Everyone knows the picture: the slow start, the takeoff, the S-shaped curve as a new product sweeps a market. The model behind it is genuinely useful for forecasting. But the story most marketers tell about it, find the special "early adopters" and "influencers" and the rest will follow, has largely failed the evidence.
The foundational picture is Rogers's, and it has earned its longevity. Plot the cumulative number of people who have adopted a new product or practice over time and you get an S-curve: slow at first, then a steep takeoff as adoption feeds on itself, then a levelling as the market saturates. Slice the population by when they adopt and Rogers labelled five groups, the innovators (the first, roughly 2.5%), early adopters (the next 13.5%), the early and late majority (34% each), and the laggards (the last 16%). And he identified five perceived attributes of an innovation that predict how fast it spreads: relative advantage (is it better?), compatibility (does it fit my life?), complexity (is it hard to understand?), trialability (can I test it cheaply?), and observability (can I see others using it?) (Rogers, 2003).
Bass gave the same phenomenon a clean mathematical engine. His model treats adoption as the sum of two forces: innovation, external influence like advertising and media that reaches people independently, and imitation, internal social influence as adopters prompt non-adopters around them. Tune those two coefficients and the model reproduces the S-curve and forecasts a new product's take-up, which is why, decades on, it remains the standard tool for new-product forecasting (Bass, 1969).
Then comes what has changed, and it is substantial. The first shift is bandwidth. When Rogers wrote in 1962, information about an innovation travelled slowly, person to person, so awareness was often the bottleneck. On today's web, awareness of a new thing can be near-instant and global, so the constraint moves downstream, to access, price, and trust rather than knowledge. Knowing a product exists is rarely the barrier now; being able to get it, afford it, and believe it is.
The second and most consequential shift is the collapse of the influentials story. The intuitive idea, descended from Katz and Lazarsfeld's two-step flow (Katz and Lazarsfeld, 1955) and popularised by Gladwell's The Tipping Point (Gladwell, 2000), is that a few special, highly-connected people tip trends, so the game is to find and win those influentials. It is a lovely story, and it has not survived testing. Using simulations and data, Watts and Dodds showed that large cascades are usually triggered not by rare influentials but by a critical mass of easily-influenced ordinary people: whether something takes off depends far more on the susceptibility of the many than on the reach of the few (Watts and Dodds, 2007). Aral and Walker, running field experiments on a social platform, confirmed that influence is structurally distributed across a network rather than concentrated in special nodes (Aral and Walker, 2012). The influencer sits, at best, on top of a wave made of ordinary people, and is often just early or lucky rather than causal.
The third shift reframes the why beneath the S-curve from personality to network structure, which is the territory of L4-07: simple contagions (information, viral content) spread easily along loose, far-reaching ties, while complex contagions (costly or norm-laden behaviours) need reinforcement from several connected people and so spread through dense, clustered networks (Centola, 2018). The S-curve still fits, but its engine is now understood as a network-and-reinforcement story, not a tale of five personality types.
The honest cautions all point the same way: the descriptive vocabulary is useful, the causal strategies built on it mostly are not.
The adopter categories are descriptive, not causal, and this trips up a lot of practice. "Innovator" and "early adopter" are defined purely by when someone adopts, so they are labels you can only apply after the fact, not a pre-existing segment you can identify and target before launch. The famous percentages (2.5%, 13.5%, and so on) are just slices of a bell curve of adoption timing, not a discovered typology of personalities. Building a launch strategy around "targeting the innovators" is therefore chasing a group you can only see in the rear-view mirror.
The influencer strategy overpromises for the same underlying reason: if influence is distributed and cascades ride on ordinary susceptibility, then paying a premium to a handful of "key influencers" is often poor value compared with reaching many ordinary people, and the hits that get attributed to an influencer were frequently going to happen anyway. And the simple-versus-complex distinction warns against the commonest error of all: assuming a costly, norm-laden behaviour (adopting a new financial product, changing a health habit) will diffuse like a funny video. It will not, because complex behaviours need reinforcement that memes do not.
