Colton Herta jokes that “nothing sticks” when asked about what he’s learned from the melding of analytical brainpower and brute horsepower being attempted this season at Andretti Autosport.
But there was one thing that stood out about the founder of Zapata Computing, a company spun out of Harvard that wants to use sophisticated, quantum-ready algorithms to make Andretti a more efficient and successful race team.
“That was pretty wild when he was like, ‘Yeah, I made Siri,’ ” Herta told NBC Sports about meeting Zapata CEO Christopher Savoie. “I was like, ‘What?’ That’s pretty insane.”
It might not be so simple as saying, “Siri, devise a strategy to win this IndyCar race” into a smartphone app.
But the goal of the Andretti-Zapata partnership is guided by the same predictive analytics and concepts that are rooted in artificial intelligence technology and big data.
INDYCAR AT LAGUNA SECA: Details, schedules for watching the season finale on NBC
With millions of data points compiled by thousands of sensors that produce roughly a terabyte of information per car during every race weekend, there already is too much information for Andretti Autosport’s 38 engineers to digest with six cars (including two from its alliance with Meyer Shank Racing).
“We do all the same things that a Formula One team does except with a tenth of the people,” Andretti Autosport technical director Eric Bretzman told NBC Sports. “It’s massive overload for us to take all the data and make a decision in a second (with) what the tires are doing, the aero is doing and the human element of what the driver is doing. There’s so much more data now than there’s ever been to really define this model and start to draw some strong trends.”
So the team joined forces this year with Zapata, which has embedded a handful of its data scientists at every IndyCar race this season. The self-described “gearhead computer geeks” (as Savoie calls them) have participated during engineering debriefs with Andretti drivers inside the team’s Race Analytics Command Center (also known as the RACC, it’s an engineering trailer rebranded with Zapata logos and themes).
By understanding the decision-making processes made on car setups and race strategy, Zapata has been helping Andretti upgrade its computer modeling infrastructure to handle more variables – mainly focusing on building race strategies around predicting tire degradation, rival teams’ tactics and optimal fuel consumption.
The goal is to have an app that could be used by the team in races next season.
“We’ve got to get it in the right format, start with some fundamentals and build on that for the future, but already we’re starting to see models that will be useful in making those predictions,” Savoie told NBC Sports. “It’s a phenomenal opportunity for us because it’s fun getting the play by play across six cars.
“There’s a lot of work to do, though, and this needs to be proved out. But I hope we can get to that point of pointing out look at this tire decision, here was the model, where we looked at the data, and we made this decision, and here’s your podium.”
The Andretti-Zapata partnership is unique in that its long-term goal is building a foundation for the use of quantum computers, which are years away from practical use in everyday life.
But virtually all major-league racing teams (including Andretti) already are using predictive strategy simulations with “classical” hardware.
“In today’s world, we rely on software development programs to help us achieve things we can’t achieve on our own,” Chip Ganassi Racing managing director Mike Hull told NBC Sports. “We do the same thing they’re doing. I think most teams do. Some don’t talk about that. You see Formula One doing it with their remote war rooms helping them try to decide what to do.
“I think what Andretti Autosport is doing is a great thing, and they’re also promoting and exposing a software development company based on data analysis. If they can eventually develop a strong enough database that includes the surprises, then they probably can make better decisions on what they need to be doing when they call a race, absolutely.
“The problem with motor racing is it’s full of surprises and just when you think you have it right, you don’t.”
Zapata’s interest in Andretti started with Savoie, who has tinkered with cars since his teens (starting with a double-barrel Datsun 210). In addition to inventing the natural language interface used to help build Siri before joining Zapata, Savoie also worked on predictive battery analytics at Nissan and developed the data platform around the Leaf model.
After attending the inaugural Music City Grand Prix in Nashville, Tennessee, Savoie (a published scholar in medicine, biochemistry and computer science) was connected with Andretti management by a team member who thought the organization wanted to upgrade its analytics.
He met with Bryan Herta at the 2021 season finale in Long Beach and began brokering a multiyear deal that was announced in February this year. It’s a unique sponsorship for Zapata, a global company that has nearly a third of its engineering staff based in Boston.
In exchange for providing analytics expertise, Zapata can promote and market its work with Andretti to the public – a level of exposure it’s not typically allowed by its clients. Much of its work is with defense contractors and multibillion-dollar companies (British Petroleum is a prominent customer) that prefer to keep a low profile on their projects.
“This allows us the opportunity in a high-speed and big data analytics platform to talk about the successes in a very public way that’s win-win,” Savoie said. “Of course, we’re not going to reveal all the really cool things we learn about strategy, but we can talk about the kinds of problems we’re solving. (Andretti) needed an uplift on the infrastructure and a partner who would help them do some really advanced analytics that they wanted.”
Said Colton Herta: “It’s a great sponsorship because we both benefit each other. You don’t really get that in racing. People pay and get branding and business to business opportunities, but it’s never like this where they’re actually on the team, and we get all this help from them.”
Savoie said Zapata is benefiting from being in the high-pressure real-world conditions of racing, which offers data sets as large as some of its biggest clients but without a controlled environment.
Being at the mercy of wireless connectivity that can vary across tracks, along with the rigors of travel, creates challenges unseen in the business world (where Zapata uses the Orquestra quantum software platform).
