Ant farmers have a goal: keep the ant colony alive. The ant farmer’s job is to keep the ant colony thriving in the face of predators, disease, poison and human beings stomping them to death. But despite the best efforts of their enemies, the ants usually survive because the ant farmers designed a robust colony, one that could survive stomping and poisoning and predators and battles with other ant colonies.
Huh? What do ant farmers have to do with airplanes?
More than five years ago, when Ed and Nancy Iacobucci’s DayJet concept began gelling into a plan to deploy a huge fleet of very light jet Eclipse 500s to serve business travel markets in the southeast U.S., a chance meeting brought two “ant farmers” to the Iacobuccis’ attention.
DayJet board member Michael Brown, formerly Microsoft’s CFO and then Nasdaq chairman, one day decided to drop in on Bruce Sawhill and Jim Herriot. Sawhill, a particle physicist, and Herriot, a computer scientist and complexity scientist, had spent a lot of time designing models of complex systems such as the Nasdaq stock market, Southwest Airlines and huge product supply chains, among other things.
Their modeling efforts were focused on learning how to predict the behavior of complex systems, in a way that would allow them to test disruptions to the systems without disrupting the actual system. In other words, Sawhill and Herriot threw virtual obstacles at simulated ant farms and observed the results. “What if you made a change in the stock market that destabilized it?” Sawhill asked. “The losses would be enormous, so it’s much better to do the experiment in software than in a real market.”
It didn’t take Ed Iacobucci long to figure out that complexity science might be a useful tool to apply to DayJet. “The ant farmers are kind of a secret weapon that gives us a crystal ball,” he explained. “It’s a fantastic tool that allows us to understand our business before it’s there.”
Sim City on Steroids
When Sawhill and Herriot began working full-time for DayJet about five years ago, they started looking for data to help flesh out the virtual world they were building to model the behavior of 50 million citizens living in 11 states in the Southeast. They liken the world they designed to a digital terrarium, where the owner can watch how bugs and plants interact inside the closed environment.
The DayJet terrarium is a simulation of the world inhabited by the business travelers in the Southeast U.S. Each day these travelers make 500,000 business trips of more than 100 miles.
Each business-trip-making traveler is simulated, too, as part of what the ant farmers call an agent-based model. As Sawhill explained, “You build simulations of the individuals such that they react to external stimuli and you set them loose and see what happens.”
The concept of virtual people acting autonomously will be familiar to anyone who has seen the computer game called The Sims. Sims are virtual humans living in a virtual world, but they are programmed to act as individuals. While the person at the keyboard can nudge the virtual people in the game in certain directions by supplying resources (food, housing, jobs and so on), the Sims make their own decisions and sometimes surprise the game players with unique behaviors.
The same is true of DayJet’s agent-based models and the world they inhabit. “The possibilities of interaction are so large that you can’t have foreseen them all,” Sawhill explained. “We feel we understand how individuals make travel decisions. We don’t understand how an entire economy of people making travel decisions works, but we can build it up from small pieces that we do understand and then set it loose and see what happens. And that’s the Sim City on steroids that we built.”
The purpose of the simulated DayJet world isn’t just to see how prospective customers react; the simulation also helps DayJet figure out where customers want to travel and how and where the company should expand. “What the ant farmers have done,” Iacobucci said, “is build a robust multilevel model that lets us test our pricing, our market deployment and ultimately test our reservation and operations systems against a realistic view of what the world is going to look like before we get there.”
Technology Eases Company Launch
Without modern computers, a company like DayJet could exist, but a lack of information would certainly hobble its launch. “DayJet is a kind of convergence of a number of technologies,” said Herriot, “not only VLJ technology, but also what we call computational mathematics, the mathematics that is possible only by using very large computer programs.”
Without a model to analyze, Sawhill explained, a company could certainly gather the necessary resources–pilots, airplanes, mechanics, customer service people, managers, a Part 135 certificate–and put an advertisement in the yellow pages (or create a Web site), then launch the business and hope that customers decide to try the service and come back for more. That’s how most businesses start.
