On 9 January 2007 a phone was announced; it went on sale on 29 June at $499 or $599 on a two-year contract. About 270,000 were sold in the first week, and in September the price of the larger model fell to $399. Six years earlier, on 3 December 2001, a two-wheeled personal transporter was unveiled after weeks of speculation that it could be as big as the personal computer. Its price was $5,000. Over its whole life about 140,000 were sold, and production of the original model ended in 2020.
Both were clever machines. One changed an industry and the other found a niche. The subject of this piece is the thinking that tries to tell them apart before the money is spent: where products come from, how adoption spreads, how to know you have fit, how to launch, and why so many launches miss. The honest answer, as you will see, is that the field can describe the patterns better than it can predict the cases.
New products come from three directions. Technology push: a lab makes something possible and a company hunts for a use. Market pull: customers have a problem and someone builds the answer. And users themselves: MIT's Eric von Hippel showed that people who use a product in demanding ways often build the first version, and that the "lead users" (whose needs run ahead of the market) are a rich source of ideas. In one of his multi-industry studies, lead users were the pioneering developers of about half of the innovations judged most important.
The best-known lens for the pull side is jobs to be done. The idea is that customers do not buy a product; they "hire" it to make progress in a situation. Clayton Christensen's milkshake story: a fast-food chain found many milkshakes sold early in the morning to solo commuters, hired for a job (make a boring drive less dull and keep hunger away until lunch), so the competitor was a banana or a bagel, not another milkshake. The method was set out in Competing Against Luck (2016, Christensen, Hall, Dillon and Duncan). The older line that people want "a quarter-inch hole, not a drill" is usually credited to Theodore Levitt, though its exact origin is murky.
The critics have a point, and the field is split. There are rival schools: Christensen's, and Tony Ulwick's "outcome-driven innovation", which defines a job as a measurable set of desired outcomes. "Job" is defined differently across them, the milkshake case is one consulting story rather than a test, and the framework is hard to falsify. Christensen's earlier theory of disruption has had similar scrutiny: Jill Lepore's 2014 New Yorker essay questioned its evidence and its reliance on selected cases. Treat jobs to be done as a useful question ("what is this person trying to get done?") rather than a proven method.
Everett Rogers's Diffusion of Innovations (1962; 5th edition 2003) gathered hundreds of studies, beginning with Ryan and Gross's 1943 study of hybrid seed corn among Iowa farmers, which took over ten years to spread even though the corn paid for itself. Rogers sorted adopters by when they adopt: innovators (first 2.5%), early adopters (next 13.5%), early majority (34%), late majority (34%) and laggards (16%). The cut-offs come from slicing a bell curve; they are a convention, not a law of nature.
Frank Bass turned the picture into one equation in 1969. At any moment, the chance a non-adopter buys has two parts: an outside push that does not depend on anyone else (p: advertising, curiosity) and an inside pull that grows with how many people already have it (q: word of mouth, imitation).
Across many consumer products, studies report an average p of about 0.03 and q of about 0.38 per year. With those, sales peak at t* = ln(q/p) ÷ (p+q), about 6.2 years. Tune it yourself.
Bass model for a market of 1,000,000 potential buyers, 15 years. Bars: new adopters each year. Line: share adopted so far. Illustrative.
Notice the lesson in the second preset: with weak outside push, a strongly social product starts slowly and then explodes. With only outside push, sales are highest on day one and fall steadily. The shape of the curve tells you what to spend on: reach in the first case, referral in the second.
Geoffrey Moore's Crossing the Chasm (1991) says the Rogers curve has a crack in it. Early adopters (visionaries who will tolerate a rough product for a big payoff) buy for different reasons from the early majority (pragmatists who want a complete solution and references from other pragmatists in their own industry). Praise from enthusiasts does not move them. Moore's advice is to pick one narrow niche, win it completely with a full solution, and use its references to cross, a bit like securing a beachhead before the main landing.
It is a famous idea with a contested evidence base. Rogers himself rejected it, writing that past research showed no support for a chasm between adopter categories. Goldenberg, Libai and Muller (2002) found a "saddle" (a dip in sales after the early peak) in a minority of consumer products, which they attribute to weak communication between segments, not to a universal gap. The toy model below has two groups, early adopters (16%) and a mainstream that mostly copies its own kind. It is mine, not Moore's.
Two-segment Bass-style model: early adopters (16%) p 0.25, q 0.5; mainstream (84%) p 0.004, own q as set, plus the chosen pull from early adopters. Illustrative, not data.
"Product–market fit" means a product that satisfies a strong demand in a market, so that customers pull it out of you. Andy Rachleff of Benchmark named the idea; Marc Andreessen popularised it in the mid-2000s with the claim that it was the only thing that mattered for a startup. How do you see it? Sean Ellis, who looked at many startups, proposed one question: how would you feel if you could no longer use this product? If at least 40% of users answer "very disappointed", you are probably there. The 40% is a rule of thumb from experience, not a law.
The trouble is noise. A survey of 30 people cannot tell 40% from 25%. And the people who answer are not everyone: users who stuck around, who were willing to fill in a form. Both biases push the number up. Use the tool to see how wide the doubt is.