“The series doesn’t allow you to create your own ad-hoc CAT5 cable network because it can be dangerous,” Savoie said. “So having servers on a truck that goes through from Indy to Long Beach and through the desert at 120 degrees, that kind of extreme environment is something we don’t see with a lot of corporate clients, so it’s actually more difficult, and it’s a faster pace. The amount of data to keep track of 24 cars at 200 mph, where the difference between first and second might be a half-second after 500 miles at 240 mph in the case of the Indy 500. It’s very extreme and the tolerances are so small, that adds a difficulty beyond what we do for BP or BASF or the government, because we don’t have to get you an answer in 0.5 seconds to make that split-second life-or-death decision.
“People might say auto racing is just yahoos going around a track really fast, but every team with this massive amount of data is doing really hard stuff. I think that’s lost on auto fans a lot of time is just how much engineering is going on. It is just as hard as working with a Fortune 100 client with huge cloud systems.
“For us, this is a marriage made in heaven. This is a great way to pressure test what we offer. It sounds ironic that it’s car racing that does that. But basically, they’re rocket ships at 200 mph on land around a track. The extreme nature and amount of data but also the variables you’re dealing with just make this the best subset for computer geeks to want to do big data stuff.”
Aside from Savoie, much of Zapata’s staff is new to auto racing
“They’re all like Bay Area tech gurus, so none of them really knew that much about IndyCar,” Herta said. “So most of it is them trying to figure out how we talk and what certain words mean. They have to build a brand-new vocabulary for racing and just how it works. Most of the guys never had seen any sort of race before. They knew the Indy 500 or Daytona 500 but didn’t really understand all that went into it.
“I think a lot of them were surprised with how advanced the technology is, especially with data. A lot of people think we just drive in circles, do pit stops that are fast, and that’s it. But there’s a lot that goes on behind the scenes, and everything we do is based around data now. We all have all these different sensors and stuff on the car to help us understand every aspect of it.”
Bretzman, who has been a racing engineer since the mid-1990s, said big data long has been part of IndyCar.
For more than 30 years, data systems on cars have been “pumping out 1,000 signals per second for each sensor,” but advancements have ramped up the volume with the advent of more electronic features (such as paddle shifting) and more reliable measurements (laser ride height sensors that once regularly overheated have become like clockwork).
“And we have new things like on-board video from every run showing every corner and every movement of the drivers’ hands,” Bretzman said. “That’s a whole new form of analytics we’re still inventing every day.
“Every time a problem comes up in motorsports, we make a solution. Another problem comes up, it’s another unique solution. Another analytics method. Well, we’ve got databases scattered everywhere now.
The hope is that Zapata can help Andretti bring order to the chaos, while also incorporating its wind tunnel and testing data to provide the most comprehensive and streamlined approach to optimizing setups and strategy.
“We’re trying to get all these databases and all this information – our history and current data — into a pipeline that’s on one platform,” Bretzman said. “So it’s going to be in a usable platform that we can access very quickly to maybe start putting more things together than what we’ve been able to do.
“We’re hiring more software engineers now than we are mechanical engineers or aero engineers or material science engineers. It’s progressing really quick now, and there are opportunities beyond what mechanical engineers know and what our environment is. And that’s where Zapata comes in as an instant shot in the arm of continuing education. They’re on the forefront of advanced techniques for data analytics.”
Before qualifying at Nashville last month, Andretti’s Race Analytics Command Center was humming with team members, engineers and drivers hunched over five flat screens and a dozen laptops.
The RACC trailer is where much of Zapata’s work takes place. Aside from asking the team about tire degradation, downforce adjustments and handling characteristics, Savoie said the Zapata engineers sometimes listen to arguments between engineers and drivers about fuel consumption and the timing of yellow flags because “those are really informative as to what decisions points are being made, and that helps inform the building of the model.
“We need to know what they’re doing in those decisions, so we know what the decision trees are to feed into our machine-learning models that will in help these kinds of decisions (such as) when we believe the pit windows will be and when we believe we can overcut or undercut (on a pit stop),” Savoie said. “Their models and engineering are absolutely incredible behind the scenes, but those decisions still happen to be very heuristic. Good engineers make great race decisions more often than not, but they make better decisions and can be more confident if they know there’s data backing that.”
There are some races in which the impact could be blunted by the forces of nature.
If the forecast is for extreme heat in Monterey, California, strategy Sunday’s season finale at Laguna Seca (where Herta is the defending race winner) could be much trickier to anticipate. Herta’s May 14 victory in the monsoon-esque GMR Grand Prix at Indianapolis Motor Speedway (which Bretzman called a “once in a lifetime anomaly”) is a good example of how unlikely conditions can challenge the efficacy of predictive analytics.
But Savoie also said “being data-driven is always better, so that’s really the goal. Not that we may make the final decision, but there at least would be a dashboard to say, ‘OK, here’s the percentage chance on making this call or that.’ ”
Though many of the questions he gets from the Zapata engineers are rudimentary, Herta said “they’re definitely asking the right questions” about why a car is handling a certain way or how it’s impacted by a mechanical element.
“I’ve been really impressed with just how smart they are — not even talking about racing,” Herta said. “They’re just really smart human beings. It’s pretty crazy.
“But I imagine it’s the same way when I talk to them about racing as when they talk to me about what they do with quantum physics. It’s like I don’t really understand all of what you’re saying, but it’s still really cool.”