“If you had deep pockets and were able to take a lot of losses until you figured out the right places to be and the right way to allocate resources, then you could probably get there without the model, but it would be expensive,” he added.
The model, Herriot said, is “really a ‘what-if’ machine.” It gives the company leaders “tuning knobs” for the operation, which they can adjust to see how the virtual customers react and how that affects the measures of business success: profits, revenue, customer satisfaction.
Every company, of course, has its own tuning knobs to adjust deployment of resources. But most companies set them at the beginning of a quarter, Herriot explained, then do their best to run the company at those settings, and after the quarter ends they assess the results to see if those settings worked. “What these models do,” he said, “is allow you do to a ‘what-if.’ What if we set the knobs at a different setting? [Or] serve different airports, charge different prices, have different capacity? And we can do thousands of what-ifs.”
Using data from the Department of Transportation and private vendors, the ant farmers created agent-based models that assign different demographic attributes–such as where they live and their income levels–to each of the virtual people. In other words, the agents are not all the same and don’t live in an “ideal gas” world.
“An agent,” Herriot said, “is a little tiny glob of software. We create an agent for the purposes of a mission. You can think of it as an individual waking up in the morning and saying, ‘I need to go from Tallahassee to Tuscaloosa because I have a three-hour meeting at a certain time of day.’ So now I have a mission, and that agent is created for the purpose of carrying out this travel mission.”
The missions are derived from DOT and other data that provides records of how people travel. In the digital terrarium, the ant farmers use a trip generator to create missions for the 50 million inhabitants.
First they tested the accuracy of the system without DayJet in the mix, by verifying that the actual trips taken by real people in the DOT data reflected with a high degree of fidelity the way the agents behave when the trip generator is switched on. If the model skews away from reality, the ant farmers can adjust the system until it more closely matches how southeastern U.S. citizens travel on business (car, train, airline, charter). Once those results accurately reflect reality, Sawhill said, “We’re inclined to believe what happens when we introduce a new option, namely DayJet.”
Every day, the model creates 500,000 missions for the digital denizens of DayJet World. “Each mission is computed individually,” said Herriot. “The individual tasked with this mission says, ‘Let’s see, I could drive, but it would take this long and my time is worth this amount, or I could fly Delta and go through Atlanta, or I could take a charter or I could take DayJet.’
“There’s a menu of possibilities, the same one that real individuals face when they have a travel mission in front of them. And they make the decision based on the cost of the trip, the expenses, the value of their time and so on. Taken together, they make a decision about what you might call the best choice for that particular mission profile and for that particular individual profile, namely what their time is worth and how important it is to save time.”
The detailed model adjusts itself based on a number of factors, including some that might seem insignificant. For example, if a virtual user picks an airline flight and arrives at the destination around lunchtime, the model builds the time and cost for lunch into its calculations. The agent may, for this trip, decide that the time and effort involved in flying the airline, renting the car and getting lunch is not worthwhile and that the DayJet flight is the better option.
The key result of all this computational effort is not just to see if the agents choose to fly DayJet but if they choose to do so when the model matches reality as closely as possible. If the model’s fidelity is accurate without DayJet, said Herriot, “then we drop DayJet [in] kind of like a sponge and see which missions we soak up based on people’s preferences.”
“If we soak up enough, then there’s a real business there,” said Sawhill. And once DayJet begins operating (likely this month, if it hasn’t already launched by now), actual data from real operations will help the ant farmers further refine their model.
No Black Jelly Beans
Ed Iacobucci uses black jelly beans to represent disruptions, based on his and his wife Nancy’s experience running an ordinary high-end charter operation called Winged Foot (which has since been absorbed into DayJet or vice versa, because DayJet adopted the charter operation’s certificate).
Customers who fly on Winged Foot’s Challenger or Learjet can request black jelly beans (or jelly beans of any color for that matter), and that is part of the service the charter company provides. That is not true for DayJet customers. As Iacobucci sees it, black jelly beans can throw a monkey wrench into the DayJet works.