A fair caveat is that the foundational debunking work predates the modern creator economy: Watts and Dodds wrote in 2007 and Aral and Walker in 2012, before Instagram, TikTok, and paid influencer marketing became a mass industry, so it is worth asking what newer evidence shows. It refines the picture rather than reversing it. Watts's own follow-up, run on Twitter, found that individual influence stayed very hard to predict in advance and that the most cost-effective approach was to enlist many ordinary "influencers" rather than bet on a few large ones (Bakshy et al., 2011). More recent marketing research confirms that paid influencer campaigns do have real, measurable effects on engagement and sales, but that those effects are strongly moderated, by the fit between influencer and brand, the kind of followers they have, and how openly the sponsorship is disclosed, and that smaller "micro-influencers" often beat mega-influencers on trust and engagement (Hughes, Swaminathan and Brooks, 2019; Leung et al., 2022). The upshot refines the original finding rather than reversing it. The newer evidence does not resurrect the romance of one special person tipping a market; if anything it confirms the distributed picture, breadth, fit, many small voices over a few large ones, and how hard the breakout still is to predict. So "influencers don't matter" would be wrong, they are a genuine channel, but "find the one key influencer who tips your category" remains the myth.
Bass forecasting works best where there is history to calibrate against and worst for genuinely novel categories, exactly when a forecast is most wanted. Much of the influentials-debunking evidence comes from specific online platforms and may not capture every offline dynamic. And diffusion research still leans on Western and consumer-technology cases, which matters because adoption speed, network structure, and the access-price-trust bottleneck differ sharply across markets, developing economies often leapfrog stages (going straight to mobile, skipping fixed infrastructure) in ways the classic S-curve was not built to describe.
How much does the influentials picture change offline, in high-trust interpersonal and community networks rather than social platforms? Can diffusion be forecast well for truly new-to-the-world categories, or only interpolated from analogues? And how do the classic patterns bend in leapfrogging and mobile-first markets that skip the sequence the model assumes?
The usable core: the S-curve and the Bass model are good for forecasting how something spreads, but the popular targeting strategies, find the innovators, buy the influencers, are mostly built on a misreading, because adoption is driven by ordinary susceptibility and network structure, not by rare special people.
Keep the forecasting, drop the personas. Forecast take-up with the Bass model (external influence plus imitation) rather than with an "early adopter persona", because the adopter categories are descriptive labels applied after adoption, not a segment you can find and target before launch (Rogers, 2003). The Bass model earns its keep as a real planning tool: fit its two coefficients from the history of an analogous product when your own is too new to have data, and use the resulting curve to time production capacity, inventory, and media weight, the decisions that actually turn on when the takeoff comes. Its honest limit is that it interpolates from analogues and struggles with genuinely new-to-the-world categories, so treat the forecast as a planning aid with error bars, not a promise.
There is one part of Rogers you can act on before launch, and it is not the adopter categories but the five perceived attributes that predict how fast anything spreads. Engineer for them directly: sharpen the relative advantage so the product is clearly better at a job people already do, raise compatibility by fitting it to existing habits rather than demanding new ones, cut complexity so there are fewer steps to first value, build in trialability through free trials, samples, and low-commitment first use, and raise observability by making adoption visible so others can see it in use (Rogers, 2003). These are levers you hold before a single unit ships, which is more than the persona targeting ever offered.
