When launching early is risky, learn early instead

Dropbox’s first product was a video. In April 2007, long before the service was ready for the public, Drew Houston posted a three-minute screencast to Hacker News showing files moving between computers as if by magic. The comments that came back told him more about what people wanted than months of private building would have. About a year later, a second video, posted to Digg, pushed the beta waiting list from about 5,000 people to 75,000 in a single day.

Houston laid this out in his talk at the Startup Lessons Learned conference in April 2010, and “Learn early, learn often” was one of the slides. The phrase is a deliberate edit of the software maxim “release early, release often”. Dropbox could not follow the original. A syncing service that loses or corrupts a user’s files does not get a second chance, so shipping a rough version to the public and fixing it live was out. The learning had to happen some other way.

The rework separates two things the original maxim bundles together. Releasing early is one route to learning, and often the best one, but it is not the only route. A video, a landing page with a sign-up box, a private beta for a few hundred people, or long conversations with target users can each answer a question before the product is exposed to everyone. The goal is the earliest possible understanding of customers, and the release is a means to it.

The approach suits products where failure in public is expensive: anything that holds data, money or health information, or hardware that cannot be patched after it ships. It also suits founders who worry, as Houston did, that the biggest risk is building something nobody wants. A screencast costs days, and a year of engineering on the wrong product costs the company.

The limit is that some things can only be learned from real use. A waiting list measures curiosity. It says nothing about whether people keep using a product after the first week, or whether they will pay for it. Dropbox itself learned some of its hardest lessons after launch. The same slides record that paid search advertising cost between $233 and $388 to win a customer for a $99 product, a mistake no demo could have revealed. The replacement, a referral program that paid both sides in extra storage, could only be judged the same way, on live users, and by early 2010 it brought in about a third of daily sign-ups.

There is also a risk in the other direction. “Learning” can become a reason never to ship, with each new survey or prototype postponing the moment when the product meets the market. Houston’s version works because it is paired with an actual launch, which Dropbox made in September 2008. By January 2010 it had about four million users.

The line’s real contribution is to treat learning as the thing to schedule. Releases, demos and interviews are tools for it, chosen by how much each one costs and what it can reveal.