WORKING WITH PRECISION SYSTEMS IN A WORLD OF DATA
INTRODUCTION
From Digitization to Precision
The Game
In December 2019, fans witnessed two surprising developments in the National Football League (NFL). The unheralded Baltimore Ravens surged to a championship season, while the Dallas Cowboys—“America’s Team”—imploded. What made the difference? Both teams were rich with on-field talent. But the Ravens had a secret weapon. In the coaches’ box high above the field in M&T Bank Stadium, 25-year old Daniel Stern, a Ravens “football analyst” (holding a degree from Yale in cognitive science) sat elbow to elbow with Baltimore’s offensive coordinator, Greg Roman. Play after play, they pumped Moneyball-style options down to coach John Harbaugh’s headset, each built with artificial intelligence (AI), each a prediction of the game and a prescription for what the Ravens should do next. Harbaugh would pick and choose. Game after game, the Ravens picked up key yardage gains; turned failed, four-and-out possessions into sustained drives; and put points on the board. Games that might have been lost were won.
Dallas coach Jason Garrett had no such advantage, nor did he want it. “Yeah, we don’t use those stats within the game,” he told reporters. The upshot: precision systems that mix human judgment with analytics can easily outperform gut instinct or data alone. Adapt to the new game in town or perish.
The Crime Wave
That same month, in the fading light of a winter’s day in New York City, Tessa Majors, a nineteen-year-old freshman at Barnard College, was murdered in
Morningside Park, a majestic but decaying jewel. In November, a neighbor had reported twenty-three streetlamps out in the park “for the umpteenth time,” a reporter wrote—five in the area at the base of the steps where Tessa Majors was murdered. “Most people know somebody who has gotten robbed there,” a victim said. “I was told by police that it was not a very safe place to be.” One nearby storeowner had complained to cops at the 26th Precinct station house by phone of harassment by kids. He was tired of it, and he was scared. “A lady told me to tell them to ‘move along.’ ”
If the New York Police Department (NYPD) had no idea this crime was coming, it should have. The knifepoint robbery by three teens was the latest in a neighborhood crime wave. One attacker later admitted that they were in the “habit” of robbing people in the park. Since June, five people reported being robbed near the staircase in Morningside Park where Tessa Majors was killed—including two robberies ten days before. There were twenty-one reported muggings in the park and its perimeter compared to eleven the same period a year prior.
Just in case, for some reason, cops missed these patterns, the NYPD had an advanced AI-enabled precision system called Patternizr for discovering lookalike crimes. A well-managed pattern recognition system—whether AI-driven or manual—should have exposed the attackers’ robbery habit sooner, but Patternizr didn’t. Cops might have stopped the crime spree sooner, but they didn’t. Inevitably, one mugging went wrong. Tessa Majors fought back and lost her life.
Lori Pollock, Chief of NYPD Crime Control Strategies at the time, reflected on this: “You could see this coming from a mile away. People weren’t paying attention or didn’t want to tackle it. Youth crime was becoming an epidemic. No one in the court system or the judicial system seemed to take it seriously. There were no consequences for young people to rob two or three people in an hour—just, roll people. And that’s what happened to Tessa Majors.” The failure was unusual for New York’s Finest. In all things operational and digital, the NYPD holds itself out as the gold standard in crime fighting. In movies, television, and real life, its achievements are legendary, its tools and techniques emulated everywhere.
But when it comes to putting data and AI to work even New York’s Finest can stumble. Having a precision system on hand will not flag what matters most if you don’t tell it to, nor will it change the world as a precision system should. “Fix broken process first,” and “Get your organization right and ready for precision” are imperatives you will hear throughout this book—before any precision system can work its magic.
In the end, the only pattern matching done here was by Officer Randy Ramos-Luna. On patrol the next day, he recognized a kid on the street by the clothing he was wearing: it matched the Wanted video still he’d seen on his department cellphone. Pursuit, confrontation, and arrest followed. Back at the station house, the kid gave up his two accomplices—just like in the movies.
Chasing COVID
On December 26, 2019, a Chinese researcher at Weiyuan Gene Technology in Guangzhou was browsing the latest test results from routine disease surveillance. Wuhan had sent over a lung fluid sample taken from a sixty-five-year-old man hospitalized with pneumonia-like symptoms. What the researcher saw alarmed her. The sample, she wrote a colleague, “was brimming with something that looked like SARS.” In 2002, severe acute respiratory syndrome (SARS) had killed 774 people worldwide.
“It’s no joke,” her coworker replied on WeChat. “It’s the same level as the plague.” Tests the next hour confirmed “bat_SARS_like_coronavirus.” “*Worse* than the plague,” her coworker replied.
China did not lock down Wuhan for twenty-eight days. The virus, being no respecter of politics, borders, or national pride, seeped out of China by car, train, bus, and plane. It caught governments from Washington to Tokyo unaware. Studies would later show that, had China locked down Wuhan one week earlier, it could have reduced infections across China by 66 percent, three weeks earlier by 95 percent. In the pandemic that followed, COVID-19, as the virus came to be known, went on to kill over 6.5 million people around the world.
Korea was already on the move. In 2015, it had bungled its own Middle East respiratory syndrome (MERS) outbreak—and fixed its playbook to act fast the next time it encountered a pandemic. That time was now: in January, Korea’s four confirmed COVID cases weren’t exactly a catastrophe—yet it was enough to launch the new protocols. One week in, Korea’s Disease Control and Prevention Agency (KDCA) convened twenty test manufacturers and told them to build COVID tests. The government guaranteed their cost reimbursement. KDCA gave them early access to viral samples and standardized the validation process. Korea’s Ministry of Food and Drug Safety was ready with authorizations.
