What Can the McLaren Racing Team Teach the Rest of Us?
Six laps into the 2008 Monaco Grand Prix, Lewis Hamilton’s McLaren race car skidded on the rain-slicked pavement, bumped against a barrier, and blew out its right rear tire. At the time, Hamilton, a gifted, impatient driver and one of auto racing’s biggest stars, was in second place. A puncture is a serious setback in any Formula One competition. In Monaco, the most prestigious title on the schedule, it’s a disaster: The course is laid out on the principality’s twisting, hilly streets, rather than a purpose-built racetrack, so passing is nearly impossible, and ground lost is particularly hard to regain. The three-time Formula One champion Nelson Piquet once likened the race to “riding a bicycle around your living room.” Rain only compounds the challenge.
When Hamilton clipped the barrier, 13 members of the McLaren race team were sitting in a windowless control room in the English town of Woking, 900 miles away. Outside, herons stood in the manmade lake that laps at the curving glass facade of the McLaren Technology Centre. The men and women at the banks of monitors, dressed in the same black and white uniforms as their teammates at the track, included strategists, systems engineers, performance engineers, mechanical engineers, and IT specialists; dozens of others in the building were patched in as well. Many of the decisions about the car’s setup and management over the course of the race are made here, not at the track. The team now had less than 30 seconds, the time it would take Hamilton to ease his car into the pit area, to make a very important call.
In the weeks and days leading up to the race, McLaren engineers had been running thousands of simulations, testing components, configurations, settings, and strategies. After the race started, the simulations continued to run, their predictive power improving lap by lap as information from the track was fed in. That meant there was a recommendation in the system for exactly what happened—Hamilton needing a pit stop in the sixth lap in a drizzle expected to soon taper off. Just six seconds after Hamilton called in his flat, a note of panic in his voice, the race engineer got on the radio and calmly told the pit crew to ready a set of tires—not deep-treaded “full wets,” but intermediate tires that could grip drier pavement as well. At almost the same instant the team manager told the crew to pump in extra fuel.
Both decisions were calculated gambles: The extra gas would weigh the car down, and the intermediates wouldn’t perform quite as well in the rain, but Hamilton would be able to stay out on the track past the point when his competitors would need to refuel and change their tires, gaining ground on them when they did. Within 10 laps, Hamilton had climbed back to third place, and when the two drivers ahead of him had to pit, Hamilton took over the lead. He held it until the checkered flag. It was one of the most dramatic Monacos in memory, and Hamilton would go on to win that season’s Drivers’ Championship.
“It’s all probabilistic,” says Mark Williams, McLaren’s head of vehicle engineering. “Because the system is running races live in the background, you can say to it, ‘How am I going to beat the guy in front?’ It goes off and specifically looks at all the options that he could do and you could do and comes up with the best solution based on probabilistic analysis. You may or may not beat him, but the closest you’ll get to him is by doing this strategy.”
McLaren has long had a reputation as a data-obsessed racing operation. It makes the telemetry systems for all its Formula One competitors, along with the computerized engine control units for Formula One, IndyCar, and Nascar. When a McLaren car is on the track, more than 120 sensors transmit a torrent of information on tire pressure, torque, temperature, and downforce (the vertical pressure, vitally important in cornering, created by airflow over a moving object).
The company has decided that the highly specialized expertise it’s developed in data analysis, simulation, and what it calls “decision support” is something that businesses far outside the world of racing would profit from and pay for. Five years ago, McLaren launched a consulting firm called McLaren Applied Technologies, under the leadership of Geoff McGrath, an engineer who started his career in the oil and gas industry and later moved into telecommunications. Among other projects, MAT has designed better health monitoring systems for sick children and created a scheduling system for Heathrow Airport that reduces flight delays. It’s worked with some of the world’s biggest oil and gas drillers to cut down the time they spend on fruitless exploration and with data center operators to better manage spikes in demand. It’s helping pharmaceutical companies speed up drug trials and reduce the time their production plants sit idle.
