Ai4 2025: Health care’s AI moment — scale or be left behind

Last week, the only thing hotter in Las Vegas than the 109-degree weather was AI.

At Ai4 2025 — billed as “America’s largest AI event” — business leaders gathered at the MGM Grand on the Las Vegas Strip to explore applications of AI across industries. But the spotlight shone brightest on health care, with the sheer number of participating panelists. Those leaders made it clear the sector’s AI moment has arrived, and now the race is on to scale it.

With patient data volumes soaring postpandemic and the public expecting every interaction to feel personal, medical care is at a tipping point. And companies like Northwell Health, AstraZeneca and Headspace are embedding artificial intelligence deep into their operations to meet surging consumer expectations and accelerate research.

“Consumers now expect our outreach to reflect everything they’ve already told us,” said Paul Lampson, VP of customer analytics at Northwell Health.

The theme for the health care sector was not about selling AI’s value — it was about proving value with early results and looking toward a future that can deliver those results at scale.

From personalized outreach to predictive care

During a panel called “Data-Driven Marketing With AI: Extracting Value From Customer Data,” executives from health care, vision care and AI fields said clean data and measured experimentation are key to proving AI’s value and personalization is no longer a “nice to have.”

“If I give you my information, you should know me,” explained Rachelle Kuebler-Weber, CMO of AEG Vision. She says her team has found success in layering patient and operational data with marketing performance metrics and third-party insights to reach the right population. “If you can anticipate my needs before I realize them, that’s the sweet spot.”

Having that predictive care approach can save lives. Lampson recounted how, during the COVID-19 pandemic, Northwell Health used website behavior patterns to predict spikes in cases two weeks before testing data confirmed them, a model that drew national attention.

Jeremy Lyon, director of strategic AI consulting at Uniphore, urged brands to mine data that might not even be top of mind — from call logs to campaign retrospectives — as essential fuel for fine-tuning models. “That corpus becomes your business’s reality,” he said. “It’s essential for contextualizing campaign objectives and mobilizing the right channel strategies.”

AstraZeneca’s push for agentic AI at scale

In a separate fireside chat, AstraZeneca Chief Architect Wayne Filin-Matthews, a veteran of Microsoft, Dell, HSBC and Accenture, discussed how the pharmaceutical giant is embedding AI at the core of its ambitious growth plans. In five years, AstraZeneca aims to launch 20 new medicines, double annual revenue from $50 billion to $100 billion and build a $50 billion manufacturing site in Virginia, all without a proportional increase in its 10,000-strong IT workforce.

Filin-Matthews noted a pivotal industry shift in early 2024 toward agentic AI — autonomous platforms that work together like teams — that now makes this vision possible. “Agents are the only way we can meet our science, operational and sustainability goals at this pace,” he said. But, he said, scaling these agents in a regulated industry requires building a “composability layer” so they can work across platforms securely and cost-effectively.

He also emphasized the importance of understanding the “cognitive behavioral” dynamics of AI teams, likening them to human teams with diverse personalities. Even a procrastinator agent can be useful in certain scientific collaborations, he joked. AstraZeneca is partnering with Stanford University to model these dynamics for research purposes, aiming to accelerate molecule prediction without increasing head count.

Headspace’s AI therapist gets thousands of fake patients

Headspace is using autonomous AI agents to recreate the rigor of real therapy sessions, and the app might have cracked the code on at least scaling its testing approach.

In a panel session all about the challenges and opportunities of AI agents, Headspace’s lead AI scientist, Akhil Chaturvedi, described the system as “an operating system with LLMs at the core,” designed to break down the science of psychology into elements that can be trained into autonomous agents. Those agents draw on patient memories, pull from specialized databases and deliver an ongoing therapeutic experience. Every conversation is scored using a rubric by a mix of human evaluators and AI judges.

To speed testing, Headspace built a library of thousands of synthetic mental health patients who “talk” to the AI therapist after every model tweak, “allowing company to understand “exactly how well [the AI agents] felt heard,” Chaturvedi said. About 95% of evaluations are synthetic, with 5% coming from real human conversations.

The vision is clear but not immediate

The vision is clear. But there are obstacles to delivering AI solutions across all facets of the industry at scale.

For Kuebler-Weber, the challenge is scaling across AI across AEG Vision’s more than 500 independently branded optical practices, for instance.

Despite the momentum at AstraZeneca, Filin-Matthews estimated that fully mature, large-scale agentic AI in the enterprise is still many years away. “In five years, we’ll have learned a lot, failed in some areas and still not be where we need to be in governance and control,” he said.

Still, all executives agreed, when scaling AI: Start small, ensure data quality and prove ROI early.

“Get AI use cases in market, even in a limited way,” Lampson advised. “Show the pipes can work.”