When Dr. Jong-Yoon Chun, CEO of the Korean biotech firm Seegene, got word of the pandemic, he responded quickly. Using a supercomputer equipped with AI tools, Seegene’s team analyzed the molecular structure of the new virus; worked through billions of possible permutations; and built a COVID-19 test that yielded results within four hours, a test that was standardized so any of Korea’s 120 labs could read them
Three weeks in, across Korea 700,000 test kits were ready. Doctors at Incheon Medical Center proposed innovations in testing protocol: drive-through sites would keep drivers in their cars, avoid spreading COVID in hospitals, and triple hourly screening capacity. By February 23, Chilgok Hospital in hardesthit Daegu had installed South Korea’s first drive-through testing center. The invention brought South Korea’s testing capacity to 20,000 tests per day. In the
opening weeks of the outbreak, 225,000 Koreans were tested. Over that period 8,000 were diagnosed with the disease. Sixty died, but the rapid implementation of a tracing system and social-distancing protocols made possible by the Seegene test and others undoubtedly saved many more lives.
“We are a molecular diagnosis company, specializing in medical testing kits,” Dr. Chun told CNN. “We have to prepare in advance. Even if nobody is asking us.” The upshot: it takes an ecosystem with all players aligned to make precision happen. With that, precision systems can have global impact, and fast.
Race on the Water
When the pandemic was brought under more control, and travel again became possible, international sports events resumed. In September 2021, off the coast of Cadiz, Spain, spectators might have caught sight of a flotilla of F50 sailboats racing through the Atlantic at speeds of 100 miles per hour, their ninety-fivefoot winglike masts and hydrofoils flying them above the waves. Eight teams of A-list sailors from around the world were making a run for a place in the championship Grand Final of the SailGP campaign in San Francisco the next spring and a prize of $1 million. Every boat was identical. Their crews saw and shared the same data. Each boat was connected to the Oracle cloud, streaming terabytes of data from sensors tracking their location, wind speed, boat speed, wind angle, and a host of other elements.
If each boat was identical and all the data were shared, what would be the key to winning? Talent and experience, certainly. Training, too. During the few months between the Olympics and the America’s Cup, these teams had to come together fast with little time to practice. Every minute together mattered. But the difference maker off Cadiz and later San Francisco was the playbook unique to each team. Each choreographed its own moves, starting with a sprint to the video room after practice to see what worked and fix what didn’t. That proved decisive on the open waters of San Francisco Bay where the Australia boat defeated all, dodging collisions, capsized competitors, and even a whale wandering onto the racecourse. The upshot: well-managed and built for continuous improvement, world-class precision systems deal with exigency, unleash the power of talent, and take the win.
“You’re racing against the best guys in the world,” Russell Coutts, SailGP’s CEO and five-time America’s Cup champion told a reporter. “If you give them more time against you, you’re going to get hurt, aren’t you?”
Time to the finish wasn’t just everything—it was the only thing. “Glory,” the Australia boat captain said, “lasts forever.”
What do these four stories have in common? Massive amounts of data became available, sometimes deluging decision makers. Each was a race of
teams against time and nature. Each depended on human-built and -run systems, loaded with AI, but with men and women, not machines, finally making the difference. Some used the data and the tools well, whether in the press of a game or a pandemic and changed the course of history. Others became tangled in the same old work routines and unchanged job designs.
We call this new mix of people, processes, and platforms built to shape the future precisely as managers require precision systems. And in today’s incredibly fast-paced, competitive, unpredictable world, precision systems matter—a lot.
What This Book Is About
This book provides a new blueprint for today’s leader, from a Fortune 10 business executive to a teacher, doctor, or police officer, to convert the enormous potential of today’s digital world into powerful change, not just once, but, in the words of Tom Davenport, a global leader in the embrace of analytics, “repeatedly, reliably, and quickly.” How? With precision systems—the highly engineered working arrangements of people, processes, and technologies that put big data and AI to work creating exactly the change leaders want, exactly where, when, and how they want it.
From Rosabeth Moss Kanter to John Kotter, change gurus have focused on how to get a person, a team, or an entire organization to start acting differently, one person or group at a time. That’s important because a lot of change happens at that level. More recently, Richard H. Thaler and Cass Sunstein’s Nudge suggests ways to trigger mass change. Nudges are blunt-force tools for promoting change. They are passive, imprecise, embedded in a designed “choice architecture,” and left to run, always “on.” They wait quietly for the next one or a million unsuspecting consumers, users, or readers to stumble across their trip wires and change course. Nudges wait for you at the end-of-aisle grocery store displays, or at the child-height candy shelves at checkout, in yellow warning strips on subway platforms, or in purposefully designed defaults on websites that you either opt out of or fall slave to. Whether for broad public policy goals like climate change, obesity, and addiction, or more modest purposes like painting steps toward a trash bin to encourage its use, the power of a nudge is potent but hit or miss. You have to stumble upon it to get the ball rolling.
Precision systems are different. In their highest form, precision systems don’t wait for you to find them—they seek you out—one gene at a time or from among millions of people. Political campaigns hunt down crucial singleissue voters and peel them away from parties and party-line voting; railroads scan thousands of engines, boxcar wheels, and track miles searching for a few life-threatening flaws; health authorities desperately seek the handful of truly sick among the millions of well citizens even before the few show symptoms.
If you want to win an election, improve the health of a city or a nation, or sell vastly increased amounts of your company’s products, you may have to get millions of people or things, or even just a few, behaving differently, fast and in real time, once or continuously over days, weeks, or months—precisely as you want.
The data is there—big data—but it can be a tremendous technical and operational challenge to put that data to work for the change you want, precisely where, when, and how you want it. Giant data sets like GPT-3 for language and ImageNet for images make available billions of examples that make life easier for engineers building everything from chatbots to machines that use images to sort weeds from crops (as we describe in our discussion of John Deere’s See and Spray system). Developers don’t have to start from scratch.
But these data sets remain unwieldly, costly to use all the same, and flawed— gifted with a “stunning ability to produce human-like text,” as one reporter observed of GPT-3, but also “a powerful capacity for biases, bigotry and disinformation.”
