Overcoming the organizational challenges of AI in healthcare: Reflections for executive leaders

• Always start with a clear business problem, not the AI technology itself
• Pilot small, measure impact, and involve cross-functional teams early
• Build trust through transparency, ethical design, and human oversight

As the healthcare environment evolves with the increasing speed of innovation, new artificial intelligence technologies are emerging, promising to streamline workflows and restore the focus to patient care. Yet, harnessing these powerful capabilities is not a simple matter of technical deployment. There are real struggles that come with implementing new technologies, no matter how promising. Overcoming the organizational challenges of AI in healthcare depend on the development of a clear-eyed strategy that takes into account an organization’s most pressing needs and its readiness for change.

We sat down with Shweta Maniar, Global Director for Healthcare and Life Sciences at Google, to help leaders navigate this landscape. She shares invaluable insights on how to prepare organizations, build trust in new systems, and ensure AI adoption delivers meaningful results from the start.

HT: What are some of the most impactful AI technologies transforming healthcare today?

Shweta Maniar: I would say there are four major types of AI that are changing the way healthcare is delivered today: generative AI, ambient AI, agentic AI, and medical algorithms. 

Generative AI is changing how we interact with data, whether it’s summarizing clinical encounters, drafting patient communications, or synthesizing literature.¹ Ambient AI, on the other hand, works behind the scenes to reduce the documentation burden for caregivers.²

Agentic AI is a game-changer because it’s collaborative, not just assistive. It takes action independently, within defined parameters, while keeping a human in the loop.³ Then we have specialized medical algorithms that are already making a difference in diagnostics, triage, and personalized treatment.⁴ These are the technologies I see making the biggest impact right now.

HT: What steps should organizations take to identify which AI solutions will produce the best results for their specific needs?

Shweta Maniar: First, stop thinking about the technology and start with the problem. Too often, leaders say, “I need generative AI,” without fully understanding what they’re trying to solve. I always say: you’re not looking for a hammer and a nail, you’re trying to hang something on the wall. Focus on that.

Consider the following best practices:

• Map your current processes: Where are the inefficiencies, the human errors, the duplication?
• Define clear, measurable outcomes: Pilot something manageable—don’t try to do it all at once.
• Understand AI solutions and pick the most appropriate tool: If your goal is reducing clinician burnout, ambient AI might be a good starting point. If it’s improving diagnostic accuracy, maybe it’s a medical algorithm. 
• Experiment first: Bring in clinical and operational leaders early, and run pilots in a controlled, human-in-the-loop environment. That’s how you will find what truly works in practice.

HT: What are the most common challenges to AI in healthcare that organizations face when integrating AI technologies?

Shweta Maniar: Organizational hurdles tend to fall into three categories: technical, administrative, and data-related. On the technical side, many organizations have legacy systems not designed for the volume or type of data AI needs.⁵ Interoperability is still a major hurdle.

Administratively, there’s often a lack of executive buy-in and a missing AI strategy. Change management is complex, especially in risk-averse cultures. There’s also a talent gap—few people are fluent in both healthcare and AI. Finally, on the data side, quality, completeness, and standardization are major issues. Data silos and unstructured formats make it difficult to train effective models.⁶ And of course, ethical considerations, especially around bias, must be front and center.

HT: What steps can organizations take to overcome the challenges of AI in healthcare?

Shweta Maniar: Start small with targeted pilots that let you measure impact and adjust. Don’t go too big too fast, it’s about iterative learning.

On the administrative side, invest in upskilling, create strong change management plans, and set up a governance committee that includes not just IT and legal, but clinicians and operations staff too. You need cross-functional input at every level.

For data, prioritize quality and consider using synthetic data (data designed to mimic the statistical properties and patterns of real-world data) where needed. Set clear ethical guidelines from day one and learn from existing data governance frameworks. This isn’t optional anymore, it’s foundational.

HT: How can organizations address concerns related to data privacy and regulatory compliance when implementing AI solutions?

Shweta Maniar: This is non-negotiable. Privacy must be embedded from the start, what we call “privacy by design.” You need robust security protocols, encryption, regular audits, and a deep understanding of evolving regulations like GDPR, HIPAA, and upcoming AI-specific laws.

It is important to involve the legal and compliance teams early. Don’t just hand them a strategy at the end. Build explainability into your models so non-technical stakeholders can understand how decisions are made. If a tool tells a radiologist to look at 10 images and skip 90, that radiologist needs to know why.

Trust is built through transparency and will generate greater buy in from the beginning from all stakeholders.

HT: Thank you for taking the time to think through this topic with us. One last question: What are you seeing in terms of upcoming AI trends in healthcare, and which of these do you think will have the most significant impacts?

Shweta Maniar: Multimodal AI is really exciting – integrating data across modalities like images, text, genomics, and wearables for a more comprehensive understanding of patient health.⁷

Federated learning is another key trend. Federated learning is a method for training AI models on decentralized data while protecting privacy. In healthcare, where patient data is highly sensitive and legally restricted from being shared, this is a crucial tool. The data never leaves its original location, such as a hospital or clinic. Instead of sending the raw data to a central server, the AI model itself is sent to each institution.

There, the model is trained locally on the institution’s data. Afterward, it sends a summary of what it learned, referred to as the model’s “weights” or “parameters,” back to the central server. The server then combines these summaries from all participating institutions to create a more robust global model. This cycle repeats, allowing the model to improve over time without ever compromising patient privacy. This approach enables a level of collaboration among institutions that was previously impossible, resulting in more effective and generalized models trained on diverse patient populations.⁸

Of course, with AI, hyper-personalization is set to go even further, using real-time data from genomics, wearables, and other inputs to generate care plans that are not static, but evolve with the patient’s condition. Finally, AI is transforming drug discovery, shortening R&D timelines, and accelerating how we identify new treatments.⁹