SAN FRANCISCO — Not too many years ago, Pandora and Evernote were tiny startups with uncertain futures. But these companies anticipated growth and made steps to prepare their infrastructure for rapid adoption on both mobile and desktop.
Pandora CTO Tom Conrad recalls just one white knuckle moment.
In the early days, Pandora’s most loyal customers were asked to pay for a $36 a year subscription to listen to Internet radio on an unlimited basis. This revenue model worked well enough, but it posed a potential barrier to growth. So the executive team made the decision to offer the service for free.
“We saw 10-times growth overnight,” Conrad said onstage today at VentureBeat’s MobileBeat 2013 conference. “We weren’t ready for it.” He described how the company’s engineers spent the next 72 hours writing code, racking servers, and “sweating a lot.”
In the mid-2000s, Pandora’s Oakland, Calif.-based team made quick fixes as the company grew, and it slowly built up a scalable architecture. They’ve come a long way, given that it all started with a Linksys router.
Looking back, Conrad stands by the decision to build their own infrastructure and buy bandwidth. “We made the call to do it ourselves, and now a big part of our value proposition is reliability,” he said.
Evernote chief technology officer Dave Engberg has some practical advice of his own to share.
Engberg watched Evernote’s note-taking application grow to 50 million customers in the space of five years. To spur this growth, his team invested in development for desktop and mobile.
As the customer base grows, Engberg’s team has been able to predict and anticipate consumer behavior through research and “complicated regression analysis.” Evernote discovered that the multiplatform approach is key to sustainable growth.
Evernote found that 24 percent of people use the app on mobile and desktop. Mobile is difficult to monetize as “people don’t expect to pay as much,” but folks tend to check the app far more often on a smartphone or tablet device, often up to 30 times each day.
“How much you are using it [Evernote] at work really changes as you go from mobile-only to mobile and desktop,” Engberg explained. Multiplatform customers, he said, are worth about 20 times more than those who only use the tool on mobile or desktop.
On the topic of scale, Engberg believes that it’s about peak and not bandwidth. “That’s what scale is really all about,” he said.
Engberg also recommends building architecture from scratch. And to the dismay of our developer beat reporter, Jolie O’Dell, he believes that platform agnostic HTML5 isn’t really a viable option for Evernote. “It will get to the point where it will be good enough for 75 percent of applications,” he said. “But you don’t get the best performance for something that doesn’t take into account all the nuances of that platform.”
The speakers agree that Internet entrepreneurs need to prepare for rapid adoption, even in the first few months. As Engberg puts it, if you’re a video streaming service, anticipate a flood of traffic when a new Justin Bieber video is about to come out.
Edward Hieatt is an expert on scaling and agile development. He is the chief operating officer for developer-consulting firm Pivotal Labs, which worked with fast-growing companies like Groupon at their peak. Twitter had chronic problems with its infrastructure (hence the Twitter “fail whale” error message) until Pivotal Labs stepped in.
Hieatt recommends that startup engineering teams build “malleable” code that can be tweaked, and automate their quality assurance. In the past, Pivotal Labs has encouraged clients to use Ruby Red Labs for web design. He also suggests hiring developers that are highly disciplined, and willing to work together at a physical office.
“I think the term ‘agile’ is undefined, but it basically means building software that can scale,” said Hieatt.
Do you have other practical suggestions for scaling your applications and tools? Please leave a comment below.
VB's research team is studying web-personalization... Chime in here, and we’ll share the results.