// The Productivity Paradox

What if healthcare AI makes your best people look less productive?

Fewer cases per hour. More room to get the difficult ones right.

Sample monthly PA cohort · same 25,000 requests
Before AI25,000 human-reviewed
After AI10,000 automated · 15,000 human-reviewed
Illustrative cohort: 25,000 requests. AI automates 10,000 routine requests. Reviewers spend 27 minutes on each of 15,000 complex requests. Final reported reviews per hour: 2.22; reviewer hours: 6,750; cost per request: $39.24. Each cell represents 100 requests.
BaselineThe same 25,000 requests, all human-reviewed
AI takes routine work10,000 eligible requests automated
Reviewers go deeper+2 minutes on each complex case
The resultThe score drops. Quality has room. Cost falls.
Routine review · 11 minComplex review · 25 minComplex review · 27 minGot right first timeAutomated by AI
Each cell = 100 requests
Reviews per reviewer hour2.22-28.1% · the reported score
Reviewer hours6,750+500 hr reinvested in complex reviews
Cost per request$39.2412.8% lower · $144,000 a month

Fewer reviews per hour. Fewer avoidable returns for clarification or correction. Lower cost per request.

// What the numbers say

The Score Drops. The Work Gets Better.

Three views of the same month: what the productivity score reports, where reviewer time actually goes, and how quality can improve when the difficult cases get the time they need.

The score drops

Completed reviews per hands-on reviewer hour-28.1%The reported score
3.092.22per hour
Reviewer time per human-reviewed request
19.427minutes

Before: 10,000 at 11 min + 15,000 at 25 min.

Where the time goes

Time per complex review
2527minutes

Deliberate choice: +2 min (+8%). 500 hours reinvested in review.

Reviewer hours
Before AI8,083.3 hr
Before AI: 8,083.3 reviewer hours, including 1,833.3 routine and 6,250 complex review hours.
After AI6,750 hr
After AI: 6,750 reviewer hours, including 6,250 original complex review hours and 500 reinvested hours.
1,833.3 hr routine review removed6,250 hr original complex review500 hr reinvested in reviews

How quality can improve

Use the extra 2 minutes

Resolve evidence /
criteria conflicts.

Record rationale
and open issues.

The quality aim

Fewer avoidable returns
for clarification
or correction.

Automation frees time. Leadership decides how reviewers use it.

500 hours reinvested in complex reviews.Staff benefits: less rework, less pressure, greater job satisfaction.
Estimated total PA processing cost
$45.00$39.24per request
12.8% lower$1.73M per year

Before: $485,000 review + $640,000 other stages = $1,125,000
After: $405,000 review + $20,000 AI + $556,000 other stages = $981,000

Cohort assumptions
25,000requests. Same incoming volume and mix
$60per reviewer hour. Loaded cost per productive hour
$2per automated request. AI covers three stages
PA processing cost by stageSeparate, non-overlapping functions. Quality gains are not monetized.
StageBeforeAfter
Review + automationCost before AI: $485,000. Cost after AI: $425,000. Each cell represents $10,000.$485,000$425,000
IntakeCost before AI: $150,000. Cost after AI: $102,000. Each cell represents $10,000.$150,000$102,000
CoordinationCost before AI: $120,000. Cost after AI: $84,000. Each cell represents $10,000.$120,000$84,000
Physician escalationCost before AI: $250,000. Cost after AI: $250,000. Each cell represents $10,000.$250,000$250,000
Systems and oversightCost before AI: $120,000. Cost after AI: $120,000. Each cell represents $10,000.$120,000$120,000
Cost after AIRemoved
Each cell = $10,000
For 10,000 eligible routine requests, AI covers intake, coordination and review. Human intake: 10 to 2 min; coordination: 8 to 2 min, both at $36/hr. Complex support unchanged.Physician: 10% × 40 min at $150/hr. Systems/oversight: $4.80/request; AI fee separate. Eligible routine path; additional exceptions and rework not included.Incremental implementation, integration, policy configuration and workflow change excluded.
Context: Health Affairs Scholar (2023), $40–$50 / private-payer submission. Sample inputs are assumptions.Source: PMC10986268