Uber Applying Machine Learning to Improve the Customer Pickup Experience Case Study Solution is the case from Kellogg School of Management. The case can be interpreted from the viewpoint of digitization, Statistical analysis, digital strategy, and consumer experience enhancement point of view. We have solved all the questions here.
For more solutions to cases such as Uber Applying Machine Learning to Improve the Customer Pickup Experience, refer to our case solutions here
What are the potential points of the pickup experience for the different customer personas? Use the personal information in the case, the narrative in Exhibit 5, and your personal experience to identify pain points for each of the rider personas, as well as for the driver?
In this case, there is a specific mention of Five Personas. Apart from it, I was able to break into a few other personas:
1. Premium: Users who Uber Black or Uber X. They don’t really care about the money and they should ideally be picked from the exact location of their starting points. The driver should become as close to the point of pick-up and drop as close to the drop-off.
Rider: Should be able to choose from a host of premium cars and ready to pay with various other modes
2. HFR: The app should easily pick up frequent areas where the rider commutes like this who use Uber many times in crowded cities like NY and New Delhi. The economy also matters to them and hence bulk pricing with minimal surge will be an added value proposition for them
Rider: Should be able to get a cab in less than 2 minutes without a lot of surge pricing
3. Commuters: Exact pick-up and drop locations should be the value propositions for them as time is of importance as they use Uber for commuting to the office. Drivers should have the least wait time as well. A flexible payment system and a pass are also desired
4. Rider: SHould get a cab immediately without waiting or calling the driver frequently for a location. Flexible payment systems are also desired. Also sharing the cab should also be made possible
5. Social: For availability and surge pricing is the biggest issue. They should be able to use it in low-density areas and hours and also during the high surge
Driver: Should have a queue in logic and can come to location as fast as possible. Should also be able to pick up a correct user using a unique verification system like QR or OTP
5. Traveler: Locating specific meeting points is needed as they are new cities. The app should be able to guide them specifically to a location where a pool of cars is available
Driver: Should have a queue in logic and can come to location as fast as possible. Should also be able to pick up a correct user using a unique verification system like QR or OTP
6. Moto (Additional)– Both Cost and availability should be a concern for them. A separate user map should be available for them as it will benefit their time and notional experience
Driver: Should adhere to strong safety needs. Telemetry data should be fed in for speed violations and penalized
7. Infrequent Users (Additional)– People who use Uber when their car breaks down. For them, they should be able to download the app and use it right away without worrying too much about location and user history and an easy payment interface like Cash
Driver: Should be able to pick up and should be incentivized for good service so that the new user can come to the Uber ecosystem
8. Drivers (Additional): They should have the least wait time and mapping with users should be done based on the average user wait time. For example, if a user needs 7 minutes on average to board the ride should not be mapped with a driver who can be available in 2 Minutes
For more solutions to cases such as Uber Applying Machine Learning to Improve the Customer Pickup Experience, refer to our case solutions here
Define the ideal pickup experience for each of the rider personas, as well as the driver persona. Use the information provided in the case regarding the priorities and expectations of individual personas to define the ideal pickup experience for each one. Create a short description of the ideal experience, as well as a list of two or three outcome expectations for each persona?
