Presentation is loading. Please wait.

Presentation is loading. Please wait.

1 O&D Forecasting Issues, Challenges, and Forecasting Results John D. Salch PROS Revenue Management, Inc.

Similar presentations


Presentation on theme: "1 O&D Forecasting Issues, Challenges, and Forecasting Results John D. Salch PROS Revenue Management, Inc."— Presentation transcript:

1 1 O&D Forecasting Issues, Challenges, and Forecasting Results John D. Salch PROS Revenue Management, Inc. jsalch@prosrm.com

2 2 Forecasting Issues / Challenges  data  processing time  modeling  dynamics

3 3 Data …There Is More Than We Know What to Do With

4 4 Data Collection  Data Sources (Assume 1000 flights per day) u PNR (Touched and Flown) ~ 250,000 per day u Flight level inventory ~ 150,000 per day u Schedule ~ 20,000 per day u Agent, Customer, etc… ~ ? (your mileage may vary…)

5 5 Data To Collect: Some Examples  PNR Record Locator  Passenger Name  Creation Time  Creation Date  Creation DOW  Holiday  Special Events  Airline Code(s)  Origin Airport  Origin City  Origin Country  Origin Continent  Destination Airport  Destination City  Destination Country  Destination Continent  Path Airport  Path City  Departure Date(s) all legs  Departure Time(s) all legs  Point of Sale City  Point of Sale Country  Point of Sale Continent  Booking Office  Group Identifier  Passenger Type (Freq. Flier Type?)  Frequent Flier Number  Fare Classes all legs  Number of Passengers  Number Protected  No Show Identifier  No Show Reason  Go Show Identifier  Go Show Booking Time before Departure  Connection from Airline  Connection to Airline  Original Point of Departure  Final Destination  Cancellation Identifier  Cancellation Date  Cancellation Time  Cancellation Reason  Flight Numbers all legs  Confirmation Codes all legs  Fare (Base, Airport Chg, Tax)  Ticketing Information  Currency (Type/Exchange Rate)  Fare Basis Code  Special Service  Passenger Address  OAL Booked By  OAL segment(s)  Tour Segment  Hotel Segment  Car Segment  Group Name  Number of Passengers in PNR  Ticket Type  Denied Boardings Code  Form Of Payment Info  Agent Iata#  Tel #  Other Supp. Info Messages  Protected History (all legs bkd)  Received From (PNR modifying person)  Arrival Times all legs  OAL segment

6 6 Data Challenges  Rich source of data u It will take many years to find all of the gems  Large volumes of data u Processing time is the binding constraint  Cleaning / Massaging u Lots of cleaning required

7 7 Forecast Modeling It Must Be Fast, Fast, Fast….

8 8 Forecast Updating  Unconstrain Actuals  Update Models

9 9 Unconstraining  Methods for adjustment u Projection Methods u Iterative Methods  Inputs u Constraint Probability u Bookings / Cancels / Waitlist

10 10 Forecast Modeling  Bayesian forecasting paradigm  Correlation adjustments  Seasonality Adjustments  Hierarchical Correlation  Component Relationship

11 11 Bayesian Forecasting  Simple updating  Minimal data history required u Uses all history, but minimize database  Dynamic to changing data u exponential smoothing

12 12 Bayesian Forecasting  components: u reservations (arrivals model) u cancellations (rate model) u go-shows u no-shows u booking curve  Each component poses new challenges!

13 13 Correlation Adjustment  remove model assumptions of independence across time slices u adjust based on correlation model u early surge in bookings/cancels may result in lower or higher bookings later in cycle u significant reduction in errors

14 14 Seasonality Adjustments  Model cyclical patterns u day of week patterns u monthly patterns u year over year patterns u significant reduction in errors

15 15 Hierarchical Adjustments  remove model assumptions of independence between entities u relate entities through hierarchy u reduce “small numbers” problem u high demand in one itinerary may imply high/low demand in another (spill) u significant reduction in errors

16 16 Component Relationship  “Blend”: u blend different models to form “out” passenger forecasts, demand to come u relate forecasts, e.g. cancels and no-shows

17 17 Accuracy: “The Forest and the Trees”  Small numbers accurate, but... u aggregations need to be accurate, as well u Feedback mechanism u proper model tuning u bad aggregate forecasts can bias bid prices

18 18 Holidays / Special Events  Accounted for in models  Discount from “non-holiday” forecasts  Incorporate user knowledge

19 19 Dynamics Everything Is Always Changing…

20 20 Dynamics  Schedule changes u Reduce impact of frequent changes in the flight network u Maintain “relevant” history u Create a “schedule-free” network  Accounting for new markets u sponsorship

21 21 Hard Work Pays off... Forecasting Results

22 22

23 23

24 24

25 25

26 26

27 27 A Forecast "Tonight's forecast: dark. Continuing dark throughout the night and turning to widely scattered light in the morning." - George CarlinGeorge Carlin


Download ppt "1 O&D Forecasting Issues, Challenges, and Forecasting Results John D. Salch PROS Revenue Management, Inc."

Similar presentations


Ads by Google