Why good ideas don't spread evenly — and the "chasm" that quietly kills most new products, tools, and habits before they ever reach the mainstream.
When something new shows up — a product, a tool, a habit, an idea — it doesn't get adopted all at once. It moves through a population in a predictable shape: a few people jump on it immediately, a slightly larger group follows once they see it work, a big middle chunk waits for proof, and a sizable tail resists until they have no choice.
Sociologist Everett Rogers mapped this into five adopter groups in his 1962 book Diffusion of Innovations. Nearly thirty years later, consultant Geoffrey Moore pointed out the single most important thing about that curve for anyone building something new: there's a gap — a chasm — between the early believers and the cautious middle, and most new products die trying to cross it.
Each group adopts for a different reason and needs a different kind of proof. Tap any bar to see who they are and what actually convinces them.
You're launching, pitching, or rolling out something genuinely new and need to know who to court first, what will stall adoption, and how to read where you actually are in the rollout — not where you hope you are.
Adoption isn't voluntary (a legal mandate, a compliance deadline), the decision is a single one-off choice rather than a population spreading over time, or the "innovation" is really just a minor tweak nobody needs to be convinced of.
"Who has actually adopted this because they needed it to work — not because they liked being first?"
Everett Rogers, Diffusion of Innovations (1962), building on earlier agricultural-adoption research. Extended by Geoffrey Moore in Crossing the Chasm (1991) for technology products specifically.
Each group isn't just "faster" or "slower" at adopting — they're convinced by genuinely different evidence. Click through them in order.
A mid-size company wants every team using a new AI writing assistant within six months. Here's how the curve actually plays out, step by step.
Rogers introduced the five categories and the S-curve shape in 1962, drawing heavily on earlier studies of how Iowa farmers adopted hybrid seed corn — a slow-moving, clearly observable innovation where the data held up well. The percentages above (2.5% / 13.5% / 34% / 34% / 16%) come from treating adoption as a bell curve split at standard-deviation points; they're a useful reference shape, not a law of nature — real products diffuse faster, slower, or lopsided depending on the category.
Geoffrey Moore's 1991 contribution — the chasm — came from watching 1980s and 1990s enterprise tech launches succeed or quietly die between the early and mainstream markets, and it's the part of the model most cited in product and go-to-market circles today.
Rogers himself later acknowledged what researchers call a pro-innovation bias in the theory: it was built to explain why things spread, which can tempt people into assuming every innovation should spread, or that faster adoption is always better. The model describes the shape of adoption; it doesn't tell you whether the thing being adopted is actually good for the people adopting it.
2.5% / 13.5% / 34% / 34% / 16% is a reference shape from Rogers' original bell-curve split, not a measured constant. Real adoption curves are lumpier and shift by category, culture, and price.
"Revolutionary" and "cutting-edge" excite Innovators and scare off the Early Majority, who are listening for "proven," "supported," and "everyone like me already uses this."
A handful of vocal early adopters loving something is real signal about them — not proof the mainstream will follow. The chasm exists precisely because that inference fails.
Early adopters will tolerate a rough, DIY experience. The Early Majority won't adopt a bare feature — they need support, documentation, integration, and a safety net around it.
The same shape governs internal tool rollouts, new management practices, public health behaviors, and AI adoption inside a team — anywhere a population, not an individual, has to move together.
The model explains spread; it doesn't certify merit. Plenty of things diffuse widely that shouldn't, and plenty of good ideas stall for reasons unrelated to quality.
If a regulation, a legal deadline, or a contractual requirement forces adoption, there's no voluntary "early vs. late" behavior to model — everyone moves on the same forced timeline.
This is a population-spread model. If you're weighing one personal decision once (not tracking how a group adopts over time), a decision-under-uncertainty model fits better than an adoption-curve one.
Some tools are correctly niche forever — built for a small expert audience by design. Forcing a chasm-crossing strategy onto something that was never meant to mainstream wastes effort chasing an early majority that doesn't exist.
Fill in your own rollout or launch, and get a short diagnostic prompt you can think through yourself — or hand to an AI assistant for a second opinion.