It can be easily established from the case that Uber is trying to create an ecosystem where both riders and Drivers are equally valued. There are primarily potential customers/users for Uber:
A. Drivers (as they form a major part of the ecosystem)
B. Commuters/Riders
Based on these the following Personas can be created
Drivers:
1. Hobbyists Drivers – Like the one in the US
2. Uberx/Black Drivers – Specialized vehicles. Can be closely compared to chauffeur-driven cars of the USA like Limousine
3. Moto Drivers- Mentioned in the Case – In Countries like India and Asia
4. Regular Drivers – Who drives Uber for living
Riders:
1. Frequent Riders – Office Goers
2. Using Uber for Functions / Concerts
3. Infrequent Uber Users
4. Someone visiting a new city
5. New Riders who use Uber from Airport and Railway stations
The following points can be identified and their association with personas can be mentioned:
1. Exact Location Cannot be identified due to GPS error in high-density areas. As GPS accuracy decreases in areas with high buildings – For All
2. For New Riders Finding the area of pickup from the Airport and Stations can be tough
3. For Infrequent riders- Since Customer Profile is not mapped, difficult to find a regular pickup point
4. Drivers can end up on the wrong side of the road – For Riders and Drivers
5. For Concert Goers and Social Users – Time and congestion as usage picks up significantly during the end or the start of concerts
6. For Hobbyists Drivers – Driver Scores are not mapped well as the usage volume is low
7. For Moto Drivers – The scale of the App Usage can be a problem
8. For Infrequent Users – Not a lot of users’ data can be easily mapped leading to satisfaction scores being low
9. For the High Density of Pickups in famous areas like Central Square or Connaught Place in Delhi – finding an exact pickup location can be a problem
10. For Uber Drivers who drive Uber for a living longer wait time means a significant loss of revenue and wastage of fuel due to wrong pickups
For more solutions to cases such as Uber Applying Machine Learning to Improve the Customer Pickup Experience, refer to our case solutions here
The Following Hypothesis can be used to enhance the experience:
1. Predictive Modelling: Like if a commuter uses Uber on a Saturday to visit the Church, it should be able to remind it and make it a more seamless experience
2. Match Singularities: Only GPS data from Mobile should not be the only match criteria. If the rider’s GPS data and cab location meet then rides can be rolled out. In case of high mismatch say -20%, voice command triggers should be used
3. Mapping Usage Pattern: For HFR and Commuters this can be used. If a rider has booked a location more than 5 times, it should be marked and should be provided to the driver as a landmark for pickup
4. Telemetry Data: Speed and Acceleration Parameters would be critical to identify driver behavior specifically for Moto and Uber X/Black as safety is paramount there
5. Subsidiary Data Pooling: Subsidiary data from services like Uber Pool can be used also to monitor data
6. New User: For someone new in the uber ecosystem, the simplicity of the app and rider experience should be the only criteria as it will incentivize him to use the data again
7. Multidimensional Feedback Mechanism via Multivariate Regressions: A Multidimensional feedback mechanism can be used where exact matches can be made. Multiple Regressions of real-time data can exactly map the user with the driver
For more solutions to cases such as Uber Applying Machine Learning to Improve the Customer Pickup Experience, refer to our case solutions here
Create a quantitative pickup quality metric using attributes derived from the passive, active, and third-party signals available to Uber. Discuss why you believe your selected attributes represent a robust pickup quality metric. What weights would you assign to the features you chose for your pickup model?
Based on your pickup quality metric, what actions can Uber operators take to improve the pickup experience?
There is mention of a specific metric in the case, which I think can be expanded significantly to increase the efficacy of the ecosystem. We would name it the “PickUp Grand Score”
The following parameters can be used to enhance the experience:
Driver Side:
1. Driver Wait time – Active
2. Distance Covered to meet the rider – Active
3. Ease of Finding the location (Qualitative) – Passive
4. Driver to deny duty if passenger score is low – Passive
5. Mapping of High ranked drivers to high-ranked riders- Active
Rider Side:
1. Steps required to meet the driver- Active
2. Wait time to book a cab – Active
3. Calls / Texts made to reach the location- Active
4. Ease of Payments-Passive
Third-Party Signals
1. Telemetry data from Uber Vehicles
2. Age and Sex of the rider – Priorities for Female and child riders
3. Congestion Metric – Google Maps API
4. Time Spent on the Uber App -Should be as low as possible
5. Entertainment apps usage- Use of Value added services for the entertainment devices installed in the cars – Tie up opportunities with Netflix and Spotify
Weights Assigned
1. Driver Wait time – Active -15%
2. Distance Covered to meet the rider – Active-10%
3. Ease of Finding the location (Qualitative) – Passive-5%
4. Driver to deny duty if passenger score is low – PAssive – 5%
5. Mapping of High ranked drivers to high-ranked riders- Active-5%
Rider Side:
1. Steps required to meet the driver- Active-20%
2. Wait time to book a cab – Active-10%
3. Calls / Texts made to reach the location- Active-10%
4. Ease of Payments-Passive-5%
Rest 15 Percent could be allocated to third-party signals like congestions or spending during commutes which can enhance tipping and additional revenue for riders
For more solutions to cases such as Uber Applying Machine Learning to Improve the Customer Pickup Experience, refer to our case solutions here
How would you improve the pickup experience at venues such as sporting events and concerts, which typically see temporary surges in demand for rides, as well as temporary parking restrictions and traffic congestion?
