This document discusses building an analytical framework with three steps: define, describe, and deploy. It provides examples from Lyft of how they used data science to grow rides by defining problems like delays in driver pickups, describing data through exploration and visualization, and deploying solutions through experimentation. The framework emphasizes clear problem definitions, learning from data, and activating insights through experiment design and measurement to inform business decisions.
3. Case Study: Rideshare
Data Science at Lyft
Lyft used data science to grow rides 10X from
30M rides in 2015 to 350M in 2018.
● Drivers can be delayed during
their pickup of passengers when a
ride is requested, resulting in an
arrival time that is later than
promised in the product
● The longer a driver is delayed in
picking up a passenger, the less
likely that passenger is to remain
equally active on the app,
resulting in fewer rides taken the
following period
Rider Experience
● Acknowledging this broken
promise to the passenger and
offering coupons as part of the
apology was an example of an
experimental intervention to
measure the impact on the
business
● A combined no cost and cost
intervention added $1.2M in ride
revenue within 30 days
4.
5. Define
What am I trying to solve?
...need to understand opportunities
within core functions and open
spaces to effectively prioritize
based on impact, which begins with
a problem statement.
Business Leaders
Data Dictionary
Data Measurement
Business Problem
Organizations need to
have a shared set of
terms with common
definitions across
business units in order to
communicate effectively
and measure impact.
The assets of an
organization’s data are
the items being
measured, and represent
opportunities for learning
which means an inventory
of them is essential for
action.
The clear articulation of
the problem an
organization is facing is
central to the ability to
think about it analytically,
which allows teams to
generate relevant areas
of investigation.
6. Define
What am I trying to solve?
Can we collect additional lead
information predictive of future
behavior so that we can optimize
interventions to accelerate and
increase activations?
Business Question
Driver applicants that have a friend or family member
driving with Lyft are more likely to activate and
activate earlier agnostic of channel.
65% increase in activation rate among applicants with
a friend driving for Lyft compared with applicants
without a friend.
7. Describe
What can I learn from my data.
...begins with how data is being
collected, what is being measured,
and what relationships am I
interested in for deeper
understanding and leading to
insights from a model.
Data Exploration
Visualization
Summary Statistics
Inference
Representing data
visually enhances our
ability to discover trends
and other relationships
essential to building a
base of knowledge about
the business.
Making progress toward
an improved organization
requires identifying the
current state including
what the current metrics
are to measure against.
Identifying relationships
and outcomes of interest
from modeled data
empowers an
organization with insights
to action on.
8. Describe
What can I learn from my data.
Learning
Can we leverage the excitement
and intent of our software trialers to
deliver high paid conversion rates?
Shorter trial durations capitalize on meeting the
expectations of high intent trial users and limiting the
opportunity for disappointment within the trial period.
56% increase in paid trial conversion among 14 and 7
day trial duration compared with the standard 30 day
trial.
GoToMeeting
Trial Conversion Rate
30 Day 14 Day 7 Day
9. Deploy
How will I action on my learnings?
...means informing tactics and
strategy gained through data
driven learnings when taking
actions on business questions,
including experimentation.
Activating on Evidence
Experiment Design
Audience
Impact
A/B tests offer
organizations a high
confidence method for
measuring impact on
business questions
including acquisition and
retention.
Experiments should have
a clearly targeted
audience based on the
inferences made during
the data exploration step.
Audience size and effect
can be used to estimate
opportunity size to the
business for planning and
prioritization.
10. Deploy
How will I action on my learnings?
...experimentation is the method
that gives the highest confidence
for an organization to measure the
impact of insights and adopt a
change that the business can rely
on.
Activating on Evidence
Measure
Recommend
Deploy
12. Analytic Products
What does a analytic product look like?
...is a foundational skill for analytic
thinking and offers a digestible way
for all levels of users to consume
information to inform business
performance.
Data Visualization
13. Analytic Products
What does a analytic product look like?
...is the highest confidence method
for actioning on a learning to
measure its impact a key metric of
the business.
Experimentation
14. Evidence In Action
Tools that empower.
...is an open source language that
allows programmers to build
secure, dynamic and personalized
data visualization to meet the
needs of analysts and business
leaders alike.
Shiny