Making it rain to make it rain
Our goal at Uber is to get you a clean, comfortable ride regardless of rain, sleet, or snow. Unfortunately Uber’s demand is not flat: it varies by city, neighborhood, time of day, day of week, local weather patterns, and so on. All of these variables, and more, go into predicting demand for Uber.
In order to make sure you get a ride when you need it we need to accurately forecast demand. We know that demand spikes on evenings and weekends, as well as during workweek mornings. Those are predictable. Early on we at Uber also noticed that when it rains, demand surges. No one likes walking around in the rain trying to catch a bus or hail a cab. When it rains, busses and subways become crammed with wet, unhappy people. So people turn to Uber instead. But weather is difficult to predict, and weather-related surges in demand negatively impact a city’s transportation systems and impact Uber’s demand prediction models.
We don’t like complex models; we like predictability, drivers like predictability, and you like predictability. When it rains without much notice, demand spikes, and we can quickly hit capacity during peak hours. This means it’s harder to get a ride, your car takes longer to show up, and our algorithms activate surge pricing (making everything more expensive).
Rain saps up excess supply. While this is bad during peak hours, it’s great during the mid-day lull when our supply is underutilized. Ideally we’d be able to even out these spikes, but we can’t change your work hours and we can’t change closing times (yet). So what can we do? Some things are just out of control. Right?
But come on. It’s 2013. Humanity is achieving technological wonder after technological wonder: we are ushering in de-extinction, we are on the cusp of decoding the brain, and we’re even getting self-pics back from our robot on Mars.
It is time for Uber to embrace the greatest technological breakthroughs and minimize the uncertainties of nature.
Thus, it is with great pleasure that we at Uber announce the first stage of our equal-demand-at-all-times campaign: “Making it rain to make it rain”.
Just like China did for the 2008 Olympic Summer Games, Uber has been conducting an experiment in cloud-seeding over San Francisco. Armed with a collection of Uber-branded anti-aircraft guns, rocket launchers, and planes, we have been firing rockets filled with sodium or potassium salts into the sky to induce precipitation. While traditional cloud-seeding technologies make use of beta silver iodide (β-AgI) crystals to provide a particulate nucleus around which ice crystals can form to then fall as precipitation (known as glaciogenic seeding), this only works for high, cold clouds. Anyone who’s spent time in San Francisco knows that our clouds are lower and much warmer than freezing. Therefore our Uber scientists have been using hygroscopic seeding, which makes smaller water droplets coalesce into fatter drops to fall as precipitation. (For further information, see this Nature 2008 News Feature on weather modification.)
While over the short term cloud-seeding causes rain, the mid-term effect may be to stabilize and control unpredictable weather patterns.
That is, by creating brief windows of severe rain, Uber is simultaneously making weather-induced surges in demand more predictable while reducing weather-related uncertainty in the mid-term. By way of analogy, think of how firefighters perform controlled burns to prevent future uncontrollable fires. Uber is causing rain when and where it wants to smooth out demand surges.
Recent reports suggest that,
Rain output from seeded storms, on average, was more than double (110 percent more) than that from nearby untreated storms. Single, isolated thunderstorms that were seeded produced an estimated 144,472 acre-feet above and beyond what could have been expected without intervention. More complex thunderstorm clusters that were seeded yielded an estimated 2,520,772 additional acre-feet of rainwater.
Cloud seeding costs are relatively low (estimated at around $20 per acre-foot of precipitation). This means that for a city as small as San Francisco, relatively small increases in precipitation would be low-cost and, if properly timed to occur during period of low demand, can yield up to 150% ROI.
We’re uber-excited and encouraged by these early successes and are continuing to explore new models for making demand prediction easier and demand less variable. So we’ll be continuing on with experiments in manipulating the outcome of Giants games to make them win even more frequently and in making sure that bars just never close. If we can eliminate the post Giants game surge in demand after a loss, or remove the closing time spike, we’ll be that much closer to perfect demand prediction.