And there’s a catch: the world is rich with events that comprise only a handful of images or words. “I’ve built AI systems of hundreds of millions of images,” AI pioneer Andrew Ng told a reporter. Ng was founder of Google’s Google Brain group, chief scientist at Baidu, and now CEO of Landing AI. “Those techniques don’t really work when you have only 50 images.”
Now new techniques championed by Ng and others can train precision models from small data—a handful of images or words. GE Global Researchers recently used only forty-five training sample images to locate kidneys in ultrasound images, for example. These novel techniques open the door to detection of thousands of rarely occurring manufacturing flaws or predictions of lowfrequency natural disasters or disease, no matter how scarce the data. In that respect, they expand the reach of precision systems even to those events that are exceedingly rare.
Big data or small, “born digital” tech titans like Facebook, Amazon, Apple, and Netflix make real-time precision look easy—practically automatic. Their AI-driven recommendation systems nudge you to the best next choice, personalized for you—or at least for people like you—millions of times each day. While by no means perfect, those highly automated precision systems are good enough, and fast enough, to have turned Amazon, Apple, and others into multi-trillion-dollar corporations. It wasn’t easy getting there; their battles of years gone by are legendary. But today, at the so-called FAANG firms— Facebook, Amazon, Apple, Netflix, and Google, everyone is a believer; the skill sets, job designs, and organization setups are long proved; and the ability to test, prove, and continuously improve automated decision making built long ago into systems and platforms, operations, and culture. The born-digitals among us still have many issues, but getting to precision isn’t one of them.
For those organizations not born digital—and most today are not—precision systems require a far different kind of approach by managers. Everywhere, they are putting precision systems to work on a smaller scale, and not just personalized but particularized to a specific business problem—to find an elusive pattern, perform a repetitive task, target a payload, or tailor a message. Depending on the nature of the business, the value proposition is entirely different each time: to accelerate mathematics learning among seventh-graders, to prevent workplace injury in mines and manufacturing, or to deepen reader engagement and increase digital subscriptions. Across these diverse missions, the leadership challenge for executive managers looks a lot like the classic work of managers everywhere who bring corporate or organization assets together for a purpose, with results and outcomes that matter. But the nature of precision changes everything, comprising as it does unique technologies, new talents, and prospects for change impossible to achieve otherwise—from the moment managers conceive the precision system, then sell it, make it work, and deliver on its promise.
That’s where Precisely comes in. This book explores how leaders in every domain are dealing with these challenges and notching wins for their organizations by taking real-time precision systems into the marketplace, the combat space, the political race. We provide insights that will help any leader who is considering the move to precision to choose the system that’s right for them, decide which problem to tackle first, sell the importance of precision to stakeholders, power up the people and the technology, accomplish change that delivers precisely what’s needed every time—and do it all responsibly.
In this first chapter, we will look at three precision systems, each vastly different but sharing essential traits common to all. The first chapter will give us an early look at some important lessons every manager can learn from.
Zipline: Delivering Precision, Saving Lives
Since the uncrewed aerial vehicles known as drones were first developed, autonomous delivery has been a dream of innovators for everything from household products to medicines. It turns out to be incredibly difficult, fraught with risk and danger, but when it happens, transforms the world.
Zipline cofounder Keller Rinaudo created the San Francisco–based company with a specific mission in mind: “To provide every human on earth with access to vital medical supplies.” One-third of the world has limited access to essential blood products, vaccines, and injectables—none more so, Rinaudo discovered, than in sub-Saharan Africa. In Tanzania, Uganda, Rwanda, Malawi, Ghana, and the Democratic Republic of the Congo, for example, stockouts and expiration rates are high, and cold-chain and short-shelf-life commodities fail
against the brutal realities of conflict, heat, and rough access. Rinaudo found the governments of these countries willing to let him experiment with delivery by Zipline. Early networking and hustling found responsive leaders. “We were meeting with the presidents of countries off of a sketch deck,” CTO Keenan Wyrobek recalled. “The market ‘pull’ was there.” By 2019, three years after its start, Zipline was delivering 75 percent of Rwanda’s blood supply to 2,500 hospitals and health facilities far from Kigali, its principal city, and operating from six distribution centers that covered 100 percent of Rwanda and 50 percent of neighboring Ghana.
Success scaled massively. Through 2022, Zipline operated a network of autonomous drones in Africa, Asia, and North America. Although each instance of success was personal, when massively scaled, it conferred vast social benefit. Thanks to Zipline, for example, an urgent order for blood from a regional hospital for a mother bleeding out during childbirth can be dispatched and arrive in as little as an hour. The drone drops the parachuted package at the destination and flies directly back to home base for reload and the next job. Zipline calls it “last-mile, on-demand emergency medical delivery.” We call it a precision system in action.
“I used to see the drones fly and think they must be mad,” Alice Mutimutuje, a Rwandan mother said. “Until the same drone saved my life.”
If the achievement is astounding, Zipline’s end-to-end supply chain is precise to a fare-thee-well. Its two Rwanda distribution centers, for example, each run eight cold-chain storage devices at different temperatures and agitation levels, each fed by three supply chains, among them the national blood bank system. Zipline updates its blood inventory each evening for the next day’s deliveries, based on forecasts that Zipline’s algorithms built with continuous improvement to 99 percent accuracy.
An optimized pick-and-pack system whittles the time from “order received” to “drone dispatched” down to five minutes on average and to as little as one minute. The drones themselves are remarkable feats of engineering; lightweight and streamlined, their frames maximize efficiency on the ground and in-flight. Every surface is tagged with a QR code for preflight inspections. Technicians point their phones at the QR codes; the plane activates the control surface, technicians image it, and algorithms predict any issues and pass or fail each control surface. Launches are tightly controlled, with each drone launched the same way every time: with an electric catapult that accelerates the plane to 110 kilometers per hour cruising speed in a quarter of a second. “That’s why we call it ‘zip,’ ” Wyrobek explains with a grin.
