The Marketing You Never Saw

We’ve all seen more advertisements than sunsets. And every one of them was made by a company that already had traction.

Which means there is a kind of marketing most of us have never seen. Every company with traction once had none, and its first customers were won with messages addressed to a select few, hidden in plain sight.

That small audience who saw those messages, and the even smaller group who acted on them, had little in common with the mass of users who followed. Sixty years of adoption research describes them consistently: they accept risks later users won’t touch, they care about the problem more than any brand, and usually they’ve already cobbled together a homemade answer, because waiting was worse than building. Later users buy when they see other people buying.¹

That’s why founders make the mistake of marketing to an audience that doesn’t exist. Everyone who figures that out has a shot at connecting with their seed users.

Drew Houston began in 2007 with a product no one cared about. His first customers were the people who already hated the problem as much as he did, emailing files to themselves at midnight, carrying thumb drives everywhere. You don’t find people like that through targeting. You find them where they hang out, and in 2007 that meant Digg and Hacker News. So he posted a three-minute demo there, laced with jokes only those readers would get. He couldn’t show that other people used Dropbox; they didn’t. He could show he was one of them. The waitlist went from 5,000 to 75,000 overnight. Those 75,000 weren’t revenue. They were proof, the thing his next audience would ask to see.

Line up every customer your product will ever have, in the order they say yes. That line has a shape, and it was mapped decades before you started selling.

Everett Rogers, who grew up watching his father resist hybrid seed corn until the drought of 1936 settled the argument, spent his career measuring how new things spread, through farms, hospitals, and villages. Everywhere he looked he found the same picture: a few people arrive very early, a broad middle arrives together, and a stubborn tail arrives late or never. Graphed, the line makes a bell. Height makes the same shape, and so do test scores; most people pile up near the average, and the counts thin out evenly toward both ends.

And the bell is predictable. Mark off distance from the middle in equal steps, and each step fences off a known share of the whole. Statisticians call the steps standard deviations, SD for short, but the shares are what matter.

Two honest notes: the fences are drawn for convenience, since the curve itself is smooth, and the counts describe eventual adopters, not everyone alive.²

The last column, read top to bottom, is the whole model: the standard of proof rises as you move right. Seed users accept a promise. Opinion leaders need the product to work, because a public endorsement spends their credibility, and they punish anyone who wastes it. The early mainstream needs to see people they respect already inside, and they are harder to move than founders expect: buyers overvalue what they already have by about a factor of three, and builders overvalue what they’ve built by about the same, a nine-fold gap between your excitement and theirs.³

One definition matters here. Opinion leaders means what the original 1955 research meant: the relative people consult before buying, the coworker who always finds things first.⁴ They exist in every social class and influence sideways, among people like themselves. The term has never meant professional influencers. And the research on who actually drives cascades is more contested than most marketing advice admits: influentials are less important than the structure of the network they sit in. But the practical implication is the same: winning the earliest adopters is not the same as winning the people they talk to.

The curve is not a scoreboard, and adoption is not a verdict on the product. A health worker named Nelida once spent two years in a Peruvian village of two hundred families, persuading them to boil their drinking water, with medicine and the truth on her side. Eleven families said yes.⁵

This chapter stays on the left half of the map. And the map makes a prediction: if each band waits for its proof, then skipping bands should fail for anyone who tries it, at any budget.

Nike’s marketing began in 1964, Phil Knight selling Japanese running shoes out of the trunk of his green Plymouth Valiant at track meets across the Pacific Northwest. “Just Do It” arrived twenty-four years later. Mailchimp ran for thirteen years before the podcast ads everyone remembers. Dollar Shave Club’s launch video preceded its Super Bowl commercial by four.⁶ The marketing that built these companies and the marketing everyone can name are separated by years, sometimes decades.

When a company tries to skip those years, it looks like Quibi. Jeffrey Katzenberg and Meg Whitman raised $1.75 billion and launched a streaming service in April 2020 with marketing built for the later mainstream: a Super Bowl commercial before launch, celebrity shows, a campaign sized for a crowd. There was no crowd. No seed users had vouched, no opinion leaders had spent credibility, and the audience the campaign reached wanted proof the campaign could not carry. Quibi offered ninety-day free trials; when they expired, ninety-two percent walked, and the founders announced the shutdown six and a half months after launch.⁷ The pandemic and the product took their share of blame, and deserved some; neither explains an audience that showed up, sampled for ninety days, and left.

Google made the opposite mistake with Wave and got the same result. Wave was a collaboration tool, launched in 2009 behind 100,000 scarce invitations, the trick that had built Gmail’s mystique. But Gmail is useful alone and a collaboration tool is not. The lucky few got in, found nobody to collaborate with, and left. Wave had other flaws, but the empty rooms meant nobody stayed long enough to find them. Within a year of launch, Google stopped development.⁸

Neither team was careless, and that is the point. One had run movie studios; the other owned the largest distribution machine on earth. They concluded what everyone concludes, because all the marketing anyone has ever seen begins after traction.

The rule runs in the other direction too. When Post-it Notes flopped in four test markets in 1977, 3M didn’t buy more advertising. It flew teams to Boise and buried the city’s offices in free samples, and mailed pads to the secretaries of Fortune 500 chief executives; ninety percent of the people who tried them said they would buy, double the best launch in company history, from a seventy-year-old conglomerate running a seed round.⁹

The first round can’t be inherited. It has to be designed, and designing it comes down to five questions.

Every marketing plan answers five questions: who, where, what, how, and what for. Marketing built for traction answers them one way. Seed users force different answers on all five.

