5 - 6 2 0 -0 3 4
REV: F EBRUARY 18, 2022
T E A C H I N G N OT E
Innovation at Uber: The Launch of Express POOL
Case Synopsis
The case “Innovation at Uber: The Launch of Express POOL” focuses on Uber’s approach to product
innovation, primarily how the company uses its technology platform to continuously assess and
modify its ridesharing products. Using the illustrative example of a new product launch, the case
describes three major types of data generation tools used for product innovation: survey evidence,
simulation, and experimentation. Within experimentation, the case offers an opportunity to discuss
how experiments need to be carefully designed to derive unbiased estimates of the effects of new
product features. In particular, the case allows for a comparison among several types of experimental
designs (user A/B, switchback, and synthetic control experiments) that Uber deploys to quantify the
impacts of product changes.
Set in 2018, the case follows a group of data scientists, engineers, product managers, and product
operations specialists as they develop and launch Express POOL (Express). The Express product offers
a reduced fare to passengers willing to carpool, wait a few extra minutes before being matched to a
driver, and then walk a short distance to/from their pick-up and drop-off points. Before Express, all
Uber products matched each ride request to the closest available driver within seconds. With Express,
ride requests occurring within a short time window are pooled together and matched with the available
cars. This increase in matching efficiency comes at the expense of making customers wait before a
match is confirmed and walk from their desired location, so Uber’s data scientists are exploring how
to balance efficiency with customer experience.
The case describes the sequence of events that led to the launch of Express POOL. Given the strategic
priority to improve Uber POOL to make it profitable, a team of engineers, data scientists, and product
managers set out to collect data and test hypotheses to identify which product tweaks could improve
efficiency without hurting customer experience too much. The case describes the collection of surveys,
the implementation of simulations, and the launch of two different experiments. The first experiment
(launch experiment), which started two weeks earlier, had launched Express in six cities while another
six cities were held as a control group. The launch had set the passenger’s maximum match wait time at
2 minutes (i.e., a passenger would wait up to 2 minutes before being matched to a driver). At the time
of the case results come in from a second experiment (match wait time experiment) measuring the benefits
,620-034 Teaching Note—Innovation at Uber: The Launch of Express POOL
of increasing the maximum match wait times from 2 to 5 minutes. Increasing match wait times
increased passenger cancellations but reduced Uber’s costs per ride. What was important is not the
direction of these effects—all ex-ante fairly predictable—but their magnitude, which was essential to
quantify the trade-off between efficiency and customer experience. The product managers must now
decide whether to keep the launch experiment unmodified for the remaining three weeks or increase
Express match wait times to 5 minutes in the middle of the experiment.
Case Positioning and Purpose
At Harvard Business School (HBS), the case has been used in a first-year required course called
Technology and Operations Management (TOM) in the module on product development and
innovation. The case has also been used in elective and executive education courses focused on data
science, innovation, experimentation, and product development. Finally, given the additional data
supplement (HBS courseware no. 619-702), the case can also be a medium to teach data analytics
techniques – including exploratory data analysis, linear regressions, and hypothesis testing – and the
role of experimentation for evaluating the impact of product or process changes. If instructors use the
case to discuss data-driven product innovation, it can fit well in a sequence of three cases, preceded by
“Team New Zealand” (HBS no. 697-040) and followed by “Booking.com” (HBS no. 619-015). Team
New Zeland sets up the basics of simulation and experimentation and their trade-offs, Uber adds
surveys and presents the complexities of experimentation when two user groups interact on a platform,
Booking.com raises important leadership challenges for companies that make large-scale
experimentation a critical component of every decision.
Learning Objectives
The case allows classroom discussion to focus on surveys, simulations, and particularly experiments
to support product innovation. Students learn to appreciate how platforms rely on a data-driven
approach to innovation and how they can use the scientific method as an integral part of their
operations. Despite the opportunities that data provide to generate and test hypotheses, evaluating the
impact of innovation is not straightforward, even at platform companies like Uber. The case discussion
helps deliver the following learning objectives:
• Identify the trade-off between the quality and quantity of transactions on a platform-based
firm. Innovation at Uber is often designed to increase transactions (number of completed rides)
and improve their average quality (for example, how long it takes for the driver to pick up the
passenger). Although both the quantity and quality of exchanges on a platform positively affect
platform growth, platforms often face a trade-off between quantity and quality. Matching a
buyer to its ideal seller (at Uber, it could be the driver who’s closest to the passenger) might
prevent valuable matches arising among the other buyers and sellers. A platform may thus
prefer to match that one buyer to a slightly less ideal seller, perhaps for a lower price, to create
more matches.
