A team spends a year on a product and launches it on a Tuesday with a press release, a social post and a cake. Traffic spikes, the dashboard glows, and by the following Tuesday the line has fallen to a tenth of the peak. Was the launch a success? That depends on a question the cake cannot answer: of the people who came, how many stayed, and did they bring anyone?
Another team launches quietly in one country, to a few thousand people, with a switch that can turn the feature off. Nothing happens on the news. Six weeks later they can see which users come back every week, what they call the product, and what they would pay. Launching is less one event than a sequence of decisions about who sees what, when, and how you will know.
- Who takes it
- Part 2 of the Product Management minor, a short route that sits beside many majors. A minor is not a degree of its own; it adds a skill to a degree. Pairs with: Business Administration, Marketing, Computer Science, Economics. Launch and growth are taught in business-school courses on entrepreneurship, marketing and technology strategy, and in the product and growth electives that sit beside computer science.
- What it is for
- Getting a product to the right people safely, finding out whether it is wanted, and growing it without fooling yourself or your users.
- Comes after
- Product Strategy, Roadmaps & Prioritization (36/97) and Product Innovation & Go-to-Market (34/63).
- Leads to
- No later subject lists it as a prerequisite; it closes the minor. It connects back to Product Metrics & Experimentation (36/98), whose retention curves and tests are how you judge a launch.
- What this piece gives you
- You will not run a launch alone after this. You will be able to choose a launch size, score a product-market-fit survey with its uncertainty, see why a loop fizzles or compounds, compare two ways to charge, and fill in a launch checklist. Easier first reads: Product Innovation & Go-to-Market (34/63), which covers channels and why launches fail (this piece does not repeat it), and Pricing without fear (04/03).
Launches come in sizes
It helps to separate three things that get called "launch". Releasing is putting the code in front of some users. Learning is measuring what happens. Telling is announcing it to the world. A mature team can do the three at different times: ship a feature behind a flag to 1% of users (36/76), watch for a week, widen to 10% and then 100%, and announce it only when it holds. A dark launch runs a feature for real traffic without showing it. A beta is a named group that agrees to be early. A soft launch opens in one market first; Spotify opened in October 2008 in Sweden, Finland, Norway, France and Spain, and reached the United States in July 2011.
Many companies sort launches into tiers, so that a copy change does not get a press event and a new product does not get a quiet email. The names and cut-offs are each company's own convention; there is no standard. The usual questions are the same: does it reach a new audience, does it change price or billing, does it alter how current users work, and would it be hard to reverse? Try the sorter. The rules in it are invented, to show the reasoning.
Tick whatever is true for your change.
Without JavaScript: 0 to 1 ticks suggests a quiet staged release, 2 to 3 a standard launch with a plan and a support brief, and 4 to 5 a major launch with a named owner, a rollback plan and a post-launch review.
A launch checklist is the tool that carries these decisions: who owns it, what success looks like, how to roll back, who must be told. You will find one to fill in at the end of this piece. Most items on it are boring, and boring is the point: Heathrow's Terminal 5 opened in March 2008 with a baggage-system breakdown and many cancelled flights, and HealthCare.gov was overwhelmed when it opened in October 2013. Reviews of both pointed to testing and rehearsal that came too late. The common thread in such stories is rarely a missing genius; it is a missed rehearsal.
Product-market fit: a feeling you can measure
Marc Andreessen wrote on 25 June 2007 that "product/market fit means being in a good market with a product that can satisfy that market," and that it is the one thing that matters for a startup. He credited the phrase to Andy Rachleff, co-founder of Benchmark Capital. Founders describe the feeling: before fit, you push and users drift away; after it, usage grows faster than you can handle and people tell friends.
Sean Ellis, who led early growth at Dropbox, LogMeIn and Eventbrite, proposed a measurable test. Ask active users one question, "How would you feel if you could no longer use this product?", with the answers "very disappointed", "somewhat disappointed" and "not disappointed". After comparing nearly a hundred startups, Ellis found that those with at least 40% answering "very disappointed" tended to have traction, and those below it struggled. Rahul Vohra of Superhuman used it in 2017 and reported that his product started at 22%. The 40% is a rule of thumb from one practitioner's sample, not a physical constant, and it should be asked of people who have actually used the core of the product, not of everyone on a list.