In flight, continuous global positioning system (GPS) signaling and radar allow each drone to “know” its own location in three-dimensional (3D) space as well as the locations of any nearby drones and planes. That’s essential for regulators, safety, and scaling into busy airspace. Zipline drones drop their
supplies by parachute, taking in terabytes of data in real time, making ballistic calculations and dynamically adjust ing for wind speed and direction so that the package drifts with the wind right where the customer wants it—no clinic needs any infrastructure to receive the shipment. Back at home base, a capture system stretches a wire across two goalpost-like towers; as the drone approaches, it communicates with radio receivers, navigates between the posts, extends its tail hook to grab the wire and powers down: mission accomplished. All without a human pilot, all guided by algorithm, the entire voyage mapped to precise three-meter by three-meter coordinates provided by What3Words— the astonishing app that has mapped the globe with 57 trillion such precise coordinates—“the national addressing system of dozens of countries,” Wryobek says. “We’d literally just get a GPS coordinate for hospitals and clinics, go visit them, and ask, ‘Hey, so where do you want this to drop?’ The future of Africa is getting solved by What3Words.”
In combination, these elements add up to a precision system dubbed by Wyrobek an “instant logistics infrastructure.” Replete with AI and physiotechnical assets managed by doctors, pharmacists, aviators, engineers, and logistics professionals, Zipline delivers exactly what a patient needs, exactly where it is needed, at exactly the right moment—and does it repeatedly, at scale, with utter reliability. It takes a system to bring health and safety precisely where and when it is needed most.
Good leadership pivots robust precision systems fast to new possibilities. In 2020, Zipline faced a new test. COVID-19 was spreading in Ghana. In its midst, a presidential election was creating an unprecedented demand for rapid delivery of health-care supplies. “Polls were opening in forty-eight hours,” Zipline reported. “Poll workers lacked masks and personal protective equipment. Tens of thousands of face masks needed to be distributed in some thirty-three districts over an area of more than nineteen thousand square kilometers.” The Zipline team mobilized, pivoting its operations from health to voting rights. Within twenty-four hours, Zipline flew over 160 drone flights to twenty-nine locations and delivered 18,000 face masks to poll workers. Hands-free drop-off minimized human contact and reduced the risk of transmission of the virus.
Pivots like this emerge unexpectedly as the value of on-demand drone deliveries becomes clearer. Zipline modifies its success metrics accordingly. For example, in its partnership with Novant, a North Carolina–based health and hospital system, Zipline claims an average reduction of $169 for patient costs by telehealth visit, an 18 percent increase in medication adherence with ondemand delivery, and a 28 percent reduction in transportation costs compared to ground transport, all with thirty times fewer carbon dioxide (CO2) emissions per mile than an electric vehicle.
This is the power of precision—not simply an incremental improvement in the way things have usually been done, but a fundamental transformation of a
delivery system, changing life on the ground precisely as Renaudo envisioned for millions, one patient, one delivery at a time. No improvement to a regional trucking network could ever produce the exponential increase in efficiency and impact that Zipline achieved. “We set out five years ago with a goal to come as close to teleportation as possible—a goal many people found crazy at the time,” he told reporters. Suddenly the way forward “to provide every human on earth with access to vital medical supplies” is within view.
Zest: Using Precision Systems to Transform the Payday Loan Market
For a different look into the power of precision systems, let’s look at another example—a small corner of the financial services industry that has been transformed by a new, precision-enabled business model revolutionary in its day.
“What would you have done if I hadn’t answered the phone?” Former Google CIO Douglas Merrill was incredulous. His sister-in-law had just called for help paying for a new tire. Mother of three, with a full-time job, and a full-time student, she was broke. “I would have taken out a payday loan,” she said.
At the time, 19 million U.S. households did the same—unbanked, at-risk, they turned to one of America’s 25,000 payday lenders to cover needs. Thirtythree percent of Americans couldn’t lay their hands on $1,000 when they needed to. They were prepared to pay payday lenders over 400 percent annual percentage rate (APR)—$15 to $30 on a $100 loan. It all added up to about $30 billion in short-term credit each year.
“I thought that wasn’t fair,” Merrill said. “Isn’t there a way to provide loans to these people that’s not so expensive?”
By now, millions of gig workers have entered the payday-borrowing market. As more and more people take up new types of work far removed from the traditional, steady, nine-to-five job, many find they need short-term cash to get through periods between gigs or while waiting for the first check to arrive from a new client. This is where access to a reliable source of credit can be a lifesaver.
Until recently, gig workers and their need for credit were sidelined, largely ignored by financial marketplaces that had no way (and not much desire) to understand them, price products and services for them, or connect the buyers with the sellers. But now the markets have discovered these gems and are unleashing their value. How? By listening to millions of new signals that turned these lumps of untouched consumer coal into diamonds.
For decades, the credit scores calculated by the agency originally known as Fair, Isaac and Company and now called FICO have largely shaped the decisions made by loan underwriters. With the invention of standardized credit
scores, lenders saw risk better and could increase both credit availability and profit. But FICO scoring was a blunt instrument, freezing out the underbanked and the unbanked—people with scant or no formal financial histories. Their “thin files” gave FICO scorers little good-caliber banking activity to go on. Risk-averse lenders hedged their bets, which meant that a bankruptcy or credit card default could ruin a consumer’s record for a lifetime.
Merrill believed that the data needed to calculate fair lending terms for underbanked customers wasn’t scarce. “All data is credit data,” Merrill said. “We just don’t know how to use it yet. If you could get access to thousands of signals, you could correct that.”
Merrill created Zest Cash and began scooping up and feeding in millions of signals from online data, building algorithms and models, helping his clients re-predict risk and make loans to the unbanked and underbanked. The more signals, the better the models, the more insight into real risk, and the higher the likelihood of payback from folks who until then were frozen into a group called the “unbankable” and thought of as “uncreditworthy.”