• Identify the trade-offs inherent to serving multiple user groups in a multi-sided platform.
Given that buyers’ choices to use a platform are directly affected by sellers’ participation
decisions, product features must meet the often conflicting needs of both sides of a platform
company—buyers and sellers—while boosting the company’s revenues. Product alterations at
Uber must make the service more attractive for one user group (e.g., passengers) without
making the experience of the other group (e.g., drivers) significantly less convenient, while also
allowing Uber itself to capture value.
2
,Teaching Note—Innovation at Uber: The Launch of Express POOL 620-034
• Compare the pros and cons of multiple approaches to gather data to guide innovation in a
platform-based firm. This case allows participants to appreciate how data-driven Uber’s
product development process is. Going from idea to launch and post-launch improvements,
Uber generates insights from surveys, simulations, and experiments. Lessons emerge from the
comparison of different data sources (and how to use them jointly) to guide product
development.
• Understand how experimentation can be used to make operational decisions in platform-
based companies. A/B tests are often powerfully simple because they yield robust
experimental evidence by randomly assigning some users to product feature A and others to
product feature B. But when users interact with each other, the decisions of one agent affect the
experience and the decisions of many other agents using the platform, which would
contaminate the results of a naïve A/B test. Thus, experiments need to be carefully designed to
derive unbiased estimates of the effects of new product features on the platform as a whole. The
case allows students to compare different experiment designs that randomize at the individual
user, hour of the day, or city level. The comparison across experiment designs is largely based
on two dimensions of the estimates of the effects: precision (low variance) and consistency (lack
of bias).
• Apply data analytics to drive innovation. The case is paired with a supplementary dataset
containing the results of a real experiment, which provides additional opportunities for
participants to put their analytical skills to use in testing hypotheses and uncovering the impact
of the increase in match wait times on market efficiency and customer satisfaction metrics.
Materials
Instructors have access to the following materials:
• The case titled “Innovation at Uber: The Launch of Express POOL” (HBS no. 619-003).
• The supplementary dataset from Uber (HBS courseware no. 619-702), which contains data from
a switchback experiment conducted in Boston to compare the Express product’s performance
with a 5-minute match wait time (treatment) against the default 2-minute match wait time
(control). A switchback experiment typically runs for two weeks in a single city. Each day is
divided into nine 160-minute time intervals. These time intervals are called switchback periods
because Uber switches between the treatment and the control condition over time. So if between
7:00 a.m. and 9:40 a.m., the algorithm makes passengers wait up to 2 minutes before being
matched, in the following 9:40 a.m.-12:20 p.m. period, passengers wait up to 5 minutes. The
design has treatment and control periods alternate within the course of a single day, and
alternate for the hour of the day across consecutive days (if today 7:00-9:40 a.m. is under
treatment, tomorrow 7:00-9:40 a.m. will be under control) and for the day of the week across
the two weeks (if Monday 7:00-9:40 a.m. is under treatment this week, Monday 7:00-9:40 a.m.
will be under control next week). Before the class discussion, instructors can distribute the
dataset for analysis, allowing participants to draw their own conclusions about the trade-offs
of maintaining the standard match wait times or increasing them.
• The 13-minute supplementary video (HBS courseware no. 620-702), which features Data
Science Manager Duncan Gilchrist, one of the case protagonists, along with Data Science
Manager Eoin O’Mahony (who does not appear in the case). In the video, Gilchrist and
O’Mahony answer questions from students in the HBS classroom. It begins with an update,
3
, 620-034 Teaching Note—Innovation at Uber: The Launch of Express POOL
wherein Gilchrist provides the resolution to the case dilemma—Uber decided to keep the
Express product fixed at 2-minute match wait times for the rest of the 5-week launch
experiment, instead of increasing match wait times to 5 minutes. The video goes on to provide
Gilchrist and O’Mahony’s perspective on a number of topics, ranging from ensuring fairness
during experimentation to persuading customers to change their behaviors. Instructors may
choose to use the video in class, either in its entirety or as individual clips of interest. See
Appendix A for a transcript of the video.
