Everyone is obsessed with Healthcare AI automating 100% of today’s workload.
The real breakthrough is eliminating work that shouldn’t exist.
My work in AI-enabled inpatient utilization management has led me to question: Are we transforming the review process—or just building a faster version of the status quo?
Many admissions present with undeniable clinical evidence. Yet legacy workflows require highly skilled utilization review (UR) nurses to navigate extensive guideline criteria to justify approvals—even when the clinical picture is crystal clear.
When we bring AI into the mix, we shouldn’t ask, “How do we replicate this entire process?”
We should ask: “Why does every case need the same depth of review?”
We risk confusing replication of an old process with actual transformation.
What if we applied the Pareto principle to inpatient UM?
👉 Automate the obvious. Identify straightforward cases where medical necessity can be established with high precision and reliability.
👉 Preserve human review for complexity. Free highly skilled UR nurses and payer medical directors to focus on borderline cases, conflicting evidence, and complex exceptions—where their clinical judgment adds the greatest value.
👉 Build trust through evidence. Start small. Independently audit AI decisions, measure accuracy and downstream outcomes, and expand automation only when evidence supports it.
The goal isn’t to reproduce every step of the legacy workflow. It’s to recognize when criteria-by-criteria analysis adds value—and when it simply adds work without improving the decision.
Why is this so difficult?
1️⃣ The Replication Trap: We reproduce existing workflows instead of questioning whether they still make sense.
2️⃣ The Trust Gap: Organizations hesitate to trust AI, even in narrowly defined cohorts where performance can be independently validated.
3️⃣ Misaligned Incentives: Operating and financial models may not reward eliminating unnecessary reviews or converting savings into capacity.
The opportunity isn’t to eliminate clinical oversight. It’s to use it where it matters most.
The breakthrough isn’t automating every decision. It’s recognizing which decisions no longer need the same amount of work.
Are we transforming utilization management—or simply making the legacy workflow faster? Is value stream mapping (VSM) dead in healthcare?
What happened to all the value stream mappers? Has it become a lost art the moment AI entered healthcare?