I have an entry over on the INFORMS blog regarding overbooking of hotels. Here it is, though I recommend following the INFORMS blog for the next few days:
Fellow blogger Guillaume Roels wrote that the hotel he reserved overbooked, so he has been exiled to a remote location and he bemoaned the lack of customer service in this transaction. Something similar was obviously going on in my hotel, the Hilton (the main hotel for the conference). Throughout the checkin yesterday, the desk clerks were looking for volunteers to be exiled, offering various incentives (”Free transportation! A drink vouncher! Big, big discounts, just for you!”) for people to move. They weren’t getting any takers while I was there, so I fear the late check-ins were similarly sent off to the boondocks.
I bet the hotels got into a mess because they misestimated the number of people who showed up (or overestimated the “melt”: people who canceled in the final week or two). If they simply took an average “no show” or “cancel in the last week” rate, I bet conference participants do so at a much lower rate. After all, the vast majority of us have preregistered for the conference, so late cancellation means forfeiting some or all of the conference registration fee. We have great incentives to figure out early whether we are going to be here or not. And, perhaps people in OR or other analytic fields tend to not cancel or cancel earlier due to the organized, steel-trap-like minds we all have! We know what we are doing, so we don’t cancel in the last week.
Of course, whether or not that is true is an empirical question, and one that can be best answered by data mining methods. Over the course of drinks last night, a senior researcher for a large business analytics firm pointed out the disconnect we have in our field between data mining and optimization. Often (though not always), these are seem as two phases of the “operations research process”. Instead, there is a need for much better integration between these approaches. Data mining should be constantly in use predicting cancellations and melt, driving the revenue management optimization approaches.
For those who were bumped by the hotels last night, you have my sympathies. Perhaps during your rides into the conference, you can plan how to integrate data mining and revenue management better in order to let hotels avoid these issues in the future.
It seems to me that INFORMS could provide historical data on canceled registrations and no-show presenters, which would let the conference hotels better estimate their melt rates.
Although it’s not an overbooking issue, there’s another data (mis)management aspect. Apparently neither INFORMS nor the convention centers convey to proximate restaurants the number of expected attendees. We’re typically at a convention center, convention centers are typically downtown, and downtown restaurants often are closed on Sundays (since most offices and many businesses are). The remaining eateries apparently are not warned to ramp up their staffs. So we repeatedly have a logjam for the Sunday lunch hour.
The comment that “Instead, there is a need for much better integration between these approaches” really stroke a chord with me. My question is: should companies create two jobs, one for statisticians in one department and the other for operations researcher in another department and hope that these two departments will work closely together or they should look for guys who have knowledge on both fields.