• If the case is taught in a data analytics course, instructors can give the following assignment
focused on hypothesis testing:
• The assignment (HBS courseware no. 622-053) asks students to generate variables, select
subsets of the data, and run hypothesis tests to compare treatment and control groups in
the switchback experiment separately for rush and non-rush hours.
• The solutions document (HBS courseware no. 622-054) provides solutions to the
assignment. The results of the hypothesis tests required to complete the assignment are
presented both in R and Excel.
• The 26-minute assignment solutions video (HBS courseware no. 622-704) provides
solutions to the data analytics assignment using the R computing software.
Lastly, if the case is taught within a product development module, instructors may want to assign
the technical note entitled “Product Development Fundamentals” (HBS no. 617-024) as a complement
to the Uber case. If the case is instead taught with data analytics in mind, instructors may want to assign
the technical note entitled “Comparing Two Groups: Sampling and T-Testing” (HBS no. 621-044) as an
introduction to statistical inference and hypothesis testing.
Suggested Assignment Questions
To guide participants in reading the case and preparing for the class discussion, instructors may
want to assign the questions below. (Note: Question #4 presumes that participants have been provided
with the supplementary dataset [HBS courseware no. 619-702] prior to class.) If instructors plan to use
Question #5 with students who have not received the supplementary dataset, instructors will need to
provide them with the results of the data analysis such as histograms (Slides 1 and 2 in TN Exhibit 2),
or t-tests (Slide 3 in TN Exhibit 2).
1. How does Uber innovate? What is the role of data science for innovation at Uber?
2. Why does Uber have so many different ways to run experiments? What are the pros and cons
of each type?
3. Evaluate the Express development project. What did Uber do well on this project? What could
have been improved?
4. Considering the supplementary dataset, what is the effect of extending match wait times from
two to five minutes on the total number of shared rides completed (that is, rides taken via both
the existing shared rides product—UberPOOL—and the new shared rides product, Express),
the proportion of shared rides that were matched, and driver payout per trip?
4
REV: F EBRUARY 18, 2022
T E A C H I N G N OT E
Innovation at Uber: The Launch of Express POOL
Case Synopsis
The case “Innovation at Uber: The Launch of Express POOL” focuses on Uber’s approach to product
innovation, primarily how the company uses its technology platform to continuously assess and
modify its ridesharing products. Using the illustrative example of a new product launch, the case
describes three major types of data generation tools used for product innovation: survey evidence,
simulation, and experimentation. Within experimentation, the case offers an opportunity to discuss
how experiments need to be carefully designed to derive unbiased estimates of the effects of new
product features. In particular, the case allows for a comparison among several types of experimental
designs (user A/B, switchback, and synthetic control experiments) that Uber deploys to quantify the
impacts of product changes.
Set in 2018, the case follows a group of data scientists, engineers, product managers, and product
operations specialists as they develop and launch Express POOL (Express). The Express product offers
a reduced fare to passengers willing to carpool, wait a few extra minutes before being matched to a
driver, and then walk a short distance to/from their pick-up and drop-off points. Before Express, all
Uber products matched each ride request to the closest available driver within seconds. With Express,
ride requests occurring within a short time window are pooled together and matched with the available
cars. This increase in matching efficiency comes at the expense of making customers wait before a
match is confirmed and walk from their desired location, so Uber’s data scientists are exploring how
to balance efficiency with customer experience.
The case describes the sequence of events that led to the launch of Express POOL. Given the strategic
priority to improve Uber POOL to make it profitable, a team of engineers, data scientists, and product
managers set out to collect data and test hypotheses to identify which product tweaks could improve
efficiency without hurting customer experience too much. The case describes the collection of surveys,
the implementation of simulations, and the launch of two different experiments. The first experiment
(launch experiment), which started two weeks earlier, had launched Express in six cities while another
six cities were held as a control group. The launch had set the passenger’s maximum match wait time at
2 minutes (i.e., a passenger would wait up to 2 minutes before being matched to a driver). At the time
of the case results come in from a second experiment (match wait time experiment) measuring the benefits
,620-034 Teaching Note—Innovation at Uber: The Launch of Express POOL
of increasing the maximum match wait times from 2 to 5 minutes. Increasing match wait times
increased passenger cancellations but reduced Uber’s costs per ride. What was important is not the
direction of these effects—all ex-ante fairly predictable—but their magnitude, which was essential to
quantify the trade-off between efficiency and customer experience. The product managers must now
decide whether to keep the launch experiment unmodified for the remaining three weeks or increase
Express match wait times to 5 minutes in the middle of the experiment.