Some, true, were awful risks. But they were mixed in with perfectly sound risks. If Merrill could “liquefy” this frozen mass, divide them into segments, and find the gems among them, he could do for them what FICO did for everyone else: price loans according to true risk, making credit more affordable where warranted.
Sifting through all the big data Merrill could get his hands on, Zest added a bevy of small signals to those used by traditional credit scoring: whether an applicant’s Social Security number belonged to a dead person, for example; how much time an applicant spent reading the fine print on the Zest website or filling out the application. They also used contrarian signals. When customers dialed in to let them know they’d be late on a payment, banks saw risk, but Zest analyzed their subsequent behavior and discovered these folks could in fact be diligent and good bets for full repayment. The more signals that Merrill incorporated into his models, the better they predicted borrower risk and the better they could price the loans. “Not 50 percent better,” Merrill said, “just a little better. But take a lot of things that are a little better, and you get something that’s a lot better.”
Zest’s machines learned well. Incorporating classic scoring data and then adding 70,000 new variables boosted the predictive accuracy of the algorithms by 40 percent and increased net repayment rates by 90 percent. Add some further analytics by humans and Zest did even better. “The combination of really big data and human artistry is the underlying value here,” Merrill said. “We went into this business to save the underbanked billions of dollars,” Merrill told an interviewer. “We knew that applying machine learning and big data analysis to underwriting would make a real difference, and it has.”
Several key factors helped to explain Zest’s success:
• First, a vision, simple in the extreme: “Save the underbanked billions of dollars—and monetize that for profit.” That was Merrill’s North Star, guiding the enterprise through ups and downs, against a plan sized for action to test and prove his assumptions and discoveries.
• Second, hard resources, including talented people, loaded up with business acumen and technical skills. A platform and software from which engineers and data scientists could extract the data, analyze it, and make it ready for business action.
• Third, soft resources of motivation and leadership for “doing good” but also scoring big paydays. An organization led by a manager who was open to discovery and expected the unexpected, skilled at building models for new products and services that delivered on the promise.
Zest’s approach has spawned plenty of competition—and issues. It was good news for the unbanked around the world—half the world by some estimates, comprising 800 million people in South Asia alone. A similar approach, for example, has helped Mahindra Finance bring new credit-scoring methods and lower-cost loans to India’s nonbanked rural poor. By 2019, reported Harvard Business Review, Mahindra Finance had become the largest rural nonbanking financial company in India, serving 50 percent of villages and 6 million customers.
But the application of precision to the financial needs of the unbanked has also served to expand a payday loan industry whose practices and fees are still often labeled as rapacious, exploitative, and exorbitant. Better with Zest, but not great. And now more of it. Precision alone doesn’t guarantee results that are socially beneficial. Human judgment and ethical values remain as essential as ever.
The NFL: Precision Systems at Work
In the fall of 2021, ESPN.com’s Kevin Seifert wrote this:
Social media lit up on Sunday night in the moments after the Baltimore Ravens converted a fourth-down run to seal an exhilarating 36–35 victory over the Kansas City Chiefs. Amid the flurry of “WOWs,” shock emojis and celebratory GIFs was a tweet from Michael Lopez, the NFL’s director of football data and analytics. After watching Ravens coach John Harbaugh make a difficult decision from his team’s own 43-yard line, sending quarterback Lamar Jackson up the middle for two yards, Lopez posted the headshot of a man that only his
followers and a handful of other devotees to football analytics would recognize: Daniel Stern.
Lopez needed no words. Stern’s smiling child-like visage was well known to the NFL’s in-most crowd as one of a handful of twenty-something analysts bringing the rigors of data science and prediction to in-game decisions. Whether for a quick retooling of strategy between halves, or in-the-moment calls to pass or run on first down, to punt or go for it on fourth down, Stern and his cohort were breaking down the historic confines of data from the quiet pastures of recruitment and player development, coming into their own under the bright lights of game time. In the booth, headsets on, game on the line, Stern’s mission was plain: lay out win probabilities to Harbaugh for a handful of plays that would have the best chance for a score and a win.
And Harbaugh—how does he gulp down all this data? “I trust my eyes first,” he told a reporter. “The information confirms or opens your eyes to something.” He will look at it, consider, and make the call.
Come autumn in America, on any given Sunday, fortunes rise and fall for thirty-two NFL teams. It all depends on what twenty-two huge, strong men, eleven on either side of the football, each incredibly fast, all well-coached and -trained, do to get their hands on the ball sitting quietly on a white chalk line, waiting. In a moment, nothing in the world will matter more. They will claw and tackle and pound away for five or six seconds at a time (the average length of a football play), eleven minutes per game (the total minutes of actual contact), for sixty game minutes (on the clock), spread over 3½ hours. Stoked by beers and junk food, millions of fans cheer from the stands in the cold, rain, snow, and heat, or watch at desks, bars, and in man caves around the world. Billions in team revenues, media rights, and fan-wagering dollars are at stake. Competition is fierce, and every move by every player counts. For every coach, on every play, the challenge remains the same: run the data, perfect the models, predict performance, make the call, hope and pray that in the next five seconds the game changes just the way you want it.
For 100 years, NFL coaches have done all this in their heads. Now machines are having their say, loaded up with AI and machine learning, plumbed by analysts (and now offensive assistant) like Daniel Stern for the nugget that will turn “couldn’t be” to “maybe” to “probably,” all stitched together in precision systems that turn insight to calling plays on the field. The teams’ precision systems chow down on data coming from cameras, embedded radio frequency identity (RFID) chips in uniforms and the footballs themselves, and history, all filtered through models that let analysts predict and prescribe and give options to coaches. Built on the promise of gaining the winning edge, they offer deep insight to players on either side of the ball. They do it fast. And the decision “team” of coaches and analysts preps hard.
Harbaugh, his coaches, and Stern start Mondays talking through their approach to the coming game. They settle on principles and rules they’ll follow come game time—when Stern will remind Harbaugh of those guardrails in the headset, giving options. On every play, the coach has a few seconds to weigh the numbers and give thumbs up or down to the offensive coordinator’s play call.