With small surveys the uncertainty is the story. Enter your own counts below. The tool gives the share, a 95% confidence interval (the Wilson interval, which behaves well at small counts), and says plainly when the interval straddles 40%.
Without JavaScript: 12 of 30 is exactly 40%, but the 95% interval runs from about 25% to 58%, so the survey cannot tell. The same 40% from 300 responses runs from about 35% to 46%.
The survey is a leading signal. The test that cannot be argued with is behaviour: do cohorts flatten in the retention curve (36/98), do users return without prompting, do they pay and stay? Be wary of fit in a narrow group. Early adopters can love a product that the wider market will not, which Geoffrey Moore called the chasm in Crossing the Chasm (34/63 covers it).
Growth loops versus funnels
A funnel (36/98) is a line: people come in at the top, some reach the bottom, and to get more out you must pour more in. A growth loop is a circle: the output of one cycle becomes the input of the next. A user signs up, invites friends, some of whom sign up and invite their own. Content loops work the same way: users publish pages, search engines send visitors, some become users who publish more.
The arithmetic is short. If each active user sends i invites a month and a share c accept, each user brings in k = i × c new users a month, the "viral coefficient". Add retention r, the share of users still active next month, and the user base next month is r + k times this month's, plus whatever you buy with ads. If r + k is below 1, the loop only slows the leak and the base settles at a ceiling. At 1 or above it compounds. Real products rarely sustain a very high k: invitations are used up, the friends who say yes are the easy ones, and each cycle takes time, which matters as much as the size.
Without JavaScript: starting from 1,000 users with retention 70%, 2 invites a user and 12% joining (k = 0.24, so r + k = 0.94), plus 200 bought a month, the base heads for a ceiling near 3,333 and has about 2,220 users after 12 months. Raise the invites to 4 (k = 0.48, r + k = 1.18) and it compounds to about 14,270 in the same year.
The useful habit is to ask which number in the loop is weakest, and whether you can raise it with a product change rather than a bigger ad budget. Invites that are part of the job (sharing a document so a colleague can edit it) behave very differently from invites that interrupt it (a prompt to "tell your friends").
Product-led or sales-led, and how to charge
A product-led company lets people try the product themselves and pays its way mostly through self-serve upgrades; the product does the selling. A sales-led company puts people between buyer and product: demos, negotiation, contracts, onboarding. Neither is better in general. Self-serve fits products that are quick to understand, cheap per seat and used by individuals first. Sales-led fits expensive, complex or regulated purchases where several people must agree. Many companies combine them: a free tier pulls teams in and a sales team follows once a company's usage grows.
Pricing has two halves. The value metric is what you count to charge: seats, messages, gigabytes, transactions, projects. A good one grows when the customer gets more value, so the bill feels fair, and is easy to predict. Packaging is how features are grouped into plans. Two common entries are a freemium plan (free forever, limited) and a free trial (full access, for a while). Freemium reaches many people and costs you for those who never pay. A trial sees fewer sign up but filters for intent. Which earns more depends on your numbers. The toy below uses invented ones, so change them to see what would need to be true.
Without JavaScript: with the starting values, freemium gives 2,000 sign-ups, 60 payers and $1,200, minus $582 to serve 1,940 free users, for $618 net in month one. The trial gives 800 trials, 160 payers and $3,200, minus $192 to serve the 640 who do not pay, for $3,008. Freemium would need about 8.9% of sign-ups to pay to match it in this first month.
Read this as a way to ask questions, not as a prediction. It covers one month; freemium users may pay later, refer others or make the product more valuable to those who do pay (the next section), none of which is in the toy. The real answer comes from testing (36/98) with your own customers. For how people think about price, see Pricing without fear (04/03).
Network effects: real or just a story?
A network effect exists when each new user makes the product more valuable to the others. A telephone is useless alone. Marketplaces and payment systems have two-sided effects: more sellers attract buyers and more buyers attract sellers. The effect can also be local: a ride-hailing service works because there are drivers in this city, not because there are drivers somewhere. Andrew Chen, in The Cold Start Problem (2021), describes the hardest part as building the first small network that works on its own and then stitching those networks together.