Case Positioning and Purpose
At Harvard Business School (HBS), the case has been used in a first-year required course called
Technology and Operations Management (TOM) in the module on product development and
innovation. The case has also been used in elective and executive education courses focused on data
science, innovation, experimentation, and product development. Finally, given the additional data
supplement (HBS courseware no. 619-702), the case can also be a medium to teach data analytics
techniques – including exploratory data analysis, linear regressions, and hypothesis testing – and the
role of experimentation for evaluating the impact of product or process changes. If instructors use the
case to discuss data-driven product innovation, it can fit well in a sequence of three cases, preceded by
“Team New Zealand” (HBS no. 697-040) and followed by “Booking.com” (HBS no. 619-015). Team
New Zeland sets up the basics of simulation and experimentation and their trade-offs, Uber adds
surveys and presents the complexities of experimentation when two user groups interact on a platform,
Booking.com raises important leadership challenges for companies that make large-scale
experimentation a critical component of every decision.
Learning Objectives
The case allows classroom discussion to focus on surveys, simulations, and particularly experiments
to support product innovation. Students learn to appreciate how platforms rely on a data-driven
approach to innovation and how they can use the scientific method as an integral part of their
operations. Despite the opportunities that data provide to generate and test hypotheses, evaluating the
impact of innovation is not straightforward, even at platform companies like Uber. The case discussion
helps deliver the following learning objectives:
• Identify the trade-off between the quality and quantity of transactions on a platform-based
firm. Innovation at Uber is often designed to increase transactions (number of completed rides)
and improve their average quality (for example, how long it takes for the driver to pick up the
passenger). Although both the quantity and quality of exchanges on a platform positively affect
platform growth, platforms often face a trade-off between quantity and quality. Matching a
buyer to its ideal seller (at Uber, it could be the driver who’s closest to the passenger) might
prevent valuable matches arising among the other buyers and sellers. A platform may thus
prefer to match that one buyer to a slightly less ideal seller, perhaps for a lower price, to create
more matches.
• Identify the trade-offs inherent to serving multiple user groups in a multi-sided platform.
Given that buyers’ choices to use a platform are directly affected by sellers’ participation
decisions, product features must meet the often conflicting needs of both sides of a platform
company—buyers and sellers—while boosting the company’s revenues. Product alterations at
Uber must make the service more attractive for one user group (e.g., passengers) without
making the experience of the other group (e.g., drivers) significantly less convenient, while also
allowing Uber itself to capture value.
2
,Teaching Note—Innovation at Uber: The Launch of Express POOL 620-034
• Compare the pros and cons of multiple approaches to gather data to guide innovation in a
platform-based firm. This case allows participants to appreciate how data-driven Uber’s
product development process is. Going from idea to launch and post-launch improvements,
Uber generates insights from surveys, simulations, and experiments. Lessons emerge from the
comparison of different data sources (and how to use them jointly) to guide product
development.
• Understand how experimentation can be used to make operational decisions in platform-
based companies. A/B tests are often powerfully simple because they yield robust
experimental evidence by randomly assigning some users to product feature A and others to
product feature B. But when users interact with each other, the decisions of one agent affect the
experience and the decisions of many other agents using the platform, which would
contaminate the results of a naïve A/B test. Thus, experiments need to be carefully designed to
derive unbiased estimates of the effects of new product features on the platform as a whole. The
case allows students to compare different experiment designs that randomize at the individual
user, hour of the day, or city level. The comparison across experiment designs is largely based
on two dimensions of the estimates of the effects: precision (low variance) and consistency (lack
of bias).
• Apply data analytics to drive innovation. The case is paired with a supplementary dataset
containing the results of a real experiment, which provides additional opportunities for
participants to put their analytical skills to use in testing hypotheses and uncovering the impact
of the increase in match wait times on market efficiency and customer satisfaction metrics.