“A lot of work goes into it,” Harbaugh told a reporter. “We always have a plan every week.”
Stern learned by watching his bosses communicate with Harbaugh during games—short, terse, to the point. They speak a similar language, making it fast, talking about win probabilities and expected points added (EPAs). The upshot: “Aggressive fourth-down calls have become a weekly occurrence for the Ravens,” reported The Athletic’s Sheil Kapadia. “They’ve gotten 10 of 14 fourthdown attempts on the season, averaging 10.5 yards per play. Eight resulted in touchdowns.”
Many debate when this modern-day data wrangling all started. Some ascribe it to Ryan Paganetti, of the Philadelphia Eagles, and peg the date not to a win with data, but to a loss without it. Paganetti was the Eagles’ analyst and game management coach for six seasons, starting in 2015. By 2017 he was sitting up in the booth during games and meeting with Coach Doug Pederson at half times. Like Stern, weeks and weeks of prep, research, and engagement with coaches and players had earned him a seat and a headset. Toward the end of an early season game against the Kansas City Chiefs, Paganetti counselled Pederson to go for a dicey two-point conversion after a touchdown, rather than a sure-bet one-point kick. The coaches game-side and on the headsets freaked out—“borderline, having a confrontation,” Paganetti told reporter Liam Fox. Pederson relented, went for the sure bet, and the game was lost.
Reflecting later, Pederson and coaches committed to Paganetti’ s analytics. “From that moment on, there was never another time that entire rest of the season where anyone questioned the information I was sharing with him,” Paganetti said. It is the stuff of NFL legend that Philadelphia thereafter won nine straight games, and the Super Bowl.
The move to precision systems in the NFL was on. With the right quarterback—a Lamar Jackson, for example, who could run and pass no matter where or when—fourth down became a “go-for-it” proposition. In 2019, the Ravens ran seventeen successful fourth-down tries, a watershed development. Being the NFL, where innovation is on display every Sunday, fast-follower “copycat” NFL teams soon made the shift to a differently-talented playmaker, too—more mobile quarterbacks like Jackson rose to the top of everyone’s must-have lists. Through the first two weeks of the 2021 season across the NFL there were 88 “go-for-it” plays on fourth down, more than at any time in league history. “Pass on first down” was gaining new currency, too.
Today, all NFL coaches are competing on precision as never before. But where “old school” competed on who had the best data, all teams now have much the same basic data available, courtesy of the National Football League’s data platform and network, dubbed the NFL Player Tracking System and its customer facing product, the NFL’s Next Gen Stats. The platform comprises sensors, receivers, and displays throughout every stadium, fueled by nickelsized RFID chips in every player’s pads and the footballs themselves, on the referees, pylons, sticks, and chains, all beaming player and ball data via in-stadium transmitters and antennas ten times per second, precisely locating individuals within inches on the field, instantly calculating their speed, separations, distance travelled, and acceleration. On every play of every game this NFL data “firehose” adds more than two hundred new data points, sent up to the cloud via Amazon Web Services and back down to fans, coaches, and players around the world, everyone putting it to their own best use.
If all teams are equal in data, a new breed of Sunday warrior makes some teams more equal than others. Each team’s “data guys” run from kibitzers and Sunday hackers to pros. Top units comprise software engineers, data scientists, data analysts, and product managers. The average ones don’t do much more than read out data from commercially available, off-the-shelf algorithms. The best of them pull the data into highly customized models that integrate data for coaches who, like Harbaugh, trust their eyes but keep their minds open. Tested, proved, and fine-tuned to conditions, NFL models are proprietary and wellguarded crown jewels of high performance. But even the best analysts must elbow their way into the booths and onto the headsets and, like any coach, earn their keep with calls that win matchups and games.
It all starts with player drafts and roster construction. Models describe the fit of prospects to the coach’s strategy, as Lamar Jackson fit John Harbaugh’s like a glove. They predict outcomes of matchups against opposing teams and players. They prescribe, in turn, the general manager’s offers of incentive and compensation packages, knowing what models say the team needs from the player to win—numbers of sacks, runs after completions, or interceptions. Those incentives drop down to weight rooms and trainers’ tables where players incorporate data for gains in strength, agility, and speed, not just one time in one game but all season long and over a career. “High caliber athletes could always do that [once],” Sports Business Journal’s Ben Fischer told an audience. “What they’re trying to figure out is, ‘How do I do that 25 times in the game, and keep that performance going through the entire season?’ ”
If player longevity is everything, safety is the key. Data drove the invention of vastly improved helmets, for example, precisely configured for the shock of collisions at full tilt, then spurred player adoptions from 40 percent to 99 percent in three years. Data informed NFL rules changes by showing certain plays
as the most injurious. It helped the NFL predict impacts on player safety from adding a proposed seventeenth game to the season.
In the stadium, precision systems have transformed the season ticket-holder experience, personalizing it seat by seat, competing successfully with the comforts of the man cave, spurring renewals even for losing teams, whether by showing unique-to-the-stadium replays-on-demand from every angle on massive screens or phones and tablets, or delivering food and drink right to the seats. Precision systems underpin the lucrative Fantasy Football industry and deepen fan engagement with competition unthinkable even five years ago. Betting on NFL games has accordingly soared, with a piece of all the action going to the NFL. In-game systems have opened new vistas for coaches and players, choosing which first- and fourth-down plays to run depending, for example, on the fatigue of opponents in late quarters measured by cameras that time players who trudge back to the sidelines, or sprint.
As new users pull the data forward in astonishing ways, some players, coaches and teams gain further advantage. Every player could, but only some do. It is that unique, distinguishing, and difficult.
