How Philips is scaling AI literacy across 70,000 employees

🚀 Why company-wide AI literacy matters

I led an effort at Philips to move AI from isolated projects to a company-wide capability. For years we had highly specialized teams building AI into medical devices, consumer products, and analytics platforms. That expertise was real and powerful, but it was concentrated. If AI is going to transform how we work, compete, and serve patients, it cannot live only in pockets of expertise. It has to become part of how every one of our 70,000 employees thinks and operates.

There are three reasons I pushed for broad AI literacy across the organization.

Operational impact - AI can shave hours off recurring tasks, freeing clinicians, engineers, and customer service teams to focus on higher-value work.
Strategic innovation - When more people understand the tools, use cases naturally emerge. Innovation multiplies when knowledge is distributed.
Risk management - Widespread literacy helps people use AI responsibly and align with governance, privacy, and safety practices.

One phrase captures the progression I want people to experience: "you start playing with it, then you start working with it, and from there you start innovating with it." That is the journey we built into our rollout.

🎯 Starting from strength: Philips’ AI roots

It is important to be clear: AI is not new to Philips. We have machine learning and AI embedded in many of our products. Our specialized teams have been delivering solutions that improve imaging diagnostics, patient monitoring, and device performance. That foundation gave us two advantages:

Credibility when pushing AI adoption into the broader workforce.
An appreciation for the difference between traditional machine learning projects and the new wave of generative AI tools that people can use interactively.

But having advanced AI inside products is not the same as equipping 70,000 employees to use AI tools in their daily work. That required a deliberate shift in approach.

🧭 The rollout strategy I used

Designing a scalable adoption program meant balancing education, experimentation, governance, and incentives. I structured the rollout in a few clear phases that other organizations can replicate.

Phase 1 - Executive and leadership alignment

Before launching anything widely, I trained our top leadership team—the XCO members—personally. Executives need to model new behaviors and make decisions about risk and investment. If leaders are unclear or hesitant, adoption stalls. So I spent time showing them what these tools can do and how to think about them strategically.

Phase 2 - Selective pilots and early adopters

We deployed ChatGPT Enterprise to a few thousand users initially to learn how the tool performed in real workflows. These early pilots let us answer key operational questions: How will users adopt it? What value are they finding? Where do we need guardrails or integrations? Running pilots at scale gave us meaningful data before a broader rollout.

Phase 3 - Company-wide rollout and literacy programs

Once pilots validated value and governance, we expanded access and focused on AI literacy. The goal was not to turn everyone into data scientists. The goal was to help everyday employees use AI safely and effectively to accelerate routine tasks, create clearer documents, and prototype new ways of working.

Phase 4 - Innovation and scaling

As people began using AI in daily tasks, we encouraged them to be creative about where AI could drive impact. That included a company-wide use case challenge to surface the best ideas and a plan to integrate the most promising projects back into product and service roadmaps.

🏆 The summer challenge: turning demand into ideas

When we announced we would roll out Enterprise ChatGPT, demand surged. Licenses were limited at first, so we had to choose who got early access. Instead of a simple lottery, we turned it into an opportunity.

I launched a summer challenge: sign up and present your brightest idea for how AI could be used at Philips. The best submissions earned a license. The competition did several things for us:

Generated high-quality use cases from across functions and countries.
Built excitement and motivation to learn the tools before access was granted.
Focused scarce resources on demonstrable ROI and practical experiments.

This approach reinforced a simple truth: people want to use AI, but they also want clear purpose and outcomes when access is scarce.

🧠 Training for impact: what we taught leaders and teams

Training focused on three practical competencies.

Tool fluency - How to prompt, iterate, and use ChatGPT Enterprise productively while protecting sensitive information.
Use case identification - How to surface eligible tasks for automation or augmentation, emphasizing administrative or repetitive activities that free time for higher-value work.
Governance and risk awareness - Clear rules about data privacy, IP, and when to involve specialized AI teams.

Practical exercises were essential. Executives and managers practiced rewriting clinical summaries, generating meeting notes, and creating test prompts for common workflows. I emphasized that literacy is not about mastering an algorithm; it is about understanding how to apply a tool responsibly and creatively.

💡 Use cases that surfaced quickly

Across pilots and early adopters, some use cases repeatedly delivered immediate benefits. These became our low-hanging fruit for company-wide adoption.

