.png)
Duke Health
Healthcare · USA
13% more accurate
Duke Health uses ML to predict surgery duration 13% more accurately, saving overtime and OR capacity
Model trained on 33,000+ surgeries. $79,000 in staffing cost savings in 4 months.
Friction
Duke Health's OR had large variation in duration estimates, causing surgeries to overrun and generating overtime costs and replanning needs.
Breakthrough
Supervised ML trained on historical operation data (procedure type, surgeon, team, patient info) to generate reliable duration estimates, integrated directly into the scheduling workflow.
Impact
13% more accurate than human estimates. $79,000+ in projected staffing savings in 4 months. Applied to 33,000+ surgeries in production.
Unlock the full analysis with breakthrough, impact, what made it smart and its technical approach below!
Problem
Duke Health's OR had large variance in surgery duration estimates, causing frequent overruns, overtime costs, and replanning. Planners relied on experience, a limited and inconsistent resource.
What made it smart
Three different ML models were trained and compared on thousands of surgeries. The best-performing model was integrated directly into the scheduling workflow. Planners receive a predicted duration the moment a surgery is booked.
Technical approach
Supervised machine learning on historical operation data: procedure type, surgeon, team, patient info, and historical duration. Models are retrained as new operation data becomes available. Human validation remains for outliers; the system flags deviations rather than replacing judgment.
Strategic lesson
In operations with high variability, a model that is 13% more accurate than human intuition, consistently and at scale, creates compounding value.
Reflection question
Which planning decisions in your organisation still rely on gut feeling or experience? What would change if you had a model that was consistently 13% more accurate?