How Uber Is Trying To Turn Everyone In the Company Into A Data Scientist

Just as every business is now a tech, so every employee must be, to some degree, a wizard. JL


Venture Beat reports:

By unifying data science with their cross-functional counterparts in engineering, product, and design, Uber has built internal tools and platforms that others in the company can leverage, bringing cutting-edge expertise to anyone within the company. The company gets hundreds of millions of signals continuously. Uber needs to know whether the app is functioning as it should be.The vision of the platform is to provide forecasts at the push of a button. No forecasting expertise [is] required. The input that’s needed from the user is historical data, as well as the forecast horizon.
Uber’s Director of Data Science, Franziska Bell, would like everyone who works at Uber to have the ability to execute their role with the same insight as a data scientist. “Our vision is to provide data science expertise at the touch of a button,” Bell said during her talk at Transform 2019 this month. By unifying the teams in her organization — the data science ninjas, as she calls them — with their cross-functional counterparts in engineering, product, and design, Uber has built internal tools and platforms that others in the company can leverage, bringing cutting-edge expertise to anyone within the company.
One example she pointed to is the forecasting platform they’ve developed, or as she described it, “platformization at ease.”

“The grand vision of the forecasting platform is to provide forecasts at the push of a button,” said Bell. “Absolutely no forecasting expertise [is] required. The only input that’s needed from the user is historical data, whether it’s in the form of a CSV file or a link to a query, as well as the forecast horizon. How far do you want to forecast? Everything else is done completely underneath the hood.”
Before developing the platform, Bell mapped out the three questions the team examines to determine if the investment in any kind platformization will be worthwhile. Number one: Is this area of data science going to greatly enhance the user experiences? Two: Are there multiple different use cases across the company? Three: How reusable are the various different methodologies or modules from use case to use case?
Forecasting was a slam dunk across the board, and she went on to explain its many use cases across Uber, including supply and demand, real-time outage detection, and hardware capacity planning.

“Obviously, we want to provide great user experiences, and in order to do so, we want to accurately forecast supply and demand-related metrics in a spatial/temporal, fine-granular fashion,” Bell explained. With the vast number of metrics, paired with the fact that many of these are supply- and demand-driven, it would be impossible to set up those thresholds by humans and keep them up-to-date over time.
Bell discussed how real-time outage detection is crucial to Uber’s business. The company gets hundreds of millions of signals continuously, whether from back end systems it is tracking or marketplace health indicators. Uber needs to know whether the app is not functioning as it should be, if people can’t sign in or sign up, or if people can’t take a trip as expected.
Hardware capacity planning formed the third leg of forecasting, using the platform to monitor closely how much hardware to purchase and provision for. “Particularly tricky is the fact that we have some very high-demand days, such as New Year’s Eve or Halloween,” she explained. “We’re a fairly young company. These are very much day-of-the-week dependent. Again, our forecasting platform and expertise can be leveraged on this front in order to ensure that we’re not over-provisioning, which would be fiscally irresponsible, but also not under-provisioning, which could cause an outage and erode the trust of our rider and driver partners.”
Leveraging its forecasting expertise, Uber has now built a completely automated tool said Bell. “People just have to bring us a list of business metrics they want to track, or backend metrics, and then we automatically and dynamically set these thresholds for them, and also keep them updated.”