Presentation is loading. Please wait.

Presentation is loading. Please wait.

O & d Forecasting for O & D Control Arjan Westerhof Decision Support.

Similar presentations


Presentation on theme: "O & d Forecasting for O & D Control Arjan Westerhof Decision Support."— Presentation transcript:

1 o & d Forecasting for O & D Control Arjan Westerhof Decision Support

2 May 5, 2015 AGIFORS Yield Management June Outline Introduction: O&D control and forecasting Why O&D’s are usually o&d’s 3 alternatives for handling o&d’s in the forecast Conclusions and discussion

3 May 5, 2015 AGIFORS Yield Management June O&D Control KLM implemented O&D revenue management in 2000 (first sub-networks) / 2001 (entire network) Systems based on –O&D demand forecasting –O&D fare forecasting –network optimization Organization based on O&D / Point of Sale

4 May 5, 2015 AGIFORS Yield Management June Bottom Up vs Top Down KLM uses bottom up demand forecasting This seems more powerful to capture the various cultural differences in KLM’s home market than top down forecasting

5 May 5, 2015 AGIFORS Yield Management June From Agifors YM 2002

6 May 5, 2015 AGIFORS Yield Management June KLM versus LH LH KL Are we almost the same ?

7 May 5, 2015 AGIFORS Yield Management June Every customer is different! O&D is not the only dimension … … other dimensions that might influence booking behavior and passenger revenue: –Customer or Point of sale (agent / country / corporate account / …) –Booking class / ticket restrictions –Departure Day of Week –Departure season –Special events –Etc. Define ‘product’ at a more detailed level

8 May 5, 2015 AGIFORS Yield Management June The product curve We are different!? %Pax %Products LH Products (?) KL Products %O&D’s %Pax KL O&D’s

9 May 5, 2015 AGIFORS Yield Management June Why is KLM different from LH? Though Amsterdam is a great place to visit ±70% of our passengers use Amsterdam only for connecting to other destinations

10 May 5, 2015 AGIFORS Yield Management June ‘Medium’ ‘Small’ ‘BIG’ Introducing... The o&d World %products %Pax — KL Products

11 May 5, 2015 AGIFORS Yield Management June O&D or o&d? > 70% of the products are ‘exotic’ o&d’s (not sold regularly) If forecasts are created for these o&d’s: –The quality of these forecasts can hardly be measured –> 70% of computation time will be involving ‘meaningless’ numbers –The forecast will be confusing to the users

12 May 5, 2015 AGIFORS Yield Management June Solutions for o&d Forecasting Do nothing special Forecast aggregation Forecast elimination

13 May 5, 2015 AGIFORS Yield Management June Do nothing special Network optimization will ‘aggregate’ the o&d’s to the leg-level when determining bid prices and buckets Advantages: –All detailed information is available in the forecast –Acceptance/rejection in the RES system aligned with forecast/optimization system Disadvantages: –Forecast quality can not really be measured –Much data with little information (user/computing)

14 May 5, 2015 AGIFORS Yield Management June Solutions for o&d Forecasting  Do nothing special Forecast aggregation Forecast elimination

15 May 5, 2015 AGIFORS Yield Management June Forecast aggregation Drop one or more of the dimensions in the product definition for products with insufficient volume For example: –Drop O&D dimension by splitting o&d’s in legs –Drop point of sale dimension –Drop booking class dimension or aggregate to cabin level

16 May 5, 2015 AGIFORS Yield Management June Forecast Aggregation Advantages –Quality of aggregated forecasts can be measured –Helps to focus on important flows Disadvantages –Bookings are evaluated in the RES system with different values than used in optimization –Unconstraining uses different revenue values than the ones used during passenger acceptance –Products with different booking behavior might be aggregated

17 May 5, 2015 AGIFORS Yield Management June Forecast quality? Hard to measure: Many O&D’s/Flights are constrained during some time in the booking cycle There are almost no stable reference periods anymore (Sep 11 / War / SARS / …) Evaluating forecasts on the leg level might bias the evaluation to the benefit of the aggregated forecasts

18 May 5, 2015 AGIFORS Yield Management June Forecast quality Leg/cabin level, open flights on two lines Note: even on an open flight some products may not be for sale due to constraints on other flights Not much difference in forecast Which one is better? TIME Rem. demand

19 May 5, 2015 AGIFORS Yield Management June TIM E Average bid price Aggregation vs. Do Nothing Some differences but not too big with 50% fewer forecast products

20 May 5, 2015 AGIFORS Yield Management June Solutions for o&d Forecasting  Do nothing special  Forecast aggregation Forecast elimination

21 May 5, 2015 AGIFORS Yield Management June Forecast elimination Throw out the o&d’s Re-map these o&d’s to products with significant demand (O&D’s) Experimental results indicate that this does not result in a good forecast

22 May 5, 2015 AGIFORS Yield Management June Conclusion Most of the products that are sold are ‘exotic’, there are much more o&d’s than O&D’s If these exotic products are being forecast, they are polluting the system Aggregation solves the small number problem but the quality of the forecast is not always better As a result, the choice between aggregating or not aggregating seems mainly a matter of personal preference (?)

23 May 5, 2015 AGIFORS Yield Management June Questions ?


Download ppt "O & d Forecasting for O & D Control Arjan Westerhof Decision Support."

Similar presentations


Ads by Google