Airline revenue model
Occupancy and revenue per flight, inferred from the one thing an airline publishes: price.
The situation
An airline's fares are the output of a revenue-management system that raises price as seats sell. The price curve quietly leaks occupancy. The client needed flight-level load factor and revenue across an entire network, none of which is published.
How it works
Observe
distributed scrapers behind rotating proxies capture fare, booking class, days to departure and observation time across every route and date, about 500,000 readings a day, each stored as a new timestamped row so every flight's price path can be reconstructed.
Decode the ladder
the discrete price buckets and the rules that trigger a jump. A jump fires when a bucket sells out, so the sequence of jumps maps back to seats gone. Calibrated against known aircraft capacity per route, that yields load factor inside a tight band.
Value it
estimated passengers times the fares actually paid across the booking curve gives revenue per flight, per route, then network-wide. A gradient-boosted model produces the seats-sold estimate, validated against the carrier's published load-factor figures.
Outcome
500,000 observations a day · full route network · validated against published load factors.
Hardest part
The whole model rests on inferring something the airline never publishes, seats sold, from the one thing it does publish, price. That only holds if the fare ladder is reconstructed correctly, and airlines move buckets, run promotions and price differently by route and season, so a rule that fit last month quietly breaks. The hard part was separating a real "seats sold, price stepped up" signal from noise, a promo, a schedule change, a competitor move, across thousands of flights, well enough that the estimate held against the few real numbers we could check.
See how this maps to your business.
walk through this build · no obligation