95% Wilson interval for a proportion. The re-run draws random samples from the "true share" you set: each press is a fresh simulated survey of the same people.
A launch is a set of choices: who the first customers are, what promise you make them, where they will hear it, what you charge and when you go. A cheap way to learn is a staged launch: a small group first, then a wider one, so mistakes are small. For channels, Gabriel Weinberg and Justin Mares's Traction (2015) lists 19 (search ads, content, partnerships, sales, referrals and more) and suggests a "bullseye": brainstorm all, test three cheaply, then focus on the one that works.
To compare channels you need two numbers. Customer acquisition cost (CAC) is what you spend to win one paying customer. Lifetime value (LTV) is the profit that customer brings before leaving. Investors in subscription businesses often quote rules of thumb: an LTV of at least three times CAC, and a payback of under a year. They are conventions, not laws, and they vary by industry.
Presets are invented, illustrative numbers, not benchmarks for any real channel.
You will often read that "95% of new products fail", usually credited to Christensen, or that 80% or 90% do. Be careful. In a 2013 review in the Journal of Product Innovation Management, George Castellion and Stephen Markham traced the best-known failure rates and reported that the very high figures lack solid support, and that careful surveys of companies' own launches imply a much lower rate. I could not retrieve the paper's exact numbers while writing, so no number is quoted here.
The deeper problem is that "fail" means many things:
| Definition of failure | Gives a rate that is |
|---|---|
| Missed the internal sales target | High: targets are often hopeful |
| Pulled from the market within a few years | Moderate |
| Never recovered its development cost | Lower than the first, higher than you think |
A different measure with real data: the US Bureau of Labor Statistics follows every private-sector business establishment (new products are not tracked separately). About 47% of establishments that opened in the year to March 2021 were still open five years later, and about 34% of those opened in the year to March 2016 survived ten years. That is about business survival in general, not the same as product failure, but it fits the same pattern: failure is common, though not 95%.
| Product | What the record shows | What the usual story gets wrong |
|---|---|---|
| Edsel (Ford), 1957 | Launched 4 Sep 1957, cancelled 19 Nov 1959; about 116,000 sold, under half the break-even point. | "Research killed it" is too simple: a recession, prices overlapping Mercury, a divisive grille and quality complaints all contributed. |
| New Coke, 1985 | Launched 23 Apr; "Classic" back 11 Jul (79 days). The new taste beat both Coke and Pepsi in blind tastings. | Tastings missed attachment to the old brand. Executives denied a deliberate plot ("we're not that dumb, and we're not that smart"). |
| Segway, 2001 | $5,000 at launch; about 140,000 sold in its lifetime; the original model ended 2020. | The hype, not the machine, was the failure; its descendants live on as scooters and in tourism and security. |
| Google Glass, 2013 | $1,500 for "Explorers"; consumer sales ended Jan 2015; business editions ended 2023. | It was a social acceptance problem as much as a technical one. |
| Quibi, 2020 | Raised $1.75 billion; launched 6 Apr; about 500,000 subscribers; shut down by Dec 2020. | Its founder blamed the pandemic; others point to the product and the content. |
A product succeeds when it does a job someone cares about, reaches them through the right people in the right order, and costs less to win than it earns. The models (jobs, Bass, the chasm, fit, unit economics) are good questions rather than forecasts, and the famous failure rates and flop stories deserve a second look.
Try this: pick a product you started using in the last year. Write down the job you hired it for, who told you about it, and what you stopped using.
Sources. Bass: Frank Bass, "A New Product Growth Model for Consumer Durables", Management Science 15(5), 1969; closed form F(t) = (1 − e−(p+q)t) ÷ (1 + (q/p) e−(p+q)t), peak t* = ln(q/p) ÷ (p+q); averages p 0.03 and q 0.38 from Mahajan, Muller and Bass (1995) as summarised in standard references. Rogers, Diffusion of Innovations (1962; 5th ed. 2003); Ryan and Gross (1943) hybrid corn in Iowa; adopter shares 2.5/13.5/34/34/16 are Rogers's standard categories. Moore, Crossing the Chasm (1991, revised 1999 and 2014); Rogers's rejection of a chasm; Goldenberg, Libai and Muller, "Riding the saddle", Journal of Marketing (2002), summarised from memory, share of products with a saddle not stated. Von Hippel lead-user work (1986; the nine-industry 54.4% figure is as reported in secondary sources). Rachleff and Andreessen on product–market fit; Sean Ellis 40% test (rule of thumb). Lepore, The New Yorker (2014). Castellion and Markham (2013), Journal of Product Innovation Management: cited from memory, not re-read. BLS Business Employment Dynamics, establishment survival (accessed 2026): 47.2% at five years for the year to March 2021, 34.2% at ten years for the year to March 2016. iPhone: announced 9 January 2007, released 29 June 2007, $499/$599, price cut 5 September 2007, about 270,000 sold in the first week. Edsel, New Coke, Segway, Google Glass and Quibi dates and figures from standard references (Wikipedia summaries of contemporary press and company statements). Every number in the interactives was computed in node; the toy two-segment model is the author's own.