris describes one innovative solution developed with a faculty member who’s spent her career helping students think about their habits with patients.
“When you approach a patient,” wonders Dr. Ferris, “How can you be systematic and make sure you’re approaching the conversation in a thoughtful way? It requires skill.”
Stanford used AI to develop a new tool that augments the teaching of faculty members in this domain. It listens in on conversations between faculty members and students and helps point out where they might have missed something.
Workflow integration determines success or failure
Perhaps the most emphatic message from both leaders is that technology alone doesn’t determine success. The key lies in how well AI integrates into existing workflows.
“I’ve been in health IT and informatics for over 15 years now,” says Dr. Pfeffer, “and I can tell you that it’s all about the workflow, the implementation, and the training. The technology is the easy part.”
This insight shaped Stanford Medicine’s approach to implementing AI ambient scribes, which took over a year to fully deploy across all specialties.
It’s all about the workflow, the implementation, and the training. The technology is the easy part.
The careful attention to workflow extends to patient interactions. Before implementing any patient-facing AI tool, Stanford Medicine considers how it will affect the care experience. The organization trains physicians on how to introduce AI tools to patients thoughtfully, always respecting patient preferences about their use. This approach has resulted in high acceptance rates, with patients appreciating how AI enhances (rather than disrupts) their care.
Embracing the probabilistic nature of medicine and AI
Healthcare professionals are accustomed to working with uncertainty. This cultural reality actually facilitates AI adoption, as Dr. Ferris explains: “Medicine is inherently a probabilistic endeavor. When lab tests come back, there is some rate of error in them. Clinicians are trained to deal in probabilities, so they’re very accustomed to uncertainty.”
This perspective helps Stanford Medicine set realistic expectations for AI tools. Rather than seeking perfection, they focus on improvement. If an AI model can provide better insights than current methods — even if not perfect — it can add value. Stanford Medicine has developed frameworks like MedHelm to measure how generative models perform on healthcare tasks, ensuring continuous improvement while acknowledging inherent limitations.
Governance frameworks that enable innovation while ensuring safety
With patient safety at stake, Stanford Medicine has developed comprehensive governance structures that don’t stifle innovation. Every AI implementation undergoes ethics assessments and risk evaluations. More importantly, the organization has built systems for continuous monitoring.
“Models are taking inputs. A health system is a living, breathing organism, lab tests change, the way we record information changes,” Dr. Ferris notes. “So we need to be always monitoring that and making sure that it’s behaving the way we expect it to behave.”
Governance extends beyond technical performance to include value assessment. Dr. Pfeffer emphasizes the importance of measurement: “You need a framework to evaluate these things and then follow up if they’re performing as they need to be. And if they’re not, turn them off, because they will live forever — and they could create problems later.”
You need a framework to evaluate these things and then follow up if they’re performing as they need to be. And if they’re not, turn them off.
Building for a personalized future in medicine
Stanford Medicine is pushing the boundaries of what’s possible with AI in healthcare — for instance, exploring novel applications of transformer models to predict medical events and outcomes, particularly in cancer care. These initiatives point toward a future where treatment decisions will be increasingly personalized based on comprehensive analysis of patient data.
This organization’s journey offers a roadmap for healthcare organizations navigating the AI revolution. By maintaining focus on their core mission while embracing thoughtful experimentation, Stanford Medicine has created an environment where AI enhances rather than disrupts healthcare delivery. As healthcare organizations worldwide grapple with AI implementation, Stanford’s emphasis on democratization, workflow integration, and continuous governance provides a proven framework for success.
Catch the full episode
This episode of the Box AI-First Podcast covers lessons learned from integrating AI into Stanford Medicine’s innovative environment, the cultural shifts that made it possible, and how the team ensures trust and safety in every AI implementation. Watch the full episode.
