A Sip Is Not a Can
Coca-Cola ran the largest consumer research project in history, got a clear answer, and acted on it. Seventy-nine days later the company surrendered to its own customers on live television. Every organization that runs on metrics is one unexamined assumption away from the same year.
On the morning of April 23, 1985, at the Vivian Beaumont Theater in Lincoln Center, Roberto Goizueta stepped in front of two hundred journalists to announce the most confident decision of his life. Goizueta was not a reckless man. He was a Cuban-born chemical engineer who had fled Castro’s revolution with little more than his education, climbed the Coca-Cola Company over three decades, and now ran it with an engineer’s faith in evidence. And the evidence, that morning, was a fortress. His company had spent four million dollars and four years on Project Kansas, running blind taste tests on nearly two hundred thousand people across the United States and Canada, and the results kept saying the same intolerable thing: people preferred Pepsi. Sweeter, smoother, they chose it again and again in the mall taste-offs that Pepsi had weaponized into television commercials. But the lab had an answer. A reformulated Coke, softer and sweeter, beat Pepsi in the blind sips, and beat original Coca-Cola decisively. So Goizueta did what the data demanded. He announced that the company was changing the formula, the sacred, vault-kept, ninety-nine-year-old formula, and he called the change the surest move the company had ever made, because by every measurable standard it was.
What happened next has entered folklore, but the tempo of it still astonishes. Within days the company’s consumer hotline, which normally fielded a few hundred calls a day, was drowning; at the peak, more than eight thousand calls a day, nearly all of them variations on grief and betrayal. Letters arrived addressed to “Chief Dodo, The Coca-Cola Company.” A retired real estate man in Seattle named Gay Mullins founded an organization called Old Cola Drinkers of America, printed protest buttons, filed a lawsuit, and became a national media fixture despite failing, in blind tests conducted by journalists, to distinguish the two colas himself, a detail I find more illuminating than his entire campaign. People hoarded cases of the old formula in basements. Fidel Castro, of all people, publicly called New Coke a symptom of American capitalist decadence, which must have stung Goizueta doubly. And on July 11, seventy-nine days after Lincoln Center, the company gave up. Two executives stood before the cameras and announced the return of the original formula as Coca-Cola Classic. ABC’s Peter Jennings broke into General Hospital to deliver the news. Senators announced it on the floor. The stock closed up. Within months, Classic was outselling both New Coke and Pepsi, and Coca-Cola exited its worst year with momentum its marketers had not seen in a decade, which is why conspiracy theorists insisted the whole thing had been staged. Company president Donald Keough gave the only answer worth keeping: some critics say we planned it, some say we blundered into it, and the truth is we are not that dumb, and we are not that smart.
The autopsy matters more than the corpse, because the research was not sloppy. It was the best consumer research money could buy, and it failed anyway, in two nested ways that every operator of every dashboard should be able to recite. The first failure was mechanical: a sip is not a can. In a mall, taking one cold mouthful from a small cup, human beings reliably prefer the sweeter liquid; sweetness wins sprints. Over a whole can, a whole meal, a whole case, that same sweetness turns cloying, and preference quietly inverts. The company had measured the product in units it is never actually consumed in, like judging marriages by first dates, and two hundred thousand replications of the wrong unit do not correct the wrongness; they laminate it. Sample size is a treacherous comfort. It shrinks the error bars around your answer while saying absolutely nothing about whether you asked the right question, and watching the confidence interval tighten around a wrong answer is perhaps the most dangerous pleasure in all of quantitative life.
The second failure was the deep one, and it is the reason this story belongs to everyone who manages by metrics and not just to soda historians. In every blind test, the liquid was unlabeled. But nobody on earth drinks unlabeled Coca-Cola. They drink a red can that sat in the cooler at their grandfather’s fishing cabin, that came out at Christmas, that shipped with soldiers to two wars and appeared in Norman Rockwell paintings; the company itself had spent a century and billions of dollars fusing that liquid to the American identity, and then designed a research program that carefully filtered its own greatest asset out of the measurement. The tests removed the brand to be objective, and the brand was the product. When the company’s own psychologists warned that some fraction of the population would react to the formula change as a small bereavement, the warning appeared in no cell of the preference spreadsheets, so it weighed nothing. The numbers were precise about taste and silent about meaning, and the executives, trained to trust precision, mistook the silence for absence, which is the exact mechanical error, it seems to me, behind most metric-driven disasters since: the dashboard measures what is easy to instrument, the organization optimizes what the dashboard shows, and the unmeasured thing, trust, loyalty, love, safety, sits invisible in the blind spot until the day it detonates.
Once you carry this pattern around, you see its descendants everywhere, wearing modern clothes. Engagement metrics are sip tests: they measure the mouthful, the click, the session, and are structurally mute about whether anyone would mourn the product, or is quietly coming to resent it over the whole can of a year. Standardized test scores are sip tests of learning. Quarterly earnings are sip tests of durability. A/B tests, the direct heirs of Project Kansas, will tell you with immaculate statistical hygiene which button color wins the afternoon, and nothing whatsoever about which accumulated experience makes users one day switch with the special fury of the betrayed. Goodhart’s law says that when a measure becomes a target it ceases to be a good measure; New Coke teaches the quieter prior law, which is that every measure was born a proxy, and a proxy is a servant that answers only the question it was given, in the units it was given, forever, no matter how large the sample grows.
The discipline this story leaves behind is not to distrust data; Goizueta’s successors did not stop testing, and neither should you. The discipline is to interrogate the harness before admiring the horse. For any number that is about to drive a decision, ask three questions in order. In what unit was this measured, and is that the unit in which reality is actually consumed? What did the measurement deliberately strip away in the name of objectivity, and is the stripped thing possibly the whole point? And who warned about this in words rather than numbers, and were they weighted at zero because words do not fit in spreadsheets? The people of 1985 did not want a better-tasting cola; they wanted their cola, and they had told the company so for ninety-nine years in the only metric that never made it to Lincoln Center, the metric of coming back tomorrow. A sip is not a can. A can is not a habit. A habit is not love. Each level up the ladder is harder to measure and worth more, and the entire art of judgment, in business and frankly outside it, is refusing to let the measurable rungs convince you the ladder ends where your instruments do.
Notes & sources: Project Kansas, the roughly 190,000–200,000 taste tests, and the four-million-dollar research budget are documented in Thomas Oliver’s The Real Coke, The Real Story (1986) and Mark Pendergrast’s For God, Country and Coca-Cola (1993); the April 23, 1985 Lincoln Center announcement, the eight-thousand-calls-a-day hotline figure, Gay Mullins and the Old Cola Drinkers of America, and the July 11 return of Classic are from contemporaneous AP, New York Times, and Time coverage; Donald Keough’s “not that dumb, not that smart” remark is from the Classic announcement press events; the sip-versus-whole-can critique was popularized by Malcolm Gladwell in Blink (2005). Image: Wikimedia Commons, CC0.