Materials
Instructors have access to the following materials:
• The case titled “Innovation at Uber: The Launch of Express POOL” (HBS no. 619-003).
• The supplementary dataset from Uber (HBS courseware no. 619-702), which contains data from
a switchback experiment conducted in Boston to compare the Express product’s performance
with a 5-minute match wait time (treatment) against the default 2-minute match wait time
(control). A switchback experiment typically runs for two weeks in a single city. Each day is
divided into nine 160-minute time intervals. These time intervals are called switchback periods
because Uber switches between the treatment and the control condition over time. So if between
7:00 a.m. and 9:40 a.m., the algorithm makes passengers wait up to 2 minutes before being
matched, in the following 9:40 a.m.-12:20 p.m. period, passengers wait up to 5 minutes. The
design has treatment and control periods alternate within the course of a single day, and
alternate for the hour of the day across consecutive days (if today 7:00-9:40 a.m. is under
treatment, tomorrow 7:00-9:40 a.m. will be under control) and for the day of the week across
the two weeks (if Monday 7:00-9:40 a.m. is under treatment this week, Monday 7:00-9:40 a.m.
will be under control next week). Before the class discussion, instructors can distribute the
dataset for analysis, allowing participants to draw their own conclusions about the trade-offs
of maintaining the standard match wait times or increasing them.
• The 13-minute supplementary video (HBS courseware no. 620-702), which features Data
Science Manager Duncan Gilchrist, one of the case protagonists, along with Data Science
Manager Eoin O’Mahony (who does not appear in the case). In the video, Gilchrist and
O’Mahony answer questions from students in the HBS classroom. It begins with an update,
3
, 620-034 Teaching Note—Innovation at Uber: The Launch of Express POOL
wherein Gilchrist provides the resolution to the case dilemma—Uber decided to keep the
Express product fixed at 2-minute match wait times for the rest of the 5-week launch
experiment, instead of increasing match wait times to 5 minutes. The video goes on to provide
Gilchrist and O’Mahony’s perspective on a number of topics, ranging from ensuring fairness
during experimentation to persuading customers to change their behaviors. Instructors may
choose to use the video in class, either in its entirety or as individual clips of interest. See
Appendix A for a transcript of the video.
• If the case is taught in a data analytics course, instructors can give the following assignment
focused on hypothesis testing:
• The assignment (HBS courseware no. 622-053) asks students to generate variables, select
subsets of the data, and run hypothesis tests to compare treatment and control groups in
the switchback experiment separately for rush and non-rush hours.
• The solutions document (HBS courseware no. 622-054) provides solutions to the
assignment. The results of the hypothesis tests required to complete the assignment are
presented both in R and Excel.
• The 26-minute assignment solutions video (HBS courseware no. 622-704) provides
solutions to the data analytics assignment using the R computing software.
Lastly, if the case is taught within a product development module, instructors may want to assign
the technical note entitled “Product Development Fundamentals” (HBS no. 617-024) as a complement
to the Uber case. If the case is instead taught with data analytics in mind, instructors may want to assign
the technical note entitled “Comparing Two Groups: Sampling and T-Testing” (HBS no. 621-044) as an
introduction to statistical inference and hypothesis testing.
Suggested Assignment Questions
To guide participants in reading the case and preparing for the class discussion, instructors may
want to assign the questions below. (Note: Question #4 presumes that participants have been provided
with the supplementary dataset [HBS courseware no. 619-702] prior to class.) If instructors plan to use
Question #5 with students who have not received the supplementary dataset, instructors will need to
provide them with the results of the data analysis such as histograms (Slides 1 and 2 in TN Exhibit 2),
or t-tests (Slide 3 in TN Exhibit 2).
1. How does Uber innovate? What is the role of data science for innovation at Uber?
2. Why does Uber have so many different ways to run experiments? What are the pros and cons
of each type?
3. Evaluate the Express development project. What did Uber do well on this project? What could
have been improved?
4. Considering the supplementary dataset, what is the effect of extending match wait times from
two to five minutes on the total number of shared rides completed (that is, rides taken via both
the existing shared rides product—UberPOOL—and the new shared rides product, Express),
the proportion of shared rides that were matched, and driver payout per trip?
4
