Day 3 of Becker’s 11th Annual IT + Revenue Cycle Conference included sessions that cut through the noise around AI and focused on what is changing — in revenue cycle operations, IT investment strategy, workforce development, and the design of health systems built for the next decade.
Across four sessions, a consistent message emerged: the organizations winning with technology are not the ones moving fastest. They are the ones that have done the harder work of building governance, investing in people, defining problems before shopping for solutions, and treating AI not as a technology initiative but as a new operating model.
Here are 10 trends driving that conversation:
1. Automation and AI are not the same thing, and conflating them is slowing health systems down.
One of the most consistent points across Day 3 was the industry’s tendency to use “AI” and “automation” interchangeably — and the cost of that confusion. Automation handles predictable, rules-based workflows: claims scrubbing, work queue routing, document retrieval. AI does something different — it predicts, reasons and surfaces patterns that no human workflow could find at scale.
Health systems that treat every automation win as an AI story are overpromising to leadership and underinvesting in the tools that will actually deliver transformational value, leaders said. Separating the two is not a semantic exercise; it is a strategic one that shapes how systems build, evaluate and govern their technology portfolios.
“Oftentimes we confuse AI with automation, and automation with AI,” said Courtney McNamee, director of revenue cycle and reimbursement at Altru Health System in Grand Forks, N.D. “AI is really good at finding a needle in a haystack, but you still need a person to help fix it or do the work that needs to be done.”
2. Revenue cycle is getting out of the transactional business and into the predictive one.
The trajectory for revenue cycle operations is clear: Automation absorbs the transactional work, freeing experienced staff to focus on prediction, analysis and strategy. Health systems that have gone furthest on this journey describe a meaningful shift in how their teams spend their time: less on claims processing, more on identifying patterns, anticipating payer behavior and intervening before problems compound.
The ratio of transactional to analytical work is shifting, and the organizations that invest in reskilling their revenue cycle teams alongside their technology deployments are the ones seeing the most durable results, leaders said. Critically, this transition requires staff to understand not just that their work is changing, but why.
“We have about 800 people in our revenue cycle operation, and years ago 80% were dedicated to transactional work. Today I’d say we’ve flipped the script: It’s probably 35%-65% transactional to predictive and critical thinking, with the hope of getting to 50-50 in the next couple of years,” said Drew von Eschenbach, vice president of enterprise revenue cycle at UW Medicine in Seattle.
3. The payer-provider AI arms race is real and providers are still catching up.
Payers have been using AI to manage claims, downcode encounters and deny care for years, leaders said. Providers are now deploying AI to fight back: automating appeals, analyzing payer policy documents at scale and using bots to overwhelm payers with volumes of legitimate disputes they are contractually obligated to review.
The dynamic is intensifying. Some payers are attempting to insert contract language that limits providers’ use of automation and AI in the appeals process, even as they deploy it aggressively on their own side, leaders said. Panelists were clear that this is not a zero-sum game heading toward resolution; the friction will grow before it eases, and the providers that fall behind on AI capability in revenue cycle likely will face mounting financial exposure as a result.
“We need to go a lot faster. We’ve got to be sprinting a marathon to catch up, because they continuously put roadblocks in our way — while at the same time going crazy with AI themselves,” Mr. von Eschenbach said.
4. Vendor evaluation is changing: At-risk contracts and shorter replacement cycles are becoming the standard.
The vendor landscape for health IT is shifting in ways that are changing how health systems structure their partnerships. The traditional two-to-three-year contracting cycle is compressing toward 12 months as technology moves faster than any static agreement can accommodate.
Health systems are increasingly pushing for at-risk models — where vendor payment is tied to whether promised KPIs are delivered — rather than fixed-fee contracts that lock systems into underperforming tools. Peer references are becoming more important, not less; leaders are bypassing vendor-curated lists and reaching out directly to known contacts at reference organizations. And the frontier model assumption — that the most capable AI tool is always the right choice — is giving way to more sophisticated thinking about cost, scope and fit.
“I always ask: Who can I actually talk to that’s doing this? And often you find it works, but it’s not all it’s cracked up to be. We’ve hit a lot of singles and doubles. Anybody coming in saying they can hit a home run every time — I’m highly skeptical,” Mr. von Eschenbach said.
5. Every AI pilot needs an expiration date.
A Deloitte survey cited during Day 3 found that roughly one-third of health system AI initiatives have scaled, one-third remain in pilot phase, and one-third have been paused or abandoned. Panelists were not surprised by the proportion stuck in pilots and identified the absence of a defined decision point as the primary culprit.
Pilots without expiration dates become permanent fixtures that consume bandwidth, cloud storage, and staff attention without delivering value, they said. The discipline of setting a clear success threshold before a pilot launches — and being willing to kill projects that do not reach it — is one of the most important governance capabilities a health system can build. Failing fast is not failure; staying in a non-performing pilot for too long is.