At a moment when big data has become de rigueur—and when cheap sensors and computing power are making everything measurable—the challenge is to know which data are important and which are merely noise. McLaren has decades of experience figuring out that very question and, just as important, setting up systems that can do it in real time—providing a sort of scaffolding, in making high-stakes decisions under extreme pressure, for the easily overwhelmed human brain. The company has a stable of engineers used to working with enormous resources, the latest technology, and very little time. It’s betting that lots of business problems can be treated, essentially, as races, and that it has a head start.
One of McLaren’s first consulting projects, in the wake of the 2008 Beijing Olympics, was for Britain’s Olympic team. MAT worked with athletes in four summer sports and two winter ones: rowing, cycling, sailing, canoeing, the bobsled, and the skeleton (a “tea tray on skates,” as the English like to describe it, driven headfirst down an ice chute by one person, prone).
The leader of the effort was Caroline Hargrove, a mechanical engineer with more than a decade of experience in racing. Hargrove was a lecturer at Cambridge University, just beginning to question her suitability to a life in academia, when she spotted an ad for a job at McLaren Racing in an engineering journal. The Montreal native had little interest in cars, but she liked interesting problems, and she applied.
She started at McLaren in 1997 as part of a small team developing a racing simulator. Then as now, women were a rarity among Formula One engineers. Simulators were rarer still: None of the teams had one. Even at McLaren there was little enthusiasm for the project. The drivers complained about the poor graphics, and the team’s head designer at the time, Adrian Newey, let it be known that he found the technology useless. Gradually, Hargrove and the other engineers improved the software and the mechanics. Then Formula One imposed limits on the amount of time cars could be tested on the track, and suddenly every team needed a simulator.
Today, McLaren has two of the machines: full-size car bodies mounted on hydraulics surrounded by curving video screens, with robotic arms that jerk the driver’s helmet back and forth to imitate the violent G-forces of high-speed turns. One sits in the basement of the Woking headquarters, near the wind tunnel; the other is in the laboratory-like area above, where McLaren’s race cars are built and the ultrahigh-end consumer cars it recently began selling are designed. An encyclopedic range of courses and conditions can be programmed into the simulators, but they’re more than just training tools. They’re able to calculate how different components affect a car’s handling, even if those components exist only as a set of specs. That means the team can test parts on the simulator before it actually builds them, so only the promising ones would be fabricated and tried out at the track. Before McLaren started using a simulator, just 10 percent of the parts it made ended up being used in its race cars; now, 90 percent of them do.
A former ski instructor who’s played tennis, squash, ice hockey, and badminton competitively, Hargrove threw herself into the Olympics project. MAT was in its infancy when the project started, so Hargrove got colleagues from the Formula One side to donate their time. Car designers made new skeleton sleds, customizing them to the bodies of each rider. Data engineers installed a readout right in front of the sledders’ faces for use while practicing in the wind tunnel. The device told them how much drag their bodies were creating and how shifts in their head and shoulder position affected it. This was information the athletes had previously gotten only after practice, not when they could immediately respond to it and see the effect.
For pursuit cyclists, who race single-speed bikes around a velodrome, Hargrove’s team created a data-logging device that mounts under their seats and sends information to the coach on torque, speed, power, and the tilt of the bikes as they bank into turns. McLaren programmers also wrote software that plotted that data as it came in against team targets, past runs, and competitors’ results. “It used to take three weeks to analyze data from a race weekend,” Hargrove says. “Now they are able to do it automatically as they go along.”
In similar work with the English national rugby team, McLaren engineers took data that the coaches were already gathering from sensors placed on players during practice and designed algorithms to glean new information. Hargrove’s analysts were able, for example, to determine how fatigued a player was—and therefore how susceptible to injury—by how long it took him to get up after being tackled and how much his pace declined over the course of an afternoon.
Over three Olympics, the athletes McLaren worked with won 17 gold medals, and 32 medals in all. McLaren couldn’t take all the credit—these were sports in which Britain had done well before, and its pursuit cycling team has dominated the event for a decade—but it was an auspicious place to begin diversifying from cars. As it grew toward its current head count of 250, MAT hired engineers and designers of its own, many from outside the racing world.
