No-show prediction is one of the most widely deployed models in ambulatory care, and one of the least scrutinized, because its output is a scheduling decision rather than a clinical one. An internal audit at a multi-site practice group suggests that framing has been a mistake.

The model performed as advertised, identifying appointments at elevated risk of no-show with useful accuracy. The audit examined which features carried the weight and found distance to clinic, prior no-show history, and appointment lead time dominating — all of which correlate strongly with transportation access and hourly employment.

The operational policy layered on top was overbooking: high-risk slots were double-booked. When both patients arrived, one waited. The audit found that wait time above ninety minutes was concentrated in the same population the model had flagged, which is the predictable consequence of the design and had not previously been measured.

The practice group has replaced overbooking with targeted outreach — reminder calls, transportation assistance, and offered telehealth conversion — for the same flagged population. Early data shows no-show rates falling and the wait-time disparity closing. The model did not change. What was done with it did.