“You should not adopt any platform without setting an expiration date on the pilot from the start. Killing some pilots is OK; it’s better than keeping it going when no one is making a decision for a year or two and it’s not helping anyone,” said Toyosi Olutade, MD, chief medical officer of Rock Island, Ill.-based UnityPoint Health-Quad Cities.
6. AI workforce training is not a technology program, it is a change management program.
Health systems that have made the most progress on AI adoption share a common approach: They treat training as a change management exercise, not a technology rollout. One-size-fits-all AI training fails because it does not connect to the specific problems people are actually trying to solve in their daily work, leaders said.
What succeeds is tailored training anchored to real workflows, delivered to staff who understand what they are supposed to do differently when they finish. Mandatory AI safety and literacy training is becoming a baseline expectation — something that gets embedded in existing training infrastructure, such as annual cybersecurity programs — while more advanced tiers serve citizen developers and technical staff. And fear is a real variable: younger workers in particular are often more concerned about AI’s impact on their roles than their managers expect, leaders said.
“You can over-index on training. If you’re training on a tool that is suboptimal or on a process that is suboptimal, you are just automating the wrong thing. Train on the workflow being optimized — that’s what drives adoption,” said Deb Anderson, CIO of Evanston, Ill.-based Endeavor Health.
7. Critical thinking is the workforce capability most at risk from AI, and health systems are not doing enough to protect it.
As AI takes on more routine analysis and decision support, a concerning pattern is emerging: Over time, staff and clinicians begin to defer to AI outputs rather than interrogating them. Machine bias — the assumption that if the AI said it, it must be right — is appearing in systems that have deployed AI tools for two or more years, accompanied by measurable reductions in critical reasoning, leaders said.
The antidote is training people to be skeptics, building governance that requires human sign-off and audit, and treating AI outputs as a starting point rather than an answer. As liability questions around AI in healthcare remain largely unresolved, the health system whose staff has stopped questioning the output will face the most exposure, leaders said.
“You see an increase in machine bias — ‘the AI told me, so it must be true.’ And in relation to that, you see a reduction in critical reasoning skills. To be the human in the loop means you have to be the chief skeptic about what’s coming to you,” said Jeff Gautney, senior vice president and CIO of Rush University System for Health in Chicago.
8. AI is beginning to be managed as labor, not just as technology.
One of the most forward-looking ideas to emerge from Day 3 was a reframing of how health systems should think about AI: not as a tool or a technology investment, but as a form of labor that must be supervised, evaluated and held accountable.
As agentic AI takes on tasks that were previously performed by humans, the management questions shift from implementation to oversight: who supervises the agent, what happens when it breaks, and who is on call when it misbehaves at 2 a.m.? These are workforce management questions, not IT questions, leaders said. Health systems that are getting ahead of this are beginning to build structures that apply the same management discipline to AI agents that they apply to human employees and recognizing that this requires a different skill set from their leadership teams.
“If you start thinking about these opportunities not just as technology, but as management of labor — that is a very different decision-making set and managerial skill set. I think decision processes are going to change dramatically,” said Brian Hasselfeld, MD, executive medical director of digital health and innovation at Johns Hopkins Medicine in Baltimore.
9. Intelligent health systems are defined by action, not data, and most systems have more of the latter than the former.
A session on designing the intelligent health system landed on a simple definition: Intelligence is the commingling of data, information and action. The first two are achievable; the third is where most organizations stall, leader said. Dashboards surface alerts. Command centers surface throughput gaps. But knowing who should act, on what, and when — and ensuring that the workflow actually exists for them to do so — remains the hardest problem in health system operations.
Successful examples leaders shared from Day 3 included AI-generated discharge summaries that improved handoff to primary care and reduced readmissions; nurse shift summaries that reduced documentation burden at end of shift; and a diagnostic algorithm that identified structural heart disease in a patient who came in for unrelated reasons.
“Intelligence is the commingling of data, information, and action. It’s less about the tools, and more about the outcome and the actions driven by that outcome,” said Chif Umejei, senior vice president and CIO of NewYork-Presbyterian.
10. The organizations with the most AI will not be the winners; the ones that couple AI with human judgment will be.
Day 3 closed with a theme that ran through a number of sessions: AI volume is not the competitive advantage. The health systems pulling ahead are not the ones with the most tools deployed — they are the ones that have been most deliberate about pairing AI capability with human judgment, clear accountability and governance structures that can absorb the pace of change, leaders said.
“Organizations with the most AI are not the winners. It is the organization that couples the human with the AI capability that’s delivering high value. That’s where we win,” Ms. Anderson said.