Too many black jelly beans–disruptions that would affect the company’s ability to deliver service between certain airports–and the system could come to a grinding halt.
Disruptions are OK, but DayJet must manage them. Normal disruptions such as weather delays, mechanical problems and staffing challenges have been built into the company’s model.
The goal is to build a system that can andle disruptions smoothly. DayJet wants to present a reliable face to its real-life customers so they come back. The ant farmers use the word “robust” to describe DayJet and other systems that complexity scientists design using their agent-based models. A terrific example of a robust system is the Internet, which distributes multiple packets of data that split up as they are sent from one computer then reassembled at the destination. The Internet is designed such that packets that are delayed or destroyed during the journey are replicated and rerouted so that they all arrive safely together at the destination.
Preparing for the Unpredictable
An ant colony is also robust. “We understand nature and how it is that an ant colony can be so robust,” said Herriot. “Even though you stomp on the colony in your kitchen, they come back the next day. That’s robustness.
“And we’re engineering a system that is robust even though we get weather disruptions, even though people cancel their flights, even though airplanes need to be maintained and squawks [happen] and so on. When we tell the customer we’ll have an airplane there, we do. Because we engineer it so that the system has just the right amount of redundancy so if this airplane isn’t available, another one is.”
Added Sawhill, “We can’t eliminate disruptions. You can take the hard-line approach of just trying to strong-arm it, but you can also say we’re just going to have to live with this because [with] the number of aircraft we’re operating [there are] going to be a lot of disruptions. So how can we live with it and in some cases even use it to our advantage? This has been the guiding light to our system.”
So how does DayJet handle actual disruptions? These can be modeled, too, and the model shows that the way to build a robust system is to engineer in the correct amount of wiggle room.
Say a planned DayJet flight suffers a mechanical problem and has to be canceled. Does DayJet have spare Eclipse 500s sitting in reserve? No, said Sawhill, “but we have spare capacity built into the system. There are enough gaps between flights and enough spare capacity that we can squeeze it out of the system to pop out another airplane to step in the breach where one didn’t happen. And we can do that without having to change every flight plan in the whole system.”
How much wiggle room is needed? Again, DayJet looks to the computer model for the answer. “We know exactly how much wiggle room,” said Herriot, “how to quantify and measure it and how to make sure it’s distributed in the right way, and we can model it to show that it’s so.”
DayJet won’t take on as much business as the market might offer but will adjust the demand to match the capacity constraints it needs to deliver reliable service, to maintain that wiggle room. “Rather than accepting reservations up to the point where we can’t possibly fly another body, we cut it off to the point where’s there’s still wiggle room,” said Sawhill.
The point is to avoid creating a brittle, instead of robust, system. He said today’s airlines are examples of brittle systems. “When they have one failure in the system it can cascade through the whole airline.” One way to avoid airline brittleness is to avoid hub-and-spoke systems, which is the Southwest Airlines business model. DayJet has no central hub, said Sawhill, “so there is no single point of failure.”
To adjust the wiggle room, DayJet can raise prices when demand is too high or lower prices for midday trips if there are too many airplanes sitting around in the middle of the day. Resource allocation is another way to manage wiggle room.
What if, after DayJet starts up, the airlines get worried that they are losing lucrative business travelers to this upstart on-demand charter operator? Couldn’t they easily stomp all over the DayJet anthill simply by lowering their prices? The answer, of course, is yes, but the ant farmers know that ant-stomping is usually ineffective, and they have the model to prove it.
“We did one experiment,” said Sawhill, “where we set the price of all commercial aviation to zero. We wanted to know if we made all commercial aviation free, would there be anything left? Surprisingly, not only was there, but there was quite a lot. We thought it might be an error or odd, but we looked into it more specifically and it turns out that sometimes commercial aviation is so inconvenient, that just the value of time can make DayJet worthwhile. Certainly it would cut into our business if all commercial aviation were free, but we cut into their business, too.”
