Definition
Ratio of AI-predicted no-show probability to actual no-show rate at each probability band, measuring whether overbooking decisions are based on a calibrated forecast.
The Measurement Gap
The Decision That Legacy Metrics Cannot Support
If the model says there is a 40% chance Patient X will no-show, can the clinic rely on that probability for overbooking decisions, or is the model systematically over- or under-predicting no-shows?
What Goes Unmeasured Today
Ambulatory clinics have used overbooking to manage no-show losses for decades. The traditional approach (book 10-15% extra based on historical no-show rates) treats all patients as equally likely to no-show. AI-based scheduling tools promise individualized predictions: this patient has a 40% no-show probability, that one has a 5% probability. Clinics use these probabilities to make targeted overbooking decisions, concentrating extra bookings on high-probability no-show slots.
What clinics almost never measure is whether those individualized probabilities are calibrated. If the model says 40% no-show probability for a group of patients, do 40% of them actually no-show? If only 20% no-show (model overestimates), the clinic overbooks aggressively and creates overcapacity: double-booked patients all show up, waiting rooms overflow, visit times shorten, and patient satisfaction drops. If 60% no-show (model underestimates), the clinic under-books and loses revenue from empty slots.
Legacy metrics (aggregate no-show rate, same-day cancellation rate, patient wait time) measure scheduling outcomes without linking them to prediction accuracy. A clinic reporting rising patient complaints about wait times may not connect the problem to an overconfident no-show prediction model that is driving excessive overbooking. Without band-level calibration measurement, the prediction model operates as a black box influencing thousands of scheduling decisions daily.
The Invisible Failure Mode
Miscalibrated no-show predictions create operational chaos that is attributed to other causes. When the model overestimates no-shows, clinics overbook, patients wait longer, providers run behind schedule, and satisfaction scores decline. The root cause (an overconfident prediction model) is invisible because nobody checks whether 40%-probability patients actually no-show 40% of the time.
Formula
Primary Formula
NPCI(b) = Actual_No_Show_Rate(b) / Predicted_No_Show_Midpoint(b)
Variable Definitions
| Variable | Definition | Unit |
|---|---|---|
| b | Probability band (e.g., 0–20%, 20–40%, 40–60%, 60–80%, 80–100%) | Band range |
| Actual_No_Show_Rate(b) | Patients who no-showed / total patients in band b | Ratio (0–1) |
| Predicted_No_Show_Midpoint(b) | Midpoint of the probability band (e.g., 0.30 for 20–40%) | Ratio (0–1) |
Data Requirements
| Data Element | Source System | Classification |
|---|---|---|
| Predicted no-show probability per appointment | Scheduling AI / EHR scheduling module | NATIVE |
| Appointment outcome (kept, no-show, cancelled, rescheduled) | EHR scheduling module | NATIVE |
| Overbooking decisions linked to predictions | Scheduling system / clinic workflow | INTEGRATION |
Edge Cases and Exclusions
Distinguish between advance cancellations (slot can be refilled) and same-day no-shows (slot wasted). NPCI should focus on same-day no-shows since those drive overbooking decisions. Calibration may vary seasonally (flu season, holidays): report quarterly. Minimum sample: 200 appointments per band per quarter.
Proposed formula, pending validation pilots.