Most of the big social media and app companies are pretty light on hard-core technology. Happy to stand on the shoulders of the tech giants that came before, many focus instead on features, design and UI. This enrages the kind of hardcore math nerds that used to rule the Valley.
Well, they have a new geeky mascot: Uber. Uber only scales and survives with hardcore mathematicians on staff. Among its braniac hires are a rocket scientist, a computational neuroscientist and a nuclear physicist. (That’s an actual staff photo to the left.)
I have no idea what those disciplines have to do with predicting cabs arrivals and sorting cab inventory. But apparently, something.
A new chest-thumpy blog post shows that using Google’s ETAs for Manhattan cabs was leading to horrendous wait times for riders, about 3.6x off the estimates. That’s pretty much the worst possible user experience for a first time Uber user, particularly in a city where cabs are plentiful and users may never give it another try.
Uber dropped the Google API like a hot potato and developed its own algorithm. It wasn’t particularly comfortable about this, because it didn’t have much historical data to go on. But as some graphs in the post show, it immediately did better. How much better? Their quants crunched some numbers for me and found that Uber is on average 186.3 seconds more accurate than Google. On average, Uber’s ETAs were 42.50% more accurate than Google’s. And with every ride, Uber gathers more data and the estimates get better.
Do a few minutes make that much of a difference when you’re waiting on a cab? Well remember, this is the average. In some cases the differences between Google’s ETAs and Uber’s ETAs was 15 minutes or more. And if you’re standing in the rain waiting on a cab, hell yeah 186 seconds matter. Given that Seattle is one of the cities next on Uber’s launch list, this is a valuable algorithm to get right.
Uber CEO Travis Kalanick and I talked backstage at Disrupt about how his company lives and dies on its “Math Department,” as they call the team in house. The video is below. (We talk math at the four minute mark.)