Of course it would be impossible for an airline to serve the same city pairs that DayJet will fly, and that is part of DayJet’s robustness. And one route, said Sawhill, “that’s just one of thousands of city pairs that we’d serve. It’s like stomping ants, you just can’t get them all.”
“We are in effect an adaptive airline,” said Herriot. “We adapt to the demand every day. Maybe from Tuscaloosa to Asheville, there’s one person today, five tomorrow, three the next day and zero the [following] day. Well if there’s zero, we don’t fly it.”
The success of DayJet doesn’t hinge only on the ant farmers’ work, although it is a significant factor. “Once you get all the data,” Iacobucci asked, “how do you organize it and create a work plan for the day?”
There is another group of computer experts at DayJet called the rocket scientists. Their mission is to take the ant farmers’ output, which tells DayJet where the virtual travelers want to go and when, and figure out how to make it happen with DayJet’s resources, using a system called real-time optimization.
“Those are the guys who work on algorithms and finding solutions to problems quickly,” Iacobucci said. “They take the requests that come in and build the plan for the day for flying.” The requests for flying that come from real DayJet customers look to the real-time optimization system just like the requests generated by the ant farmers’ virtual customers. Ultimately, Iacobucci explained, the combination of the two technologies makes DayJet possible.
In late July, DayJet began testing its system with employees acting as customers, using the DayJet online reservation system and flying between DayJet facilities on three Eclipse 500s. DayJet had 20 type-rated pilots as of August 1 and is constantly running training classes for new pilots and employees. By the second week of August, Iacobucci expected DayJet to have 12 Eclipse 500s in the fleet, and after FAA proving runs are completed, the company should be ready to launch.
All of DayJet’s seven Eclipse 500s had received the new pitot/AOA probe modification by early last month, and the airplanes have been flying IFR and in RVSM airspace, too. During employee testing in late July one airplane flew what might be a typical DayJet day, racking up eight flight segments in one day. Flight segments have been averaging about 46 minutes, including climbing to the mid-20 flight levels because it doesn’t make sense to climb higher. “By the end of August,” Iacobucci said, “I think we should be in a pretty high level of execution.”
DayJet, which has 300 Eclipse 500s on order, said on August 7 that it “closed debt facilities totaling $140 million for the acquisition of its fleet of Eclipse 500s.” The debt facilities consist of senior debt on the aircraft and money to finance pre-delivery deposits. DayJet has now raised more than $200 million.
“We feel comfortable saying this is the largest amount raised for an on-demand carrier in history,” DayJet CFO and vice president of business operations John Staten told AIN, adding that it also places the company in the top five for pre-launch capital. “This [financing] will take us well into next year with our current delivery schedule [and] allows DayJet to focus on the launch and operations of the service.”
How Does Theoretical Science Help Build a Business Plan?
Scientists have discovered that underlying our seemingly ordered world of bricks and mortar, those things we can touch and manipulate and whose properties and interactions we can model mathematically, there is a world of tiny particles that glue everything together. Biology, chemistry and physics study particular aspects of nature, explained particle physicist and DayJet employee Bruce Sawhill.
Mathematics is used to calculate the interactions of biological, chemical and physical properties, a short-hand language “for understanding very complicated things,” said DayJet computer scientist and complexity scientist Jim Herriot.
But when delving beneath the surface of the stuff that makes up our universe, below the level of atoms and the many tinier particles that atoms are made of, mathematics is no longer useful and complexity science has to step in. Chemistry does look at the behavior of many tiny particles, but lumps them together as an ideal gas, the particles of which all behave the same.
Complexity science assumes that the items being studied–particles, decision-making humans, whatever–act based on unique factors that affect each item. A potential DayJet customer, for example, has a choice when he needs to travel between Boca Raton and Gainesville, Fla.
He can choose among the five-hour drive, a highly inconvenient airline flight, a private charter aircraft or a DayJet flight. Using complexity science tools, Herriot and Sawhill are able not only to study the behavior of people making travel decisions but they also can model this behavior and see whether, given the choice of a DayJet-type on-demand charter operation, people will choose to fly in a relatively small jet.