Skip to content

More on Air Taxis

My friend and business partner (in sports scheduling), George Nemhauser, read my post on air taxis and wrote to remind me that Georgia Tech worked with DayJet on the optimization issues that are key to the efficient running of their operation. This led me to an USA Today article by Kevin Maney that I had missed on the subject. The article covers the optimization issues very well:

Tech industry veteran Ed Iacobucci seems an improbable guy to start a new kind of airline. It’s like Donald Trump starting a chain of Laundromats, or Tom Cruise marketing an anti-depression drug.

Pretty jarring, in other words.

But he isn’t really starting an airline, much as eBay didn’t start a flea market. Iacobucci is a one-time IBM tech whiz and founder of software maker Citrix Systems. Over the past four years, he and his team have built a breakthrough computer system for solving highly complex optimization problems.

An optimization problem is like when a mom has to pick up one kid at soccer, one at dance, buy groceries, walk the dog and volunteer at church, and has to figure out the most efficient way to do them all. Now try that for hundreds of moms and hundreds of tasks all at once.

“This is hard stuff,” Iacobucci says. “There’s a lot of new science involved.”

His team is using this system to launch DayJet, the first true on-demand air service. Such a service could not exist without the new computer system. Basically, Iacobucci has started a technology company that will make its money by flying people around.

I love the phrase “a technology company that will make its money by flying people around”. I would like the phrase “an operations research company that will make its money by flying people around” even better, because that is what DayJet really is. Just like Amazon is an operations research company that makes its money by selling books and more, and FedEx is an operations research company …

The article goes on and makes a very clear point about the scale of the optimization, and the need for timely schedules:

If you have a bunch of little jets and a bunch of people in different cities who want a ride, Iacobucci thought, software should be able to figure out the most efficient way to scatter the planes so they can transport the people — while charging enough to make a profit but not nearly as much as a traditional charter plane service.

Good idea, until you start considering all the variables involved. Matching people, cities and aircraft seats is tough enough, but add in crew schedules, maintenance, fuel costs and the uncertainties of weather — plus the need to quote ticket prices before all the variables are in place — and you’ve got a computational mountain no one had yet climbed.

“When I told our team what we wanted to do, they went like this,” Iacobucci says as he makes a cross with his fingers — the way you’d ward off vampires. That’s serious, considering his team includes a couple of former Soviet rocket scientists and the complexity theory department at Georgia Tech, which helped DayJet crack the problem.

As customers put in their requests, the system continually crunches all the departure and arrival requests, plane availability, weather patterns and so on, coming up with a new best answer for schedules and prices every five seconds, always trying — as the DayJet folks say — to get the solution “within 2% of optimality.”

You have to appreciate how remarkable that is. When you’re making everyday, multi-faceted decisions — What should I make for dinner? Should I finish this report or see my kid’s soccer game? — it’s pretty unlikely you ever get within 2% of optimality. I think Donna Reed used to, but I’m sure that’s escaped every other human since.

The DayJet system crunches answers and ranges of probability until a couple of hours before jets would have to take off. “Then the schedule starts to gelatinize,” says Brad Noe, DayJet’s VP of engineering. “And it comes up with a plan.”

This is a great example of academia/business interaction in operations research to come up with businesses that could not exist twenty years ago.