Healthcare's Moneyball Moment
Billy Beane didn't help the Oakland A's win by spending more. He won by measuring differently. That mindset belongs in healthcare right now.
We still treat appointment volume, panel size, and throughput as success. Those are activity metrics, not outcomes. They don't tell you if the right patients are getting the right care at the right time-or if scarce specialty slots are going to the cases that matter most.
Why legacy metrics miss
When high-risk patients are overlooked, the costs show up later: emergency admits, avoidable complications, leakage to outside providers, and missed chances to intervene early. Broad, guideline-only outreach looks efficient on a dashboard, but it pushes resources toward patients who are easiest to reach, not those with the highest upside.
The result: low yield, wasted effort, and specialty clinics clogged with low-acuity visits that don't move outcomes or margin.
The sabermetrics of patient identification
AI flips patient identification from a passive, administrative task into a strategic function. It scans millions of data points already in the EHR-labs, vitals, demographics, medications, and histories-to surface patients who warrant immediate clinical review.
This isn't about replacing clinicians. It's about precision at scale, integrated into existing workflows so teams can act without disruption.
What AI-enabled population analysis looks like
- Combines labs, vitals, demographics, medications, and histories into risk signals.
- Surfaces patient cohorts with high likelihood of benefit from follow-up now, not months later.
- Routes prioritized lists into current EHR work queues and care navigation processes.
- Tracks yield and closes the loop so models keep improving and trust builds over time.
The clinical payoff
Organizations using advanced analytics to proactively surface patients are seeing earlier diagnosis across multiple conditions. The value isn't changing clinical practice-it's finding the right patients sooner, before care becomes invasive, costly, and risky.
- Colorectal cancer: Risk-stratified outreach delivers substantially higher diagnostic yield than untargeted screening campaigns (see research from NEJM Catalyst).
- Liver disease (MASLD): Earlier identification reduces progression to cirrhosis and helps avoid high-cost complications.
- Arrhythmias: Many atrial fibrillation cases go undiagnosed; AI-supported identification enables monitoring and timely cardiology referral before stroke or heart failure risks climb (see studies via the JAMA Network).
- Type 1 diabetes: Autoimmune diabetes is frequently misclassified; AI-enabled EHR analysis can flag high-risk patients who would be missed under standard protocols.
Across these areas, precision and earlier identification change the trajectory. Delay compounds clinical risk and financial exposure.
The financial multiplier
For CFOs, the math is straightforward. Earlier identification expands downstream revenue, cuts waste, and stabilizes high-value specialty lines-without adding headcount or disrupting schedules.
Example: In a modeled cohort of 100,000 adults ages 45-75, roughly one-third are overdue for CRC screening. By focusing outreach on the specific ~3% surfaced through AI analysis, a health system can find around five times as many cancers and generate an estimated $760,000 in incremental revenue over four years-while paying for fewer total screenings. Similar effects show up in liver disease (avoiding decompensation), cardiology (preventing stroke and HF admissions), and endocrinology (reducing emergency escalations).
- Higher diagnostic yield per outreach and per visit
- Fewer low-acuity specialty appointments
- Stronger contribution margins and in-network retention
- Improved performance in both value-based and fee-for-service contracts
How to operationalize this in 90 days
- Pick 1-2 high-yield use cases: CRC screening gaps, MASLD risk, suspected AF, or diabetes misclassification. Align with specialty capacity.
- Assemble the data: Pull recent labs, vitals, demographics, medications, problem lists, and encounter history. Define inclusion/exclusion rules and privacy controls.
- Choose models with explainability: Clinicians should see why a patient was flagged (e.g., ALT trends, BMI, A1C variability, age, prior EKGs).
- Integrate into the EHR: Deliver prioritized worklists to existing queues for care managers, navigators, and specialty schedulers.
- Stand up closed-loop workflows: Outreach → schedule → diagnosis/staging → treatment start → outcome tracking. Measure yield at each step.
- Track the right metrics: Diagnostic yield, days-to-diagnosis, no-show rates, specialty slot mix (high vs. low acuity), leakage, and contribution margin.
- Build trust and governance: Clinical review committees, bias and equity checks, periodic model recalibration, and clear escalation criteria.
- Start small, scale fast: Prove ROI in one service line, then roll to the next with shared playbooks and templates.
If you need a practical primer on tactics and stack choices, explore resources on AI for Healthcare.
The Moneyball advantage
Baseball changed when teams realized they were measuring the wrong things. Healthcare is at that same point. Full schedules and broad outreach look productive, but they're a mirage.
Real value comes from using data to find the patients who matter most-those whose care drives stronger outcomes and better margins. Every surfaced patient is a clinical win captured and a financial leak sealed.
Volume is a vanity metric. The differentiator is using data to make every clinical encounter count.
Your membership also unlocks:
