Riding the AI Tsunami | How Organizations Are Driving Cultural Change

Session Focus

Khan opened with the two questions she is asked whenever someone learns what she does: what is going to happen to my job, and what should my kids study in school. Both reflect how deeply the pace of AI progress cuts at a personal level, magnified enormously when leading organizations.

Her framing of the problem: many companies are under huge pressure to adopt AI, yet surveys show organizations still focused on pilots, which some would frame as AI slop, rather than fundamentally reimagining businesses. The panel was designed as a candid conversation about where things actually stand.

Start With Skepticism About Value

Kyle Bryant opened by insisting the value conversation comes first, citing a healthcare CFO survey finding that around 18 percent of organizations well into their AI journey cannot attribute financial outcomes to their efforts.

His two conclusions: AI is not always the right tool for the problem, and many well-intentioned efforts are not driving value while actively adding cost to the system. He acknowledged this was a downbeat opening while arguing the skeptical eye matters.

He divided the work into three categories.

Personal productivity, typically at the individual level, including coding tools, chatbots and increasingly powerful workflow and agent builders. ROI is difficult to measure here, and it demands substantial training, enablement and culture change.

Process transformation, covering core operational and service delivery workflows where reimagining or rewiring the process drives value. Revenue cycle management is where he sees the clearest split: some players simply add AI onto a current process and make things worse, while others rebuild end to end.

Product transformation, adding AI features or building new products. He singled this out as interesting from an access perspective because it changes scaling economics, creating the ability to serve customers and segments that previously could not be served well.

Asked where he does not see it working, he declined to name a use case and redirected to how initiatives are enabled: C-suite buy-in, and involving subject matter experts and end users in building the systems and workflows.

Ambient Documentation as a Doorway

David Rhew told a story from the previous day, when a physician approached him in the speakers room. Years earlier Rhew had demonstrated ambient voice technology at the Harvard School of Public Health, and afterward the students agreed it would never happen because healthcare moves too slowly. Within one year they were using it.

His point was about what made adoption fast: the tool addressed a critical pain point, saving clinicians time and removing the administrative burden of typing.

But he framed ambient documentation as an opening rather than an endpoint. Digitizing an encounter means AI can be applied to everything said and how it was said, combined with information from the medical record, supporting decision support, performance optimization and care coordination.

The same logic extends to ambient vision and computer vision, where the system analyzes, interprets and enables action in ways not previously possible.

Triage Before the Visit

Ron Walls MD began with a governance principle. Healthcare needs a different approach from other industries, and a different approach within itself: the scrutiny and oversight appropriate for a revenue cycle model differs substantially from what is required when patient care is affected.

After ambient documentation, the system identified its biggest challenge as connecting patients with primary care against a national shortage, where even patients who have a doctor cannot reach them in a timely way.

The response was MG Care Connect, built with a partner. Patients enter through Patient Gateway, the Epic MyChart interface they already use, and begin a conversation with an AI agent Walls described as astoundingly human sounding.

The agent conducts an intake interview much as a nurse would, establishing the problem and determining how quickly the patient needs to be seen and in what format: a live doctor, a virtual doctor, virtual urgent care or in-person urgent care.

The part Walls emphasized is what the physician receives afterward. Not a summary that the patient thinks they have a bladder infection, but a pre-populated Epic note reading as though the visit has already begun, describing a 23-year-old woman with urinary burning beginning the previous morning at 11 o’clock, a developing sense of urgency, no back pain, no fever, no chills. Relevant information from the medical record is scanned and pulled in.

The effect is that the physician can focus on clinical decision-making rather than history gathering, lifting burden much as ambient documentation does.

The results he reported: 77 percent of patients see a doctor the same day they access it, which he called profoundly different from the usual primary care model. Net promoter scores are very high. The program launched almost exactly a year ago, saw 36,000 patients in its first year, and now runs at 150 patients a day.

Extending the Same Infrastructure

Khan connected the triage infrastructure to a problem in her own field. Roughly 4 to 5 percent of patients enroll in trials for innovative therapies under development. The leading reason is lack of awareness, and the second is lack of real-time matching, including identifying patients who have failed one therapeutic and become eligible for another.

With Recursion running programs in late-line solid tumors and rare diseases, she raised whether infrastructure like MG Care Connect could match patients to investigational therapeutics early.

She also raised the equity dimension for Rhew: community clinics are the most disadvantaged by any quality improvement gap, with physicians and nurses already overburdened, which makes extending quality of care beyond top-tier hospitals a distinct problem from improving care within them.

Key Takeaways

1. Roughly 18 percent of organizations well into an AI journey cannot attribute financial outcomes to it.

2. Three categories carry different ROI profiles: personal productivity, process transformation and product transformation.

3. Bolting AI onto an existing process can make it worse, which is the visible split in revenue cycle deployments.

4. Adoption speed follows pain point severity. Ambient documentation spread in a year after being dismissed as science fiction.

5. Governance should scale to clinical risk, with revenue cycle models held to a different standard than care decisions.

6. AI intake produced same-day physician access for 77 percent of patients, with 36,000 seen in the first year.

7. Trial enrollment fails on awareness and real-time matching, which the same triage infrastructure could address.