Traction answers who with a market: a demographic, a persona, a size. Founders describe their early audience the same way, only smaller, and the research says it can’t be described that way at all.

The urge to adopt early is domain-specific. The same person who lines up for a new phone might be the last to try a new food. The general trait, measured directly, barely predicts adoption. Age and income predict it better, and neither predicts it well.¹⁰ Seed users are recognized by behavior, and the behavior comes in combinations. Some arrive through pain: the workaround built, the forum thread started, the three rivals churned through. Studied formally as lead users, these are people who face a need before the market does and have often built some version of the solution themselves.¹¹ Others arrive with no pain point at all, enthusiasts with a standing appetite for whatever is new inside the window they watch, and for them the risk barely registers, because nothing they depend on is at stake. The constant is the window: a domain that matters enough to keep watching.

Pinterest went nine months without reaching ten thousand users, and the ones who stayed were not the early-adopter crowd anyone would have targeted. They were collectors and planners, some recruited straight from Ben Silbermann’s Iowa address book, invisible to every demographic filter and visible only by their behavior.¹²

Traction answers where with channels: which media, how much reach. A domain-specific obsession doesn’t live in a channel. It becomes visible only where the domain gathers, because the behaviors that mark a seed user are things people do among their own. Even the search method is different: researchers hunting rare lead users found that referral chains, each expert pointing to someone more expert, reach the right person after talking to a small fraction of the people that screening a population requires.¹³

Discord launched into one room. On the day the company counts as its launch, a friend of the founders posted in the Final Fantasy XIV subreddit, asking what people thought of a new voice app, with a link to a server where the founders were waiting to talk. Fifty users arrived that day, a hundred the next, and the play ran for six months.¹⁴ RXBAR ran the physical version, walking protein bars into CrossFit gyms one at a time, because everyone who cared that much about an ingredient label was standing in the same room.¹⁵

Traction answers what with the crowd itself: testimonials, logos, the visible weight of everyone else. Before traction that evidence doesn’t exist, and seed users are the one audience that never asks for it.

What they accept instead is fluency, proof that you hold their problem the way they hold it, in their vocabulary, at their level of detail. A promise becomes admissible when the promise demonstrates comprehension.

Polish splits in two here. Competence helps everywhere; in crowdfunding, the closest thing academia has to a seed-user market, sloppy campaigns with spelling errors reliably die.¹⁶ But broadcast gloss works against the sale, because expensive production is itself a signal, and what it signals is that a crowd was the intended reader.¹⁷ A thirty-second spot produced by a creative agency signals a budget, which signals a crowd, which signals that the intended reader is someone else. The seed user knows they are not that someone.

Asking for complaints proves exactly what a seed user needs proven: that the truth about the problem matters to you more than the sale.

Slack’s first campaign was Stewart Butterfield asking friends at six to ten companies to use the product and complain about it. When Slack opened its preview in August 2013, 8,000 companies signed up in the first twenty-four hours.¹⁸

Traction answers how with broadcast, because there is no other way to address a crowd. The first round runs on conversation, and not because scrappiness is a virtue: fluency and responsiveness are the evidence, and they can only travel through exchanges that can be answered. Diffusion research sorted the channels out long ago. Mass media excels at creating awareness; interpersonal exchange is what changes minds.¹⁹ The first round lives entirely in the second job.

When Zoom was young, Eric Yuan personally emailed every customer who canceled. One accused him of being an autoresponder wearing the CEO’s name, so Yuan offered to prove it on a Zoom call.²⁰ Fifty years earlier, Nike’s first employee ran the same play on index cards, one per customer, shoe sizes and race results, encouragement mailed before big meets.²¹ The medium changes; the mechanism doesn’t.

Traction answers what for with revenue. The first round’s real output is proof: the named references, the retention, the witnesses the next band will ask to see before it moves. The first customers are nearly a byproduct of that production line.

Diffusion models measure the two forces separately, the pull of marketing and the pull of other people’s adoption, and across hundreds of studied products the second outweighs the first, on average, by roughly thirteen to one.²² Proof is the only asset that compounds, because each band’s standard is the threshold the next band waits behind.

Tinder had fewer than five thousand users when Whitney Wolfe left on a tour of college campuses. She pitched a sorority chapter, got the room to install the app, then walked to the brother fraternity, where the men opened it onto a screen of women they already knew. Each side was the other’s proof, manufactured an hour apart. She came back to fifteen thousand users and an avalanche already moving.²³

Those five answers, taken in order, are the first marketing plan: find the people who were paying attention before you arrived, go where they gather, prove you understand, send messages that can survive being answered, and collect the evidence the next band will ask to see. A company that can’t yet name its first hundred customers is still left of the second step, and this is its plan.

It runs on almost no money, because conversation is nearly free; Butterfield’s campaign cost him some pride, and Houston’s cost a weekend. What the plan consumes is attention and the willingness to be answered.

Obscurity is the other advantage. Nobody upstream is watching, which makes this the one stage where experimentation costs almost nothing: run three campaigns at once, try messages that contradict each other, skip the polished brand, because the audience that might judge any of it won’t arrive for years. Founders hold back anyway, afraid of poisoning a well they haven’t dug yet. The fear is measurably irrational: people overestimate how much others notice and remember about them,²⁴ and a company with no traction is overestimating from zero. The customers who will matter in three years can’t remember a message they never received.

A shot, though, not a guarantee. Identical products meet different fates because early adoption amplifies chance; in the best-known experiment on the question, the same songs became hits in one artificial market and flops in the next.²⁵ How much of the rest is luck, and what can be built around it, is another chapter’s territory.