Robert Metcalfe's law, first stated in 1983, says the value of a network grows with the number of possible connections, n(n − 1) ÷ 2, roughly the square of the users. In 2006 Bob Briscoe, Andrew Odlyzko and Benjamin Tilly argued in IEEE Spectrum that this overstates it, because most possible connections are never used, and that value is closer to n log n. Researchers fitting models to real data, such as studies of Facebook and Tencent, have reported results on both sides. The sober reading is that value grows faster than the number of users, by an amount that varies, and "we have network effects" is a claim to test, not a slogan. Check it yourself: do existing users get more value as others join, and do the newer cohorts retain as well as the older ones?
Metcalfe: n(n − 1) ÷ 2 n log n n (straight line)
Without JavaScript: at 100 users there are 4,950 possible links and n ln n is 461; at 1,000,000 users there are about 500 billion possible links and n ln n is about 13.8 million. The units are arbitrary, so only the shapes should be compared.
Going global: payments, languages and rules
A product built for one country meets three surprises elsewhere. The first is how people pay. Cards are not the default everywhere, and some of the largest payment systems in the world are public or telecom rails.
| System | Where | Started | What to know |
|---|---|---|---|
| M-Pesa | Kenya | March 2007 | Mobile money from Safaricom and Vodafone, built on agent shops |
| UPI | India | Pilot April 2016 | Run by NPCI, regulated by the Reserve Bank of India |
| Pix | Brazil | November 2020 | Created and run by the central bank, Banco Central do Brasil |
| WeChat Mini Programs | China | 2017 | Apps that run inside a messaging super-app |
Pix reached about 42 billion transactions in 2023, according to figures reported from the central bank, and UPI processed some 20 billion in a single month, August 2025, according to NPCI figures; a checkout without them can lose customers who would not think to ask for a card. The lesson is not "add Pix" but to find out how your customers move money before you decide.
The second is language. Internationalisation (often written i18n) is building the product so it can be adapted: no text hard-wired into images, space for longer words, right-to-left layouts for Arabic and Hebrew, dates and numbers formatted by locale. Localisation (l10n) is the adaptation itself: translation, currency, examples, tone. Plurals show why it is more than swapping words. English has two forms ("1 file", "2 files"); the Unicode CLDR data lists up to six plural categories, and Arabic uses all of them. Names are not always "first" and "last", and addresses vary widely.
The third is rules: sales taxes (VAT in the EU, GST in India and Australia), consumer law, and data protection. The GDPR in Europe, the LGPD in Brazil and China's Personal Information Protection Law (in force since November 2021) each limit what you may collect and where it may go (36/98). Launching in a new country is a project of its own, and the safest first step is often a partner who already knows it.
Sunsetting, and the ethics of growth
Products end. Google Reader was announced for closure on 13 March 2013 and shut on 1 July 2013, and many of its users never forgave it. Google+ launched on 28 June 2011; in October 2018 Google announced its consumer version would close, citing low engagement and a flaw that could have exposed user data, and it closed on 2 April 2019. Juicero, which raised about $120 million to sell a $699 juice press (cut to $399 in 2017), suspended sales in September 2017, months after a Bloomberg report showed the packs could be squeezed by hand with similar results. These are different kinds of ending: one was a planned retirement, one a retreat from a social network that never found its audience, and one a product whose price and engineering were out of step with what it did.
A good sunset is a small project: decide with data (who uses it, what do they use it for), announce early, give people a way to export what is theirs, honour what was paid for, point to alternatives, and keep the date. Cutting off people without warning is the part users remember.
