Should we give it away free?

The question behind the question

“Free” is never a business model; it is a decision about who pays. Someone always does: advertisers, sellers, the users of a premium tier, a parent company with another source of income, taxpayers or volunteers. The real question is whether that payer exists, gains as the free side grows, and will keep paying. The site’s pattern who pays shapes what gets built collects cases where the answer decided what the product became.

Free also has psychological force. In 2007 Kristina Shampanier, Nina Mazar and Dan Ariely showed in experiments that a price of zero is not just a low price: people treat free items as more valuable than a small price difference would justify, choosing a free lower-quality chocolate over a much better one at a token price. That effect drives adoption, and it also makes later charging very hard.

When free worked

Hotmail made email free in 1996, ran it in a web browser and added an invitation to sign up at the foot of every message; each user’s email recruited the next. Gmail offered a gigabyte of storage free in 2004, paid for by advertising chosen by scanning messages. Alibaba’s Taobao charged sellers nothing to list in 2003, against eBay, which had brought its global fee model to China; Taobao earned money from advertising and services and eBay withdrew. Google made its web map free in 2005 and free turn-by-turn navigation standard on Android phones in 2009, which collapsed the market for paid navigation devices.

These successes share a structure. The free product removed a barrier that had limited adoption, the payer was clear (advertisers in Google’s case, sellers buying promotion in Taobao’s), and the payer gained directly as the free side grew. In several cases the free product was a feature for a company that earned its money elsewhere, which is why rivals that sold the same thing could not respond: Garmin and TomTom could not give away navigation because navigation was their business. Red Hat showed a different path: it gave away the software and charged enterprises for certified, supported, long-maintained versions, selling the reliability that free software alone could not guarantee.

When it did not

The failures show what happens when the payer is missing or the costs grow. Twitter’s Vine let millions of people watch short videos made by a community of creators, but gave the creators no way to earn; they moved to platforms that paid them, and Vine closed. India’s decision in 2020 to make UPI payments free for merchants by law drove adoption extraordinarily fast, but left the cost of running and securing the system to subsidies and to apps with little direct income, and by 2026 the payments body was proposing fees on larger payments. France’s Minitel gave households terminals free from the early 1980s and built a thriving national online service paid for by per-minute charges; the model worked for a decade and then made the system hard to abandon when the internet arrived.

A diagnosis to start from

The diagnosis begins with three questions. Who pays, and does their willingness to pay rise as the free side grows? Do costs rise with every free user, as with storage, bandwidth, computing for AI or support, or do they stay roughly fixed? And is the product free because it is cheap to provide, or because the provider earns money elsewhere?

The guiding policy that follows from the successes is to make free the side whose adoption is hardest to win and whose growth makes the other side pay more, the logic Jean-Charles Rochet and Jean Tirole formalised in their work on two-sided markets. The coherent actions are a pricing structure that makes the paying side’s gain visible, cost controls that keep each free user cheap, and a clear line between what is free for good and what will be charged for, stated at the start.

The strongest counter-case

The strongest objection is that free can destroy value for everyone, including the giver. When news became free online, funded by advertising that later moved to search and social platforms, many publishers lost the income that paid for reporting. Free products funded by advertising have an incentive to maximise attention rather than value, a problem the site’s essays on recommendation feeds and social media describe. And in the age of AI, the cost of serving each user is no longer near zero: free tiers for AI models cost real money per question, which has pushed companies towards advertising, as OpenAI began to do in 2026, with all the incentives that brings.

The objection does not overturn the advice, but it adds a test: whether the payer’s interests align with the users’. Advertising-funded free products work best when what advertisers pay for, intent or attention, is also what users came for, as in search, and worst when the two diverge.

What the wider research says

Clayton Christensen and Michael Raynor argued in The Innovator’s Solution (2003) that when one layer of a product becomes good enough and commoditised, profits move to adjacent layers, a principle they called the conservation of attractive profits. Free is often the visible sign of that shift: a company that profits in an adjacent layer gives away the layer others sell. Chris Anderson’s Free (2009) popularised the idea that digital goods tend towards zero price, and the record supports a narrower version of it: free wins against paid when a rival with another income makes your product its feature.