Another early client was Specialized. The Northern California bicycle maker had long claimed that “smoother is faster,” meaning that the ability of its high-performance road and mountain bikes to absorb bumps not only made them more comfortable to ride but also quicker than the competition. Smoother, however, isn’t always faster: A too-flexible bike steals power from the rider’s pedaling. So how smooth was too smooth? The testing methods that even industry-leading bikemakers like Specialized used—putting people on prototypes to ride them and report back—were highly subjective and of limited use.
Hired in early 2010, the MAT team decided to build a bicycle version of its race car simulator. They quickly realized that the project raised its own set of challenges. Unlike in a car, the rider’s body was a major part of the calculation. “The human’s a big, fat, blobby mass on a very stiff, light structure,” says Duncan Bradley, the engineer and product designer who led the team. To work, the simulator had to re-create the dynamics of the rider as well as the bike.
For three years, the McLaren team gathered data on how different parts of different Specialized bikes performed, individually and in combination, with and without the rider. A serious cyclist himself, Bradley often served as the test rider, perching in spandex and helmet on a deconstructed bit of bicycle mounted on hydraulics. The rig would send calibrated jolts up the handlebars or the seat, into his hands or hindquarters, with sensors measuring the diffusion of force. It was slow work, but by mid-2013 the team succeeded in creating, essentially, a virtual bicycle: “We turned the whole system, including the rider, into a mathematical formula,” says Bradley. Now Specialized’s designers and engineers could, with a few keystrokes, change the shape or weight or stiffness of various parts, and try out the resulting bike over different road surfaces.
“We’d been taking the measurements that were used to build this model for years, but we could not put it together,” says Mark Cote, Specialized’s manager for aerodynamics research and development. “There’s a kind of analysis paralysis in the world today, with every single piece of your world being measured in some way.” McLaren, he says, laughing, “has way too many sensors on their cars, and they know how and when to interpret that data.”
By late 2011, McLaren was generating interest outside sports. Stephen Mayhew is a biochemist who leads a team of analysts that shapes R&D strategy at U.K.-based health-care giant GlaxoSmithKline. He’s long known about McLaren—Formula One is a big deal in England—but when his superiors told him they’d hired the race car builder to help him with his job, he was dubious. Mayhew, who’s exceedingly polite, puts it this way: “There was, I think, a healthy skepticism on our side.” He was sent to a meeting in Woking, at McLaren’s on-site juice bar, and it was only when Hargrove described the Olympics and rugby work that Mayhew began to think a collaboration might not be a waste of time. Drugmakers, like Olympic coaches, are interested in human performance.
For all the sophistication of the drug discovery process, trying them out on people remains a time-consuming, low-tech process. Volunteers take the drug (or a placebo) and then are monitored by a doctor through visits every few months, so data points are few and far between. Mayhew wondered whether patients could be monitored remotely, like rugby players and pursuit cyclists. If information could be constantly logged and transmitted back to the testers, a drug’s effects—or lack thereof—could be spotted much sooner, saving labor, time, and maybe a lot of money.
The GSK and MAT teams decided to focus on disorders that affect a person’s motion. The idea is to measure patients’ movements—how much they move, how often, how far, even how fluidly—over time as they are treated. If a drug is effective, researchers will be able to detect the patient’s improvement in real time. In a pilot study, McLaren engineers put sensors on 80 stroke patients. GSK is analyzing the results of the trial, but if they’re promising enough, the company will consider similar work with patients being treated for Parkinson’s disease, ALS, and arthritis.
In addition to developing cures for debilitating diseases, GSK also makes everyday items. The engineering director of one of its toothpaste plants, in Maidenhead, England, brought in McLaren to improve the efficiency of his production line. The work focused on changeovers, the brief windows when the line is stopped to change flavors. It’s analogous to a pit stop, and McLaren’s consultants treated it that way, using video analysis to refine each team member’s responsibilities and movements, rearrange how tools are laid out, and create a seven-step preparation and debriefing regimen for the teams. Over six months average changeover times dropped from 49 minutes to 15, a reduction that, by GSK’s calculation, translates into 7 million more tubes of Sensodyne and Aquafresh a year.
