Prior Auth, Denials, and Data Chaos: A CFO's Guide to Making Healthcare AI Work | Context 2026

Most health systems are buying AI. Far fewer are making it work. The difference comes down to data, governance, and the discipline to ask the right questions before writing the check. In this session from Context 2026, Kelly Caner, Managing Director of RCM Solutions at Innovaccer, sits down with Mark Cohen, CFO at University of Rochester Medical Center, for one of the most grounded conversations of the conference. From prior authorization wins to failed ortho implementations, from employee buildathons to a 12-person governance team, this is what healthcare AI actually looks like inside an $8 billion health system. In this video: → Why data fragmentation is the real ceiling for AI in healthcare → How University of Rochester structured AI governance without slowing down → What a 500-person employee buildathon taught them about scaling AI ideas → Where revenue cycle AI is delivering the clearest financial returns today → Why front-end RCM is where the biggest AI opportunity lives → How to measure ROI when the returns are not always in dollars Chapters: 00:00 Introductions — Kelly Caner and Mark Cohen 03:05 What makes AI more than another technology trend for a CFO 04:51 The biggest obstacles preventing health systems from scaling AI 06:01 Why prior auth for imaging worked and ortho failed 07:22 Foundational capabilities AI solutions actually need 10:05 University of Rochester's health lab and innovation programs 10:59 The 500-person employee buildathon and what it produced 13:26 Generating AI ideas versus operationalizing AI at scale 15:33 How governance works at University of Rochester — the 12-person team 17:02 How a CFO decides whether an AI investment deserves funding 18:17 DAX ambient listening — the rollout story and the results 21:25 Is data infrastructure investment more important than AI investment? 25:41 Operational realities that create hidden barriers in the data 27:53 Where revenue cycle AI delivers the greatest financial impact 30:53 Using back-end denial data to train front-end AI solutions 32:10 End user adoption — the blockade nobody talks about enough 33:46 What to monitor to know AI is actually improving outcomes 37:18 Good enough data — when to start versus when to wait 39:18 Top two categories of patient data for financial predictability 40:40 Closing thoughts and advice for CFOs