1. For Areas with high-density pick-ups they should be able to locate specific pinpoints
2. Voice Commands should be activated- Easier for rider and driver
3. Set of highly used locations can be used bookmarked
4. Telemetry data from various cars should be used so that the optimal route is taken by the rider
5. Uber should have a live video blog feed for the ride experience like the cleanliness of the car or the behavior of the driver- can be used for SOS calls as well
6. GPS should not be the only governing criteria. They should use other locational factors like the location of the mobile phones and map it if the user is stationary. It will help them pinpoint better
7. For Wrong pickups qualitative measures should be used like are you satisfied with the pickup
8. Narrowing down on frequently used points by the riders. Visual Markers like photographs should be made a feature of the app
9. Congestion data from google maps API should be pre-fed to the drivers so that they can avoid peak hour rush
10. Additional User Dissatisfiers like tickets should be resolved using semantics
For more solutions to cases such as Uber Applying Machine Learning to Improve the Customer Pickup Experience, refer to our case solutions here
Discuss the steps involved in setting up an ML model for automating pickups at scale. Use the framework of the seven-step model in the case (Exhibit 7) to elaborate how Uber should apply this framework to the ML model
Step 1: Define the Business Problem to be Solved:
Enhancing Pickup Experience
Step 2: Evaluate Machine Learning is important
Definitely. Uber operates in more than 400 cities globally and millions of riders book. So problems and solutions can be universal and scale can be achieved only by using ML and AI
Step3: Gather and Label Data
1. Telemetry Data
2. ETA vs ATA
3. Data from Google API
4. Wait time
5. App Crashes in congestion
6. Customer Behavioral Feedback to find problem geographical areas
7. Multi-Level GPS – Data from both mobile of the user and defined by the user should be mapped by a randomness algorithm. If the match is more than the driver should be mapped, else rider should give voice instructions
8. Many other subsidiary points like – Is the driver and user happy and so on?
Step 4: Process Data and Define Features
Use algorithms to pinpoint exact locations
See Surge and pricing algorithms
Additional Features like – Happiness meters, Alternate navigation routes, Incentivize drivers, and so on
Step 5: Select the ML approach
Use Telemetry data to initially supervise and based on the outputs and accuracy it can be used in an unsupervised environment
Step 6: Engineer Features and Improve Model
A host of additional features can be introduced as discussed in point 4
Additional revenue-generating mechanisms like VAS as discussed in question 5
Step 7: Make Decision and Design User Experience:
The shift of perspectives from 2016 onwards can be further moved towards defining the journey experience. It will help them in creating a host of other services and generate more revenue as the rider spends the most time during the commute. Music, Content, or simply conversations can enhance tipping and enhance revenues for Uber
For more solutions to cases such as Uber Applying Machine Learning to Improve the Customer Pickup Experience, refer to our case solutions here
We have given high-quality solutions to over 50 Case Studies on our Website. We would be thankful if you share it within your network 🙂
If you intend to share your experience regarding the solution, please write us a query here
Follow us on LinkedIn or Quora.
Samrat is a Delhi-based MBA from the Indian Institute of Management. He is a Strategy, AI, and Marketing Enthusiast and passionately writes about core and emerging topics in Management studies. Reach out to his LinkedIn for a discussion or follow his Quora Page