Product Manager Portfolio 1: Uber ETA

Problem:

List the top three most important variables Uber could use to estimate the ETA for passenger pickup, beyond the ETA on Google Maps.

Clarification:

Passenger pickup is the time it takes for the selected Uber car to pick up the passenger. It does NOT include the time it takes to locate a driver, and it does NOT include the time it takes to drive the passenger to their destination.

Clarification:

Do we have to count the time on the user side (e.g., the user walking to the pickup position)? Does “beyond the ETA on Google Maps” mean not using the ETA from maps at all, or making the ETA more precise?

Landscape:

Uber is one of the largest ride-sharing apps in the U.S. I do not work there, but based on my personal experience, I would assume their mission is to make transportation easier for people. If we look outside of the company itself to see the market, I would say they are located in a highly competitive market where they have direct competitors like Lyft and some indirect competitors like shared bicycles, car sellers, or public transportation. So it totally makes sense that Uber wants to improve their ETA estimation accuracy to provide a better rider experience. In this part, I’ve introduced Uber’s landscape, their mission, and the goal for this ETA improvement. I will pause here to see if you have any questions or anything else you want to add.

Even though the logic is simple as user place order → uber distribute order → driver accept order → driver pickup user finally drop user. The overall logic is simple but there still some small things could effect ETA time that beyond ETA from maps provider since user actions are unpredictable and community enviroment are varies.

Hypothesis:

I believe the time it takes for a user to go out the door to the pickup location varies depending on user habits. But normally, these habits could be grouped as the environment of the pickup location, since users going to the same place will possibly perform almost the same actions. I will also assume the community environment could affect drive time, since the starting position and pickup position could possibly be complicated, which may affect the ETA.

Given this, I will do some quick brainstorming to identify potential variables Uber could use to estimate the ETA more accurately.

So, given user actions, I will assume the environment and user actions will affect the ETA accuracy significantly. Since user actions are affected by the environment, I would like to make the environment the key topic of our variable definition.

Variables:

• Start & Pickup Location Types:
  When a driver starts from downtown or a shopping mall, the time cost will be significantly different compared to a position in a general single-family house community. So I would say this will affect it a lot.
• User Starting Location Types:
  Even though we have analyzed location types from the driver’s side, the user’s starting location could still affect the pickup ETA. This point is majorly affected by user actions, but since user actions are unpredictable, I would say people who go to the same place will have similar actions. For example, if a user places an order at a shopping mall, they are typically finishing shopping; they will go downstairs and maybe say goodbye to their friends. In this situation, the user’s location is still a strong variable affecting ETA.
• User Habits and Pickup Time:
  We could combine these two metrics, but it is actually the same variable: user habits. User habits can be explained in two ways. Macro user habits: probably users will have similar habits at the same time. I know it’s almost the same as “User Starting Location Types,” but we are talking about time right now. For example, if going back home from a nightclub on a Friday night, people might be delayed because many people choose to go to nightclubs on Friday nights, so the traffic will be horrible and the driver could be blocked on the way, affecting the ETA. In a micro way, every specific user has different habits; probably User A will spend 5 minutes getting to the door because they need to drink a cup of coffee before going out, but User B could just leave the door on time since they are ready to go out when they place the order.

Of course, there are still a lot of different variables like traffic and work time. But those are things maps’ ETA already considers. What we have to consider here is only how to improve ETA beyond the maps’ ETA to provide a better experience, so I will not take too much time to discuss that. I would like to pause here to see if you have any questions or anything you want to add.

Evaluation:

Now let’s shift to the evaluation part. I will evaluate the three metrics we just picked to see how they would affect our platform. By choosing start & pickup location types as variables, we would be able to improve accuracy by estimating drive time in a specific region, since maps normally would not consider this small region. But it could still be wrong due to some special events or random situations. As for user starting location types, we could use this to improve accuracy by predicting user actions. But as we mentioned before, user actions are unpredictable; probably before going to the pickup location, the user wants to grab a cup of coffee or go to the restroom, which could reduce accuracy in some specific cases. For user habits, this could also be affected by unpredictable user actions, but it will be generally correct since once a user’s workflow is defined, it would not change too frequently. However, this part will lead to some technical issues, like how to design the algorithm for a specific user and specific time at a specific location, which would also be a challenge.