What do you do when you’ve built a product up from nothing to the point of being acquired for hundreds of millions of dollars by a massive American corporation? You take some time off.
That’s exactly what Felix Miller and Martin Stiksel, founders of Last.fm, did in 2007 when CBS Interactive (disclaimer: a former employer of mine) bought it, but it was exactly this that led the pair to their latest idea, Lumi. A new content recommendation engine, essentially.
“We became normal consumers, private people – buy a house, do up a house, have a garden, grow vegetables, do DIY, all that stuff,” Miller explained to me. “And with any new hobby you have a lot of need for information, so we had to look up lots of stuff whether it’s books or online or whatever but the process somehow felt really clunky and time-consuming and unsatisfying. You want to spend time in your garden but then you have to spend all day researching a hoe!”
The entrepreneurs’ answer to this process was to look at drawing on their experience from Last.fm and using data that already exists anyway to generate content recommendations for things they might like to read or watch, or listen to.
“Martin put it very nicely, he said, ‘I’ve been browsing the Web for 15 years and I have no benefit from that. I have a few bookmarks and I’ve learnt how to write a good Google query but that’s it – all this knowledge I’ve produced, all these choices I’ve made, they all disappear into the ether’ [this is when we decided] this is what we should be doing,” Miller said.
So, in order to get recommendations you simply open an account (by email, or by connecting your Twitter account), install a browser add-on or extension (available for Chrome, Firefox or Safari at present) and it’ll scan your web browsing history file. Not third-party cookies, your history file, which is then anonymized.
This distinction might seem arbitrary but with a service like Lumi, Miller and Stiksel know that the public is now far more aware of anything that looks like it might invade online privacy more than ever before.
Any browsing data that is collected is completely anonymized and even the Lumi team (which consists of just eight engineers and one product manager) couldn’t match the data with the user. Needless to say, despite the fact you create an account, your profile – and therefore recommendations – are not viewable by anyone else. Collection of that data’s sole purpose is to generate content recommendations.
Cynics will be thinking “..and revenue” but the pair also promised they’d never monetize Lumi by selling the personal data of its users. In fact, when quizzed about monetization, it seemed like the last thought on their mind.
“Our focus is on a great user experience,” Miller said. Stiksel added that the hardest part of Lumi so far was keeping it simple and refining the choices and options given to the user to a bare minimum.
This isn’t the first time Lumi has been unveiled to the world, but the launch in December was deliberately a limited affair designed to gain early feedback from the 10,000 or so users that were on board and iron out any bugs, not that I’ve seen any.
Simple is the word
Using Lumi is sort of like using an RSS reader (indeed, it displays content in a similar way to some), except that you don’t need to make any effort to do it as suggestions simply come from yours – and others’ – browsing habits. So you might end up on the Guardian website, or you might end up on that small but bizarrely popular little blog that you like.
The real idea, of course, is to get you to places you wouldn’t end up normally by pooling collective browsing data – the more people who like certain topics that read certain things, the more they get suggested to other people who read the same topics. To get you to that bizarrely popular little blog that others like but you’ve never come across.
So, for example, when I joined it scanned my browsing history and identified the key topics, namely business, design, games, law, music, software, startups and tech. You can see the bar down the left-hand side in the image below, moving the mouse away from the bar hides it until you next mouse-over.
So, while these might be my automatically assigned categories (and they’re obviously pretty accurate too as they’re based on actual browsing) yours would likely be very different.
The bar also gives quick access to connected friends, your profile, your collections and your starred items. The latter in this list are the items that are displayed publicly whenever you star a page (whether it’s discovered via the Lumi interface or not) using the little Lumi logo icon that the browser extension installs alongside the address bar; there’s also a star icon in the top-right hand corner of the page if browsing an article on the Lumi website, and you can of course, change your settings so that starred items are no longer public by default.
Each item is categorized by its topic, and clicking the topic on any entry will take you to a Lumi page full of new content. So, for example, if you only want Android news, clicking the Android tag on an appropriately tagged story will give you just that. But again, this isn’t just about news, it could just as likely be suggesting links to cookery forums, extreme sports boards and blogs – pretty much anything, depending solely on your browsing history.
It’s this automatic categorization and tagging of content that allows Lumi to survive with such a small team and work in such a hands-off manner. For it to work, you never need to open it again once the browser extension is installed, but if you come back to it one day in the future, it will be populated with content that’s as fresh as it ever was based on your browsing habits.
While the site works happily on mobile phones, HTML 5 is the current focus of the team. That’s not to say there won’t be native apps one day, but for now, perfecting the Web app is the primary goal.
Whether or not people will see the benefit in yet another content discovery tool remains to be seen, but having given Lumi a quick test run, I’d say it certainly helps to cut down on the ‘noise’ within the realms of my own Web browsing and given that it takes no relearning of Web browsing behavior and no input from the user (if desired) it’s got a good chance of gaining some traction.