Treat "find the key influencers" with real scepticism. The evidence says influence is distributed, so reaching many ordinary, easily-influenced people usually beats paying a premium for a few tastemakers (Watts and Dodds, 2007; Aral and Walker, 2012), which also rhymes with the penetration lesson of L5-02 and the weak-persuasion lesson of L5-01. It is worth understanding why the influential story is so seductive, and Watts's Everything Is Obvious (Watts, 2011) is the accessible account: a cascade looks, in hindsight, as though the person at the top caused it, but that person was usually early or lucky, and our "common sense" rewrites the story afterwards to make the outcome seem inevitable and the influential seem decisive. Once you stop hunting for the hero node, influencer marketing becomes a real but conditional channel to run on its own terms. The newer evidence points to breadth and fit across many smaller, well-matched creators over a premium bet on a mega-name, and no one can reliably pick in advance which post will break out (Hughes, Swaminathan and Brooks, 2019; Leung et al., 2022). Micro-influencers often win on trust and engagement for exactly this reason. And match your mechanism to the behaviour: if you are spreading information or light content, weak-tie reach works; if you are spreading a costly or habit-changing behaviour, that is a complex contagion needing clustered reinforcement, so seed several connected adopters in the same pocket and instrument the network structure rather than hunting for a hero node (Centola, 2018; L4-07). The dual-use edge is the influencer-industrial complex. An agency selling concentrated influence, "we will find the person who tips your category", is selling a story the evidence has largely dismantled, and the audience test exposes it: does the plan serve the client's growth or the agency's roster? The honest version reaches breadth, matches creators to the brief, and engineers reinforcement instead. In practice that means spreading a launch across many well-matched smaller voices and ordinary-buyer reach, then reading the data for genuine clustering, rather than front-loading the budget onto one name and hoping it tips.
Movements spread on the same corrected logic: not a handful of charismatic influentials tipping the masses, but a critical mass of ordinary, persuadable people reinforcing each other, which is why organising density and repeated peer contact tend to matter more than a celebrity endorsement. A genuine behaviour change, turning out, switching allegiance, joining, is a complex contagion, so it needs the clustered reinforcement Centola describes rather than a single viral moment: several people someone already trusts, moving together, do what one distant famous voice cannot (Centola, 2018; L4-07). The applied error to avoid is spending a movement's budget on one big endorsement when the same money spread across dense local organising would compound. Distributed influence also reframes what to measure: instead of chasing a celebrity's reach, watch whether adoption is clustering, whether new supporters are appearing next to existing ones, because that is the signature of a complex contagion taking hold rather than a one-off spike that fades.
Diffusion of a public good, a vaccine, a benefit, a safe practice, is usually a complex contagion, which means the classic "raise awareness" campaign addresses the wrong bottleneck: awareness spreads fast, but adoption waits on access, affordability, trust, and social reinforcement. Effort is better spent lowering those barriers and building the visible, repeated local examples that complex behaviours require than on broadcasting information people already have. Centola's How Behavior Spreads is the applied case for this (Centola, 2018): behaviour change takes hold when people see several trusted others in their own network adopt, so seeding clustered, reinforcing local uptake, through community health workers, respected local figures, and visible neighbourhood adoption, beats a mass-media blitz aimed at lone individuals (L4-07). Bass forecasting has a public use too, planning the rollout of a new service or programme so that capacity matches the expected curve. The ethical line is that the aim is genuine adoption of something that serves people, not manufactured pressure, so the same clustered-reinforcement machinery must never tip into coercion. And localise the curve: adoption speed, network structure, and the access-price-trust bottleneck differ sharply by market, and developing economies often leapfrog stages, going straight to mobile and skipping fixed infrastructure, so a rollout modelled on a mature-market analogue can misjudge both the timing and where the real barrier sits.
Three habits. Use the S-curve and Bass model to forecast and to spot which of the two forces (outside influence versus imitation) is driving a spread, but do not turn the adopter categories into a targeting persona, they are a rear-view label, not a findable segment. Distrust "find the influencer" pitches: influence is distributed, so breadth usually beats betting on a few special people. And before assuming something will "go viral", ask whether it is a simple contagion (information, which spreads easily) or a complex one (a costly behaviour, which needs reinforcement), because the second does not travel like the first.
.png)