Google's failures

One of the most depressing things about looking at Google is how many failed projects there are. All those tens of thousands of engineers, supposedly the best and the brightest, all with their 20% time (even it has been scaled back a bit lately) and yet I can count almost all the good Google products on the fingers of one hand (I have to count on my fingers - I did history at university).

There are a lot of conspiracy theories about all Google’s failed projects - especially the ones that never looked for a moment like they’d work. There’s the theory that they’re a way to soak up cash and make Google look less profitable to regulators. Or, a way to soak up engineers who might otherwise get bored of working in advertising and go to Facebook or Apple. My favourite theory is that these projects are a sort of sandpit for Larry and Sergey - a way to distract the founders, keep them out of the way and stop them interfering with the core business. “Larry has another dumb idea about adwords? Send him that Space Elevator proposal!”. Or, they’re just a way for the founders to pretend they’re not running an advertising business.

The boring truth is that most of the 20% project are things we never hear about- clever optimisations of small details of search or advertising. But the repeated public failures are a puzzle - why do so many clever people screw up so many products? Base, Knol, Jaiku, Dodgeball, Answers, Squared, dMarc, Buzz, Wave…

All of the thousands of blog posts around Google’s culture home in on an engineer-led, data-led culture where even marketing people come from engineering backgrounds and everything gets a-b tested. Most famously, they tested 41 shades of blue for a single link, leading to the (rather distinguished) designer resigning in disgust.

This means that Google starts its thinking with an engineering solution and then builds the GUI. That worked well for search, where the algorithm required a single text box - even Google couldn’t screw that up. Reader or Gmail’s interfaces don’t need much thought - they certainly didn’t get much innovation. But what if you’re building a social network? Or a photo-sharing site? You don’t start with the engineering - in fact the engineering comes last. You need to get it right, of course: Friendster strangled itself at birth with poor scalability. But you start with wireframes, work out the site page by page and you design the architecture LAST, at about the time you’re deciding what radius the circles on the corners should have.

You can’t A-B test whether your new social network should share people’s private email address book with the entire world without even FUCKING ASKING. There’s no ‘correct’ engineering solution here - no interesting computer science problem. There’s no data that can tell you what to do - rather what you need is judgement.

So, my more-or-less serious theory (based on absolutely no chats with actual real Google employees) is that Google will never make a decent social product, nor any other product for which the engineering is generic and success depends on judgement calls around the UX, because Google’s culture does not appear to respect skills needed for such products.