An S-curve describes how the cumulative adoption of a new product, technology, or behavior unfolds over time: a slow, almost flat start while a small base of users experiments; a steep middle phase as word-of-mouth and network effects compound; and a flattening top as the addressable market saturates. The shape is the signature of any system that combines a positive feedback loop (each adopter recruits others) with a constraint (a finite pool of potential adopters). For growth investors the S-curve is less a precise forecasting tool than a mental discipline: it forces you to ask where on the curve a company actually sits — pre-inflection, in the explosive middle, or quietly decelerating toward saturation — because that placement, more than the trailing growth rate, governs the future. The central tension is that the same 40% growth number looks wildly different at the bottom of the S (accelerating, re-rate higher) versus near the top (about to decay, re-rate lower).

How it's formed / calculated

The cumulative curve is sigmoidal (logistic-shaped); its first derivative — new adopters per period — is the familiar bell curve of adopter categories.

Rogers' Diffusion of Innovations (1962). Everett Rogers partitioned adopters along the bell curve into five groups, with commonly cited proportions of innovators ~2.5%, early adopters ~13.5%, early majority ~34%, late majority ~34%, laggards ~16% (derived from successive standard deviations of a normal distribution; widely reproduced but a stylized convention, not a measured constant). Rogers also identified five perceived attributes that predict rate of adoption: relative advantage, compatibility, complexity, trialability, and observability.

The Bass model (Frank Bass, 1969). The most-used quantitative formalization. Adoption hazard is:

> f(t) / (1 − F(t)) = p + q·F(t)

where F(t) is the cumulative adopted fraction, p is the coefficient of innovation (external influence — advertising, media), and q is the coefficient of imitation (internal influence — word-of-mouth). A third parameter m is the total addressable market (the ceiling). Peak new-adoption timing is t\* = (ln q − ln p)/(p + q). Across a meta-analysis of published applications, rule-of-thumb averages are roughly p ≈ 0.03 and q ≈ 0.38 (years), with p typically 0.01–0.03 and q typically 0.3–0.5 (Bass-model literature; Lilien/Rangaswamy and the Sultan/Farley/Lehmann meta-analysis). The large q relative to p is why diffusion is back-loaded and steep in the middle.

Geoffrey Moore's chasm (1991). Moore argued the bell curve is discontinuous for disruptive tech: a gap separates early adopters (visionaries) from the early majority (pragmatists), and many products die in this chasm rather than gliding up the curve.

How it's used in practice

1. Locating the inflection. Investors track penetration of a named addressable population — e.g. streaming households, EV share of new-car sales, cloud as a fraction of IT spend — and watch for the acceleration that signals the steep middle. The most valuable re-ratings historically come just before or at inflection, when growth is reflexively reinforcing. 2. Sanity-checking TAM math. A credible bull case implies a future penetration of a defined market. The S-curve lets you reverse-engineer the implied path and ask whether the q (virality, network effect) and m (ceiling) are plausible. 3. Anticipating deceleration. Past ~50% penetration of the realistic market, the law of large numbers and a shrinking remaining pool mean growth must slow — the math is unforgiving regardless of execution. Mistaking late-S deceleration for a temporary stumble is a classic overpay. 4. Stacked S-curves. Durable compounders chain successive S-curves (Apple: iPod → iPhone → services). Industry-level analysts use this to distinguish a one-product saturation story from a serial-innovation platform.

Adoption, debate & evidence

The S-curve as a descriptive pattern is extremely well established — diffusion of innovations is among the most-cited frameworks in the social sciences, and the Bass model has thousands of empirical applications across consumer durables, pharma, media, and energy. Where it earns its keep is after the fact: fitting a curve to realized data.

The hard, honest caveat is that S-curves are far weaker as forward predictors than they look. The two parameters that matter most — the ceiling m (true addressable market) and the steepness/timing — are exactly the ones you cannot observe until the curve is well advanced. Early-stage Bass fits are notoriously unstable: small changes in the first few data points swing the implied peak and ceiling enormously (a documented weakness in the new-product-forecasting literature — e.g. Forecast Pro, Lilien & Rangaswamy). Practitioners often "borrow" p and q from analogous prior products precisely because they can't estimate them from sparse data. Additional failure modes: technologies that never cross the chasm (no S at all), competitive substitution that truncates the curve early, redefinitions of the market that move m, and the survivorship bias of studying only technologies that did diffuse. There is no robust evidence that S-curve fitting alone generates investment alpha; its value is as a structuring lens, not a signal.

Strengths & limitations

Works best as a diagnostic: identifying that a category is pre-inflection vs. saturating, stress-testing TAM narratives, and explaining why growth rates mean-revert. It is genuinely powerful for forcing the "where on the curve?" question and for tempering both euphoria (early) and despair (chasm).

Fails when used as a precise forecaster from thin data, when the ceiling is assumed rather than defended, and when a single curve is projected through what is actually a chasm or a competitive cliff. The #1 misuse: drawing a smooth S to a self-serving TAM and treating the steep middle as inevitable — every doomed bubble stock had a beautiful projected S-curve. Adoption is a hypothesis, not a destiny.

Sources

• Rogers, Diffusion of Innovations — summarized at Wikipedia: Diffusion of innovations and Technology adoption life cycle
• Bass model formula, p/q meanings, peak-timing equation: Wikipedia: Bass diffusion model
• Typical empirical parameter values (p ≈ 0.03 with range 0.01–0.03; q ≈ 0.38 with range 0.3–0.5, time in years): Wikipedia: Bass diffusion model (which attributes these averages to the Sultan/Farley/Lehmann meta-analysis); secondary illustration PyMC-Marketing: Bass model intro
• Forecasting instability / new-product application: Forecast Pro: The Bass Model
• Crossing the Chasm (Moore): Technology adoption life cycle
• Investing application & S-curve framing: Mawer: S-Curves and the Adoption Cycle (EP124); Umbrex: S-Curve of Industry/Technology Evolution

Dispute flags: Rogers' exact adopter percentages are a stylized convention, not a measured law. Bass p/q "averages" are rules of thumb that vary widely by category and time unit. No robust evidence that S-curve fitting alone produces investment alpha — treat as a structuring lens, not a predictive edge.