SQL, data modeling, data quality, EDA, visualization · insight-0074
Flag capacity-constrained observations before treating sales as demand
- Evidence type
- empirical finding
- Difficulty
- medium
- Roles
- Data scientist, Data analyst
- Industries
- aviation, retail, e-commerce
Summary: situation, action, reason, boundary
Situation
A forecast or price analysis uses passengers or units sold as the target, and some flights sold out or some stores ran out of stock.
Action
Mark observations where sales hit capacity or where booking classes closed. Either unconstrain them with a stated method (for example a censored-data or EM approach), model the censoring, or restrict claims to "sales" and state the limitation on every chart and forecast.
Reason
Rejected requests leave no trace in sales data; treating censored values as complete biases means and elasticities downward.
Boundary
Airline and hotel bookings, retail with stockouts, ticketing and any capacity-limited service.
Symptom or risk
On constrained days, sales stop at capacity, so the target understates demand, peak demand looks capped, and models learn that high-demand periods are flat.
Exceptions
Products that never approach capacity; verify with the constrained share rather than assuming.
Trade-offs
Unconstraining methods add assumptions about turned-away demand that cannot be fully validated.
Pitfalls
Dropping sold-out days from training, which removes exactly the high-demand periods.
Example
Synthetic airline simulator: passengers are capped by capacity, so peak-season load factors near 1 hide the demand above it.
Lessons and concepts
Claims and their sources
Historical airline booking data can understate true demand because requests are rejected by capacity or booking limits; the authors report revenue gains of roughly 2-12% from better unconstraining, based on booking data from one major US airline (a 2002 study).