The EPIC AI Wave – What CIOs Need to Know Now

Epic just announced enough AI to transform healthcare—or destroy your budget. The difference comes down to whether you chase the shiny object or build the scaffolding.

The AI Firehose from UGM

Epic dropped its biggest set of AI announcements yet at UGM, Here's a list, detailed coverage of each feature can be found pretty much anywhere, I'm not a reporter, I'm an analyst.

• Art – their ambient scribe competing directly with your Abridge, Ambience and Nuance investments
• Emmie – a patient copilot living in MyChart
• Penny – revenue cycle AI promising to handle coding and denials
• CoMET – a foundation model trained on billions of medical events from Cosmos
• Plus AI features tucked into Rover, Synopsis, and inpatient workflows

On stage it looked polished. In the real world, CIOs are asking the only question that matters: What do we actually deploy, what do we pilot, and what do we ignore?

The Epic AI Adoption Matrix

After many discussions with practicing CIO's who have navigated Epic's "revolutionary" feature roll outs over the years, here's how to separate leverage from landmines.

Quick Wins (Deploy within 6 months)

Penny for specific denial types – Start with missing documentation and routine prior auth denials. The math is simple: if you're spending $65K per FTE to work denials and Penny can handle 70% of volume, ROI is clear. Just don't expect it to handle complex medical necessity appeals on day one.

Hey Epic for simple queries – Let staff ask "what's the bed census?" or "show me today's OR schedule" without breaking anything. It's Alexa for Epic—useful but not transformative.

CoMET for retrospective analysis – Quality reviews, cohort studies, population health analytics. No clinical risk when you're analyzing last quarter's readmissions. This is where CoMET shines today. If you're not participating with Cosmos that is the first step, the value proposition is too great to sit on the sidelines. You can't win on your own.

Prove It First (Pilot carefully, measure everything)

Art (Epic's ambient scribe) – Here's the uncomfortable truth: Epic rarely nails v1.0. You've already invested in Abridge, Ambience, or Microsoft. They have 2-3 years of battle scars and clinician trust. Let other health systems debug Art while you optimize what's working. Epic's demo was Microsoft/DAX under the hood anyway—why switch to Epic's wrapper?

CoMET for clinical decision support – Start with low-risk nudges like preventive care reminders or vaccination alerts. But semantic drift and training bias make it dangerous for critical diagnoses. Remember: Epic trained CoMET on aggregated data, not your specific population. I'll bet that is coming.

Penny beyond basic denials – Some workflows will print money, others will create more work than they save. Pilot by service line and benchmark every claim type.

Strategic Patience (Watch and wait 12-18 months)

Emmie for complex patient interactions – An AI chatting with your patients about symptoms? The liability framework doesn't exist yet. The potential for hallucinated medical advice is real. Let academic medical centers with risk appetite go first.

Art as your primary scribe strategy – Keep your proven vendors until Epic demonstrates equivalent quality, clinician satisfaction, and specialty coverage. The switching cost isn't just money—it's clinician trust.

Payer Platform integrations – Epic's payer ambitions are struggling. Current adoption is minimal. Let someone else discover why payers aren't biting.

Quicksand (Just don't)

• Ripping out working solutions for Epic's unproven alternatives
• Wholesale workflow redesign based on demo promises
• Betting on capabilities Epic is "about to release"
• Forcing clinicians to unlearn muscle memory without proven ROI

Reality Check: It's Epic's first-ish swing at AI-native features. Your current ambient vendors have thousands of hours of edge cases handled. Switching to Epic's v1.0 isn't innovation—it's volunteering your clinicians as QA testers. And when it fails at 2 AM, the board won't call Judy—they'll call you.

The Epic AI Reality Checklist

Before you commit to any Epic AI rollout, answer these questions honestly:

Value & ROI

• Can you quantify the problem in dollars or hours? "Efficiency" isn't a business case
• Will you see hard ROI in 12 months, or is this another "strategic investment" that never pays off?

Decision Rights & Ownership

• Who owns success AND failure—you, clinical leadership, or the dreaded "AI committee"? (Committees don't own failures)
• When AI makes a controversial recommendation, who stands behind it publicly?

Data Readiness

• Is your data good enough to trust lives to it, or just good enough for dashboards?
• Do you have the analyst bench strength to monitor model drift—or will patient harm be your first warning?

Organizational Readiness

• Name the clinical champion who'll stake their reputation on this. No name = no go
• How long did your last major workflow change really take to achieve 80% adoption?
• Do you have change management capital to spend, or are you already overdrawn from your last EHR upgrade?

The Bottom Line: If you can't check all three boxes, you're not ready to be an Epic AI early adopter—you're volunteering to be a cautionary tale.

Governance That Actually Governs

The organizations successfully piloting Epic AI share three characteristics:

1. They have a "stop button" process – Clear escalation when AI goes sideways, including who can pull the plug at 2 AM without committee approval
2. They measure actual vs. promised performance – Not Epic's benchmarks, but their own.
3. They protect their optionality – Parallel running proven solutions while piloting Epic's alternatives. The cost of redundancy is nothing compared to the cost of failure

The Uncomfortable Truth About Epic AI

Epic's playing catch-up. They watched Nuance, Abridge, and others prove the ambient documentation market. They saw health systems achieve real ROI with third-party AI. Now they're using their platform leverage to offer "integrated" alternatives.

That's not inherently bad—integration has value. But don't confuse integration with innovation. Epic's AI is Epic-data-friendly, Epic-workflow-optimized, and Epic-value-generating. Whether it's clinician-friendly or patient-outcome-improving remains to be proven. Epic has a great track record over time, so the trick is choosing the right time.

Your Job in the AI Wave

Your role isn't to implement everything Epic announced. Your role is to:

• Identify genuine quick wins that solve real problems with minimal risk
• Design careful pilots with clear success metrics and exit strategies
• Protect proven solutions from platform FOMO

The winners in Epic's AI wave won't be the first adopters—they'll be the smart adopters who can separate transformation from transaction.

Next up in Part 2: Beyond AI – Ecosystem, Expansion, and the Payer Play. While everyone obsessed over Epic's AI demos, they missed the three strategic moves that will reshape healthcare's power dynamics. Epic isn't just adding features—they're redrawing the map.

Bill Russell is a former CIO for a 16-hospital system and founder of This Week Health and the 229 Project.