Imagine, for example, if all players managed their personal development as Cooper Kupp has managed his. Kupp ran poorly in the pre-draft player showcases and went late in the 2017 draft to the Los Angeles Rams. Riven with curiosity about the science of high performance and driven by personal thirst for success, Kupp transformed his backyard tennis court into a barn and a personal football laboratory. His goal: test and prove a new science of personalized, datadriven football. His gear, laid in for the purpose: stadium-style turf, specialized treadmills, timing gates, towing units, all to measure his speed, agility, and GPS locations as given by imaging and sensors. His team: consultants who helped Kupp winnow his personal data firehose down to “one actual step and football’s most critical movement option,” Sports Illustrated’s Greg Bishop wrote, literally “the first step off the line of scrimmage.” Kupp’s plan: achieve maximum velocity five yards off the line sooner than any receiver ever had, leaving defenders behind. Train with drills and sprints “not for every day or every session, but for every rep.”
The upshot? By his fifth season with the Rams Kupp had become a “phenomenal accelerator, aggressive in his bursts,” Bishop wrote, “amassing borderline historic numbers.” Kupp was the first receiver to have over two thousand receiving yards in a season as both a receiver and a running back, on the snap of the ball “shuffling through his repertoire of movement options en route to exactly the right place, at precisely the right time,” a gift of the gods to his quarterback and coaches.
As teams and precision systems mature, powerful new consumers like the Rams’ Kupp, or the NFL’s Health and Safety Committee, or the Ravens’ coaches emerge who exert a tidal pull on data, gaining advantage from
precision, at least for a time. As novelty wears off and innovation becomes standard issue, new equipment and in-stadium infrastructure anchor the data. New business practices embed it. New models shape it. New outcomes measure and validate it.
If all teams are created (or soon become) equal on data, advantage comes from platforms, processes, and people bringing all these systems together across the enterprise, in time, continuously improving, applying precision solutions exactly where, when and how coaches want the change, achieving highest performance of the parts and the whole, at once and over time. That takes leadership with the mindset and the skills to do it—thinking of precision not as a “nice to have,” but as the cornerstone of new strategy. And in that respect, all teams are decidedly unequal.
As we have seen, some teams, like the Dallas Cowboys, were slow to embrace the power of data-driven precision. Former head coach Jason Garrett told the press in 2019 that analytics were “not valuable” to the team during the game— an opinion generally received as moronic. Garrett had no Daniel Stern sitting side by side his own coaches up in the booth. That was Garrett’s call, and he paid for it, being fired twice from NFL coaching jobs since.
In 2019, after an abysmal 4–12 season, David Gettleman, the New York Giants’ general manager finally conceded the game begrudgingly to analytics, at least in the back office. “We have hired four computer folks, software, and we are completely redoing the back end of our college and pro scouting systems.” Nothing about in-game management, however. In fact, that same year Gettleman hired Jason Garrett as the Giants’ new offensive coordinator, then fired him as the Cowboys had, and then lost his own job after the Giants went 4–13 on the season.
Though the times are changing, any coach favoring analytics still must have some steel. Played right, on average, he’ll do better following the numbers, but on any given play or game he could lose. Then the “dumb and I’m proud” crowd will descend. When Brandon Staley’s LA Chargers “went for it” on five fourth downs but converted just two against the Kansas City Chiefs, and lost in overtime, Howie Long and Terry Bradshaw went after him on Fox. “Neither of us can spell analytics,” Long said, “but it took a beating tonight.”
By the end of the 2021 season Staley had redeemed himself rising to the top of FootballOutsiders.com’s new “Aggressiveness Index,” passing the Ravens’ Harbaugh on the way, who fell to number sixteen. In his own rearview mirror Staley could see upstarts closing fast, like the Indianapolis Colts’ head coach Frank Reich as he stepped over the still-warm football corpse of NFL workhorse coach Bill Belichick. In their game 15 matchup, Reich called for three fourth-down tries, making but two, but ultimately exhausted Belichick’s Patriots and their options. Meanwhile, with nine minutes to go in the game Belichick settled for a three-point field goal rather than
attempt a fourth-and-goal conversion for six points. Final score: Colts 27, Patriots 17.
“It’s crazy how much more the Patriots would win if he made three to four better decisions each game,” a league source told Stephen Holder of The Athletic. “Despite being perhaps the greatest strategist in the game’s history,” Holder wrote, “Belichick still has a sizeable blind spot when it comes to analytic decisions.”
Being known as a numbers-driven coach can create perverse new disadvantage, which other coaches may seize and win on. In a late season matchup between the Pittsburgh Steelers and the Ravens in 2021, Harbaugh and Jackson failed at a game-winning two-point conversion. They went home losers. Steelers head coach Mike Tomlin had studied Harbaugh’s game film and was not at all surprised by the Ravens going for the win. “They aggressively play analytics,” Tomlin said. “From that standpoint, they're predictable.”
As for Tomlin, a Super Bowl Champ, and the only man in history to not have a losing season for 15 straight years (and Zach Tumin’s favorite NFL coach), he was last on the FootballOutsiders.com “Aggressiveness Index.” And still he was finding a way to win.
Why We Wrote This Book
In this first chapter, we have book-ended an extraordinary decade with accounts like Zest, one of the earliest commercial exploitations of big data primed for market entry; the sophisticated, dexterous use of precision systems for change in Zipline; and the change brought to the venerable NFL. By 2022, as Kara Swisher had predicted, digitization was practically a commodity. As Doug Merrill blazed the trail by liquefying (a term coined by venture capitalist Marc Andreessen) credit data, a venturesome few armed with dazzling technologies liquefied whole markets and spawned some of the richest and most powerful new firms on the planet—seemingly lifting not a finger in the physical world, providing platforms rich with AI and machine learning, and designed and serviced by brilliant engineers, designers, and product teams. These platforms connect billions of buyers and sellers. “Uber, the world’s largest taxi company, owns no vehicles,” marketing guru Tom Goodwin wrote. “Facebook, the world’s most popular media owner, creates no content. Alibaba, the most valuable retailer, has no inventory. And Airbnb, the world’s largest accommodation provider, owns no real estate.” For good measure, add Shein—“the biggest fast-fashion retailer in the United States, with no stores,” says Benedict Evans. That’s the power and allure of precision systems putting product and service into the hands of billions of new customers for their consideration and choice. But if everyone everywhere has gone digital, what’s making the difference
between winners and losers in the marketplace, the political race, the battle space? It’s the ability to turn all that digital infrastructure into real-world value where people will pay premiums to get the change they want, where, when, and how they want it—an Uber rolling up to a fare on a lonely city street, same-day drone delivery of medicine to a rural hospital, or a vital payday loan in the urban core. Every solution involves a precision system built to deliver change exactly—precisely—where, when, and how customers want it and managers intend it. If we look, we can we see such precision systems in action everywhere:
• Precision marketing campaigns uniquely customize offers to millions of end users (voters, shoppers, travelers).