In another collaboration with Mayhew’s team, McLaren designed a data visualization tool for GSK’s chemists. Drug researchers go through tens of thousands of compounds before they settle on the one to put through trials, evaluating them on the basis of potency (how well it works), solubility (how well it gets to where it’s needed), selectivity (how focused its effects are), and other qualities. The problem is that altering a molecule to make it more potent can make it less selective or soluble, or vice versa. Teams can spend months tinkering with compounds only to find that they’d found the optimum balance long before. Software McLaren created for GSK takes the tables of raw chemistry data from all the compounds and translates them into intuitive graphics, so researchers can more easily tell when their efforts have reached the point of diminishing returns.
Working with major oil and gas exploration companies (McLaren won’t say which) in the North Sea, MAT is developing a data analysis and graphics program for drill operators. Guiding an undersea drill requires dexterity and sangfroid: The operations burn through a million dollars a day, so the cost of delay is high, while the cost of an error can be ecological disaster. As a drill descends, it bores through layers of hard rock and softer sand. The operator has to adjust the speed at which the drill rotates or the pressure with which it’s driven into the ground or the pace at which it clears out rubble. If he gets it wrong, the drill bit can stall—or worse, vibrate so much that it begins to whip around at the end of what can be a 10-mile-long drill chain.
To help the operator, the McLaren program takes the information coming up from sensors on the drill bit and creates a constantly updating set of recommendations for how to proceed. Those recommendations can be tailored to how conservative the operator wants to be or how much risk he’s willing to swallow. “We knew nothing about drilling,” says Hargrove. “But we realized it’s not so much the drilling process as such; it’s how do you proceed when something doesn’t quite go right.”
Air travel, too, is a realm where things often don’t quite go right. The limited supply of airport gate slots and runway space and the inevitability of poor weather combine to create a tightly coupled network where delays and bottlenecks can quickly ripple across continents. The managers at airports who coordinate arrivals and departures are at the center of this improvised, jerky dance. When they wake up in the morning, planes that took off the day before are headed their way—some already late or rerouted—and they have to figure out how best to bring them in.
London’s Heathrow Airport presents a particularly intricate puzzle. It moves more people than all but a couple of airports in the world, yet it has only two runways—Chicago’s O’Hare International, by comparison, has eight. In an added limitation, local environmental and noise regulations restrict most flights to between 6 a.m. and 11 p.m. “These are the two busiest runways anywhere,” says Jon Proudlove, the general manager for Britain’s National Air Traffic Services (NATS) at Heathrow, sitting in a conference room at the base of the airport’s control tower. Almost 300 feet above, in a glass-walled, top-hat-shaped room filled with the murmuring of coordinates and instructions, a team of 11 choreographs the constant flow of takeoffs and landings. It’s a clear day, and the procession of jetliners on final approach hangs in a long, straight line out over London and the eastern horizon.
“People will only sleep on the floor of Terminal 5 once, having missed their British Airways flight in this random world that we operate in,” says Proudlove, “and they’re going to say, ‘You know what? Sorry. I’m going to get the KLM from my local airport and go to Amsterdam,’ where they’ve got more runways than I can ever dream of.”
Proudlove is the point man on a NATS partnership with McLaren to create a better scheduling system. Previously, scheduling had relied on a computer program that looked at a few “study days” from the past season, usually idealized days in which little went wrong. Working with NATS, McLaren created a software tool that allows Heathrow’s Runway Scheduling Limits Committee to model bad days as well as good ones and to simulate the effects on global air traffic of events such as a blizzard in Frankfurt, thick fog in Singapore, or a volcanic eruption in Iceland. That’s enabled the airport to better plan for delays and, as a result, to increase its capacity.
“Before, you were looking at a blank piece of paper. Now you’ve got a picture and some sliders and some knobs,” says Mike Phillips, a former road car engineer who headed the air traffic control work. “You can play with it and start seeing the effect of your actions before you take them.”
