Sophisticated unconstrainer revenue performance in a large airline network

In this article, we use a large network (572 markets) in the sophisticated Passenger Origin‐Destination Simulator (PODS) to examine the impact on expected revenues when applying three different methods of unconstraining – Expectation Maximization (EM), Projection Detruncation and Booking Curve in one of two different approaches (‘all closed’ versus ‘zeroes only’ (ZO), or ‘multiple estimates’ (ME) versus ‘single estimate’ (SE)). Owing to the extremely competitive nature of this PODS network (four airlines competing for customers) and its allowance for customer choice, we are able to assess all the implications of both implementation variations, including the impact of spill, upgrades and recapture. We find that for the first implementation decision, the optimization engine that is currently being used dictates quite different revenue results when considering changing implementation techniques. In one case (under leg optimization), moving away from the ZO approach will lead to 0.5–1.2 per cent revenue gains; while in the other case (under network optimization), it leads to slight revenue losses. For the second implementation decision, we find that generally it is safer to use ME. There is one exception – for EM unconstraining when the carrier of interest is using leg optimization, it is best to switch to the SE approach, as it generates 0.6–1.26 per cent revenue gains.