Akio Morita had ordered a first run of 30,000 Walkmans, over doubts inside Sony, where some said a tape player that couldn’t record would never sell. At the time, selling 15,000 tape recorders a month was considered good.
The launch on June 22, 1979 put reporters on a bus from Ginza to Yoyogi Park, handed each of them a Walkman and told them to press play. Most newspapers ignored it. The Walkman went on sale on July 1 at 33,000 yen. By the end of July, about 3,000 had sold.
So Sony’s own staff took it to the street. Advertising and sales people rode Tokyo’s Yamanote loop line all day wearing Walkmans. On Sundays they walked through Shinjuku and Ginza, stopping strangers and asking them to listen.
People looked puzzled. Then they put the headphones on.
Seen first, then tried
By the end of August, all 30,000 had sold out. Sony’s history credits the street campaign alongside magazine ads with pop idols, so the walk-ups weren’t the only push. But they show the pattern. New things often spread slowly at first, because most people wait to see someone try it, then want a go themselves.
The sociologist Everett Rogers spent his career collecting studies of how innovations spread. In his synthesis, five perceived qualities matter: relative advantage over what people have, compatibility with how they already work, low complexity, trialability (you can try it without committing) and observability (others can see it and its results). A stranger on the Yamanote line in headphones was observability. Someone handing you the headphones was trialability.
Those two aren’t always the ones that count. In a 2009 review, J. W. Dearing wrote that in Rogers’ synthesis, advantage, simplicity and compatibility explain adoption most consistently, while observability and trialability matter less consistently, with trialability likely to count more for risky, expensive innovations. A 33,000-yen gadget nobody had heard of was a risky buy.
Two towns in Iowa
Rogers built on a study of corn. Bryce Ryan and Neal Gross, sociologists at Iowa State College, traced how hybrid seed spread through two farming communities. Not one farm operator there had adopted it in 1926. By 1941 every commercial operator was planting it, and the authors judged its superiority “great” since at least the early 1930s.
For many, hearing about it wasn’t the holdup. Among the 257 farmers analyzed, the latest adopters heard of the seed only about three years after the earliest, yet adopted more than seven years later. Salesmen were the most common first source, but asked what most influenced them, more named neighbors, 45.5 percent, than anything else. “I just figured I’d let the neighbors try it first,” one said.
Then they ran their own trials. Very few planted all their corn to hybrid in the first season. Farmers who started in 1939, with years of neighbors’ results to look at, still put a median of only 30 percent of their corn acreage into it. Most operators, the authors wrote, “insist upon personal experimentation before complete acceptance.”
Where the 2.5 percent comes from
The famous chart slices a bell curve: innovators 2.5 percent, early adopters 13.5, early majority 34, late majority 34, laggards 16. Those numbers appear in a 1961 extension bulletin Rogers co-wrote.
They aren’t a finding about people. Rogers defined the categories by distance from the average adoption time, in standard deviations: innovators more than two earlier than average, early adopters between one and two, and so on. Cut any normal curve at those points and you get roughly those shares, whatever the innovation. The 1961 bulletin itself warned that farmers in a small area may not fall into these proportions, and Rogers called the categories “ideal types.”
He also named a bias in his field. Most diffusion research, he wrote in 1975, “assumes the innovations studied are ‘good’ and should be adopted by everyone.” Hybrid corn paid off for every farmer in the Iowa study, but most other innovations studied lacked that advantage: “Many individuals, for their own good, should not adopt them.” A 1980 paper by Shankariah Chamala, A. W. van den Ban and Niels Röling added that labels like “laggard” blamed farmers’ psychology and ignored credit, land and access to information.
Geoffrey Moore later argued that the curve can stall in a gap between early adopters and the mainstream: the chasm.
The email that moved one person
Imagine a team lead who builds a checklist that catches the billing errors her team keeps fixing by hand. She sends a long email explaining why it’s better. Two weeks later, one person uses it.
The next convert is a colleague at the next desk who watches her close the month’s invoices with it, then asks to try it on one client. The email argued advantage. The desk made it visible and safe to try. And the holdout may have a client the checklist doesn’t fit, which is worth hearing before anyone calls him a laggard.
Faster than the PC
Alexander Bick, Adam Blandin and David Deming surveyed more than 10,000 US adults aged 18 to 64 in late 2024. In 1984, three years after the IBM PC, 25 percent of workers used a computer for their job. In 2024, two years after ChatGPT, 27 percent used generative AI for theirs. Counting home use, adoption outpaced both PCs and the internet.
Their suggested reason reads like Rogers’ list: cost and ease of use, since ChatGPT is free or cheap and needs no technical expertise.
These are self-reports from online panels, and using a tool isn’t the same as it helping. The authors estimate generative AI touched only 1 to 5 percent of US work hours. A rising curve counts adopters. Rogers’ farmers had a seed that paid every one of them, and most new things aren’t hybrid corn.