Administrative automation - Drafting reports, generating meeting summaries, formatting clinical documentation, and organizing emails.
Clinical documentation support - Helping clinicians translate notes into standardized records, which reduces back-office burden.
Customer-facing content - Creating clearer product explanations, troubleshooting scripts, and multi-language support content.
Research and synthesis - Rapidly summarizing literature, extracting key points from large documents, and preparing concise briefings for decision makers.

One concrete example stuck with me. I was in a hospital where an emergency case arrived and a clinician spent 15 minutes in the cath lab to save a life. Afterwards, they faced an equivalent amount of administrative work. If we can shave administrative time in half or more, clinicians could potentially save additional lives or spend more time with patients. That is the kind of operational leverage AI can deliver.

🔒 Balancing access with governance

Scaling access to AI requires thoughtful guardrails. We combined technical controls with policy and education.

Technical controls

We used enterprise-grade deployment of ChatGPT that included features designed for organizations: data protections, single sign-on, and administrative controls. Those capabilities let us enforce boundaries while allowing creative use.

Policy and training

Policies explained what types of data could be processed, how to handle personal health information, and how to ask for help from specialized AI teams when projects moved beyond generative assistance into product development.

Human oversight

AI assisted outputs always required human validation. We trained people to see AI as an assistant that can speed work and present options, not as an infallible oracle. This mindset is crucial in a healthcare context where errors have real consequences.

📈 Measuring adoption and value

To understand whether the program was working, we tracked a mix of quantitative and qualitative indicators.

Usage metrics - Number of active users, sessions, and prompts per user.
Efficiency gains - Time saved on administrative tasks, number of documents generated or completed faster.
Project outcomes - Use cases that progressed from pilot to production, including estimated cost savings or revenue impact.
User sentiment - Feedback from clinicians, engineers, and managers on usability and trust.

These measures helped us refine training and prioritize integrations. They also made it possible to demonstrate ROI to stakeholders and encourage further investment.

🧩 Integrating specialized AI teams and enterprise users

One tension I navigated was how to align our existing specialized AI teams with the wave of everyday AI use. The two need each other.

Specialized teams continue to build models and product-grade solutions that require deep domain knowledge and rigorous validation. At the same time, general employees using ChatGPT Enterprise can prototype ideas quickly and uncover new product opportunities for the specialists to scale.

That relationship works best when clear pathways exist:

Prototyping guidelines for employees to ensure responsible experiments.
Handoff processes for promising prototypes to the specialized AI teams.
Joint review sessions where teams and business stakeholders align on viability and compliance.

🛠️ Practical guidelines I taught teams for effective prompts

Prompt engineering does not need to be mysterious. I gave teams a few practical rules that anyone could apply:

Start with the goal - Describe the desired outcome, not the tool. For example, "Create a 300-word patient-friendly summary of this procedure" beats "Write something about this."
Provide context - Give the model the necessary background. Short excerpts, bullet lists, or a brief description of the audience help produce targeted outputs.
Iterate quickly - Ask for a draft, then request modifications. Use follow-up prompts to clarify tone, format, or detail level.
Require citations - For clinical or technical outputs, ask the model to list sources or highlight where human review is essential.
Protect sensitive data - Never include personal health information or proprietary IP in prompts unless the tool and policies explicitly permit it.

📚 Building AI literacy as a capability

Learning at scale needs structure. I treated AI literacy like a capability to build, not a one-off training event. That meant investing in:

Learning pathways for different roles
Train-the-trainer programs so managers could cascade knowledge
Resource hubs with templates, use case libraries, and governance checklists
Community channels where practitioners shared tips, prompts, and outcomes

By making literacy reusable, we empowered local teams to adapt the approach to their context. A radiology department needs different examples than a supply chain team, but both benefit from the same core principles.

🌱 Culture shift: from fear to experimentation

One of the biggest obstacles to adoption is fear—fear of job loss, fear of mistakes, and fear of regulatory backlash. I leaned into a narrative of empowerment. I framed AI as a way to reclaim time for higher-value, human-centered tasks.

"Our healthcare practitioners are spending way too much time on administrative burden. How can we take that away from them and give them time back to spend with the patient?"

That question resonated. It grounded the technology in purpose and gave people a clear metric for success: more patient time, better patient outcomes, or faster decisions.

To keep experimentation safe, we created spaces where failure was low-risk. Small pilots with clear scope allowed people to try new workflows, learn, and share lessons without fear of major consequences.