• Precision pattern search discovers hidden links across real-world events (crimes, financial irregularities, cyberattacks).
• Precision customer shaping suggests the products or services that a target consumer may want to see or buy next (films, music, clothing).
• Precision sensor systems monitor the status of machine and human activities via the networked Internet of Things (train and track integrity, oil pipeline safety, health and fitness wearables).
• Precision product and services engineers personalized offerings to suit customers’ tastes and requirements (athletic gear, automobiles, medical devices).
• Precision matching finds best fits to produce specific desired results (matching buyers to sellers, faces to identities, terrorists to locations, sentences to offenders, job candidates to jobs, curricula to students, football plays to game conditions).
• Precision scheduling smooths the operations of computer-assisted physical systems (supply chain flows, airline scheduling, and hospital bed utilization).
We’ve looked at hundreds of such efforts and spoken with hundreds of executives, managers, and teams, from New York–Presbyterian Hospital and BNSF Railways to the New York Times and the Biden 2020 campaign. Each used precision systems for their own purposes—whether to boost subscriptions and revenues, win games and elections, speed freight and passengers on their way, or improve health and save lives.
In each instance, we see precision systems comprising a range of techniques and an experimental approach that can be applied in diverse settings. Precision, like math, fits everywhere and can be used for everything. Like math, precision is a collection of techniques, facts, and methods that can be used to answer all sorts of questions—to describe the current world, model and predict the future, prescribe the best step forward.
As this perspective unfolded through our research, we learned a lot about the amazing results that precision systems make possible—the power to create just the change managers want, where, when, and how they want it. That’s
power for good, when used wisely. We’ve also discovered just how difficult precision systems are to design and implement well, and the risks of doing it badly.
We wrote this book to help managers be successful facing this new onslaught of possibilities and demands. We share what we learned from those in the forefront of precision by studying their tools, techniques, habits, and best moves forward. No matter what kind of organization you may lead or manage, you’ll find insights here that can help you see how precision can enable you to do your work better, as well as nuts-and-bolts guidance into the details of making it happen.
What You Can Expect from This Book
Over the next ten chapters, we’ll unpack the technologies, people, and strategies of precision systems to yield lessons that any executive or manager, coach or campaign chief can use right now to get ahead and stay ahead.
Chapter 2, “Six Kinds of Precision Systems,” homes in on six precision designs and their essential attributes, technical requirements, and fit to a range of business problems. It explains how the designs work, when to use them, and how to determine what kind of problems your organization currently faces.
Chapter 3, “Design, Test, and Prove: Key Steps in Building a Precision System,” brings you inside precision operations. You’ll read about and learn from managers as they test whether have found the right audiences for their campaigns, whether their pattern finding works, whether their recommendations are relevant. You’ll learn how to determine what counts as a win at the outset— big enough to matter, small enough to work.
Chapter 4, “Getting Started with Data and Platforms,” helps you decide which business problems are best suited for precision solutions and how to prioritize opportunities to get the most value possible into the right hands as soon as possible.
Chapter 5, “The Precision Vanguard,” explains the roles that organizational leaders play in driving the conversion to precision.
Chapter 6, “The Political Management of Precision,” describes the challenges and headwinds facing precision, from status quo bias to the endowment effect. You’ll learn how successful leaders have found ways to overcome these barriers.
Chapter 7, “Bootstrapping Precision,” examines the kinds of collaboration needed to bring precision to life. You’ll learn about the external and internal partnerships that precision design requires, as well as the critical issues of governance for collaboration where traditional lines of authority will be missing.
Chapter 8, “Creating and Managing Teams of Precision,” offers insights into the special challenges involved in organizing and leading the kinds of interdisciplinary, cross-functional teams that are needed to design and implement a winning precision system
Chapter 9, “Building the Precision Enterprise,” examines the best organizational architectures for precision. We organize the discussion around the critical tasks that precision requires, the formal structures needed to support it, the essential skill sets and work design that make it happen, and the cultural norms and values needed to insert new precision systems into existing business processes.
Chapter 10, “Red Zones of Precision,” examines some of the principal technical, ethical, and operating risks that the leaders of any precision initiative must address. It illustrates specific ways of measuring and minimizing these risks.
Chapter 11, “Summing Up: Some Final Lessons for Leaders,” summarizes the most important takeaways from the book. You’ll learn about the three key leadership roles involved in launching precision systems. The admiral creates a strategy for developing a portfolio of precision projects. The orchestra leader manages a group of self-sufficient professionals who must harmonize their efforts in implementing precision. The pit boss guides precision projects past implementation barriers and pitfalls to reach the end-point delivery of value.
Throughout the book, you’ll be introduced to organizations from many industries and economic sectors that have made the big leap to precision systems, overcoming obstacles and challenges along the way. Their stories will help you recognize the enormous opportunities that precision offers your organization—and they’ll likely inspire you to begin or accelerate your own journey of exploration into the amazing new world of precision systems.
“Through compelling case studies from the likes of the New York Times, Precisely takes the reader
Precisely
“Tumin and Want take us on a whirlwind tour of all the ways systems premised on