Growth has an ethical edge. A dark pattern is a design that tricks people into a choice they would not otherwise make: a subscription that is easy to start and hard to cancel, or pre-ticked boxes (36/75). The word was coined by Harry Brignull in 2010; the US Federal Trade Commission published a staff report on them in September 2022, and the EU's Digital Services Act (Regulation 2022/2065) restricts them on online platforms. The related debate is about attention-maximising design: infinite scroll, streaks, notifications. Researchers disagree about the size of the harm. Amy Orben and Andrew Przybylski reported in 2019 that technology use explained at most 0.4% of the variation in adolescent well-being across three large datasets, and argued this was too small to guide policy. Jonathan Haidt argues in The Anxious Generation (2024) that the average hides serious harm to some young people, and points to the timing of rising distress after smartphones spread. Others, such as Candice Odgers, find the evidence for a causal link weak. Product people don't have to settle that to act: if a metric rises because users feel trapped, it is a metric about trapping.
Template: a launch checklist
Tick what is true, fill in the blanks, and copy the result into your tracker. It is a starting list; delete what does not apply and add what your product needs. Nothing is stored or sent.
What a graduate just sees
- Releasing, learning and telling are three steps, and they do not have to happen on the same day.
- A launch-day spike says nothing about fit; the cohorts after it do.
- 40% "very disappointed" is a rule of thumb, and a small survey has a wide interval around it.
- A loop compounds only if retention plus new users per user reaches 1; otherwise it just leaks more slowly.
- Free versus trial is a question about numbers you can measure, not a matter of taste.
- "Network effects" is a claim to test: ask whether later users make earlier users' experience better.
- Payments, language and law are part of the product in each country.
- How you end a product is part of how people judge the company.
Where it shows up next
- Product Metrics & Experimentation (36/98): retention, cohorts and tests are how a launch is judged.
- Product Innovation & Go-to-Market (34/63) and Marketing Research & Analytics (34/59): channels, positioning and segments.
- Starting a Business (04): pricing, the first customers and the legal minimum for a new product.
- Shipping Software (36/76): feature flags, canaries and rollbacks, the machinery of a safe launch.
- UI & UX Design (36/75): dark patterns and the line between persuasion and trickery.
If you want to go deeper
- 📘 Textbooks. Andrew Chen, The Cold Start Problem (Harper Business, 2021), on network effects and how products reach critical mass; and Sean Ellis and Morgan Brown, Hacking Growth (Crown Business, 2017), on running growth experiments as a team practice.
- 📘 Gentler. Eric Ries, The Lean Startup (Crown Business, 2011); and Geoffrey Moore, Crossing the Chasm (HarperBusiness, first published 1991), on moving from early adopters to the mainstream.
- 🌐 Free course. Y Combinator's Startup School, a free online programme with lectures on launching, growth and finding product-market fit.
- 🌐 Reference. Marc Andreessen's original essay, The Only Thing That Matters, and First Round Review's how one team measured fit.
Launch in steps you can undo, and measure what stays instead of what spikes. Ask people the one question that shows whether they would miss you, keep the uncertainty in view, and look for loops where growth comes from the product itself. Adapt to the places you enter, and when you must end something, end it in a way you would want to be on the receiving end of.
Try this: pick a product you use weekly and ask yourself Ellis's question. Then write down who you would tell, how it would reach them, and what would make you stop. That is a one-person fit survey and a loop in miniature.
Sources. Andreessen, "The Pmarca Guide to Startups, Part 4: The only thing that matters," 25 June 2007; First Round Review, "How Superhuman built an engine to find product/market fit" (Ellis's 40% test; Superhuman's 22% in 2017); Briscoe, Odlyzko and Tilly, "Metcalfe's Law is Wrong," IEEE Spectrum, July 2006; Chen, The Cold Start Problem (2021); Unicode CLDR plural rules; Brignull's coinage of "dark patterns" (2010); US Federal Trade Commission, Bringing Dark Patterns to Light (staff report, September 2022); Regulation (EU) 2022/2065 (Digital Services Act); Orben and Przybylski, Nature Human Behaviour 3, 2019; Haidt, The Anxious Generation (2024); Spotify, M-Pesa, UPI, Pix, Google+ and Juicero dates from company and central-bank announcements as summarised in encyclopedia articles. All numbers inside the tools are invented for illustration.
- 📘 Chen, The Cold Start Problem
- 📘 Ellis and Brown, Hacking Growth
- 🌐 Y Combinator Startup School