📣 Communication and change management

Successful change requires clear communication. We used three communication levers.

Transparent announcements about where AI was being used, what it could and could not do, and how employees could get access.
Success stories from pilots that demonstrated tangible benefits and practical tips.
Open channels for questions and concerns, including forums where governance experts answered compliance questions.

People needed to hear both the opportunities and the constraints. Honesty about limits builds trust and helps teams make better decisions about where to invest their time.

🔁 From pilots to production: how ideas scale

Not every idea from the summer challenge or pilots should go into production. We judged scale candidates on three criteria:

Impact - Measurable time savings, cost reduction, or improved outcomes.
Feasibility - Can the idea be integrated into existing systems and workflows?
Compliance - Does it meet privacy, safety, and regulatory requirements?

When a project met these standards, we supported it with product-level engineering, quality assurance, and governance sign-off. This combination preserved innovation while ensuring reliability.

📎 A few real-world examples

To illustrate how the approach plays out, here are condensed examples inspired by early pilots.

Clinical note summarization

An emergency department team used an AI assistant to draft standardized clinical notes from clinician bullet points. The output reduced administrative time by an estimated 40 percent per case, with clinicians validating and editing the draft rather than composing from scratch.

Technical documentation and translations

A product support team used AI to create first drafts of user guides and to translate troubleshooting steps into local languages. Editors then refined the output. This cut turnaround time from days to hours in many cases.

Regulatory briefing synthesis

Regulatory affairs teams used AI to summarize long policy documents and create concise briefings for executives. That allowed faster, better-informed decisions without replacing the essential human legal review.

📌 Lessons learned

After running pilots and expanding the program, several lessons became clear.

Start with clear problems, not tools - AI adoption accelerates when the focus is on solving a specific pain point.
Leadership matters - When executives model curiosity and responsible use, adoption rises.
Make learning practical - Short, hands-on training beats long theory sessions.
Governance must be practical - Overly restrictive policies kill innovation; lax policies create risk. Find balance with clear exceptions and pathways.
Measure outcomes, not vanity metrics - Track time saved and impact on patient or customer outcomes.
Keep humans in the loop - AI amplifies human capabilities. It should not replace critical human judgment, especially in healthcare.

🔮 What comes next

Once AI literacy spreads and the early wave of experiments mature, the next phase is product integration and deeper automation. That means building validated models into clinical systems and enterprise workflows with the same rigor we apply to any medical technology.

But I also want to keep the spirit of experimentation alive. People who "start playing with it" discover friction points and new product ideas. We want that curiosity to flow into our product teams so we keep moving forward. As I said earlier, the aspiration is big: "let's go for the moon. Let's go for Mars." In practice that means dreaming big while methodically proving value, step by step.

🧭 Practical checklist for organizations starting today

If you are leading a similar effort, here is a compact checklist you can apply immediately.

Train leaders first - Give executives hands-on sessions so they can model usage and make informed decisions.
Run focused pilots - Start with a few thousand users in high-impact areas like clinical documentation or customer support.
Incentivize creative ideas - Use challenges or competitions to surface high-impact use cases.
Put governance in place - Combine technical controls with clear policies and training.
Measure outcomes - Track time saved, user satisfaction, and production readiness.
Create handoff paths - Ensure prototypes can be escalated to specialized AI teams for scaling.
Build reusable learning - Create templates, playbooks, and communities to spread best practices.

🤝 Closing thoughts

Scaling AI literacy across a large organization is as much a people challenge as a technical one. You need leadership, clear incentives, safe spaces to experiment, and guardrails that protect patients and customers. When you get the balance right, the benefits are tangible: clinicians spend more time with patients, teams produce higher-quality work faster, and the organization discovers new areas to innovate.

I still believe in a simple progression: people will start by playing with AI, then incorporate it into how they work, and finally push the boundaries of what the organization can do. If you focus on literacy, measurement, and responsible experimentation, you can turn that progression into sustained capability across tens of thousands of employees.

"AI is not new to Philips. We have AI embedded in many, many of our products. But most of that is traditional AI machine learning. We also felt it's really the time to elevate AI literacy across all of the 70,000 employees."

That remains my perspective: specialized AI teams will continue to build the next-generation products, but broad literacy creates the demand, use cases, and informed users that make enterprise-grade AI worth the investment. The work is ongoing, but the direction is clear: give people the skills, give them safe access, and watch innovation emerge from all corners of the organization.