McLaren is now working on a more ambitious version of the software that can adjust the schedule day by day, even hour by hour. Nicknamed “Race Day,” it will constantly update its predictions with actual data on the speed and position of incoming aircraft and the weather in their path, tuning the models toward reality. One of its most valuable uses would be as a damage-control tool on those days when multiple meteorological and logistical dominoes fall. If, for example, it becomes clear by midafternoon that Heathrow simply won’t be able to handle all of its remaining scheduled arrivals before the 11 p.m. cutoff, Race Day will recommend how to proceed based on what the user wants to prioritize. Is the goal to cancel the fewest flights? Preserve the most connections? Favor long-haul flights over shorter ones? The software won’t make the decisions itself—those have to be negotiated among the airlines, airports, and national aviation authorities. But it will allow them to do it with far greater awareness of the ramifications.
“These guys have the simulation capability to take onboard all of the information that is now coming out of the ether as all of these planes start to move toward Heathrow, or are at Heathrow ready to get out of Heathrow,” says Proudlove. “Then they can start to tell us, ‘Now this is what your day is going to look like. This is how it’s starting to pan out against that plan that you created.’ That is 3 million zillion miles away from where we are today.” In a simulated run, the program improved on-time performance at Heathrow by 19 percent.
While Heathrow will begin using the McLaren/NATS scheduling platform early next year, Race Day, like several of the decision-support systems MAT has created, is still in the proof-of-concept stage. Heathrow might yet decide it isn’t needed, just as GSK may decide that sticking sensors on patients in drug trials raises too many regulatory issues. MAT’s whole business model, which depends on finding ways to collect licensing fees for the technologies it creates, is an experiment with no guarantee of success.
McLaren can always go back to just racing and selling cars. Its engineers, though, seem to really like thinking about things besides downforce and tire wear. Peter van Manen was head of McLaren’s electronics division in 2009 when he gave a speech at a medical conference on what doctors could learn from car racing. Heather Duncan, an intensive care doctor at England’s Birmingham Children’s Hospital, was in the audience and approached him afterward. The two have been working together since then, trying to figure out ways to monitor very ill children.
On a Tuesday in mid-August, Duncan is leading a tour of the Birmingham hospital, a turreted Victorian pile in the heart of the old industrial city’s downtown. Children with pneumonia and septicemia, and others recovering from cardiac failure, organ transplants, and serious accidents, lie in beds separated by curtains in large open wards. They range from newborns with birth defects to teenagers—some alert and curious, others heavily sedated, almost all attended by worried parents.
The difficulty with patients like these, Duncan explains, is that they’re “normally abnormal.” The serious health problems they suffer from mean that the baseline for their various vital signs—their “normal”—can be outside the typical range to begin with, and that makes it harder to spot trouble. Their condition can deteriorate from stable to life-threatening in a few minutes—between the time a nurse checks on the child and returns to check on her again an hour later.
Van Manen tries to explain: “I mean, a lot of these kids are born slightly broken,” he says. Duncan cuts in with a reproving smile. “This engineering slang,” she says. “They’re individual. They’re very individual.”
Patients in intensive care units today already have their heart rate, oxygen levels, blood pressure, and the like measured. Duncan and Van Manen are refining software that would spot worrying trends in those indicators long before they cross the critical threshold that sets off alarms. That could help doctors head off problems. “If the data tells you there is a cardiac risk in 10 minutes’ time, that’s not particularly useful. If it tells you in three hours’ time you’re going to have a cardiac arrest, that’s extremely useful,” says Van Manen. “In sort of simplistic, nonmedical terms, if you can start to help when the body is still trying to fix it or even before the body has given up, you will be in a better position.”
Van Manen and Duncan hope what will result from this is a new combination of sensors and software, something that can be sold to hospitals far beyond Birmingham. That’s a ways off; Duncan expects they’ll be testing the technology for at least the next six years. Right now the software they’re using still has its old Formula One default settings for the categories of data it’s gathering. “It’s tire pressure, tire temperature,” Duncan says, then thinks for a moment—“and something to do with gears.”