Post on 11-Apr-2017
transcript
Ann Johnson, CEO and Co-Founder of inter|ana
Tweet @stimmlet and @interanacorp
Talking to Your Digital Customers
Your customers are talking to you in the most dependable way – actions always speak louder than words.
What kinds of things do actions tell you?
• How well is your product working?
• What features do they like?
• What features should you add?
• What marketing campaigns do they respond to?
• Is something broken in your app?
How data reveals actions
What did they do?
Why did they do it?
What will they do next?
How to change what they do?
Case Study: Tinder
• Today, Tinder operates in 196 countries, whose users
generate an average of 1.8 billion swipes and create 26
million matches per day.
Tinder’s Infrastructure
• Started with Hadoop
• Slow to answer new questions
• Only available to trained data experts
• Changed to Self-serve solution
• Fast at scale
• Easy data model
• Graphical Interface
• Behavior support
Data is not a magical temple – it’s for everyone
Retention Analysis
• What parts of the funnel have good and bad conversion?
• How are conversion rates changing over time?
• How is churn changing over time?
• What users are likely to churn?
Matching Tinder to the User
• Data shows different users use Tinder differently
• Genders
• Orientation
• Age
• Location
• Tinder iterates on small tweaks to its algorithms to find
the best matches
• Brazil’s youth may use the app to meet new friends
• 2014 World Cup boosted use in Brazil by 50%
• S. and U.K. users 25-34 use the app to meet new people for
travel, dating, and marriage
How Should I Improve My Product?
• Should a feature be promoted?
• Should a feature be removed?
Do users like a feature? Only if they use it!
A Happy Community of Tinder Users
• A small percentage of users were swiping right on every profile
• Analyzed data to understand this behavior and limit it
• Quality of matches has increased dramatically
• In September of 2015, Tinder launched Super Like
• Using data to understand its effects:
• Product adoption
• Product usability
• Quality of matches
• Effectiveness of marketing campaigns
Is my product working?
• Counts of errors
• Latency
• Much more subtle things – If people a love a feature on
iOS but never touch it on Android, maybe your Android
implementation has a bug.
• If a platform has shorter sessions, there might be a
problem lurking
Tinder Example
• Tinder received reports from a small group of users that
Tinder worked on WiFi but not 4G
• Tinder diagnosed the root cause from user data: affected
users were in the same region and shared the same
carrier.
• Tinder worked with the carrier to fix their routing issue
and restored 4G service to its users.
Not all problems can be caught in testing, so real-world
data is essential
Tinder Marketing uses Data
• Effects of press coverage
• Response to campaigns across demographics and
locations
• Real-time feedback to double-down on things that work
and stop things that don’t
Kyle Miller, Marketing Manager at Tinder, says, “I would never consider myself a data person, but now I feel like I have the ability to accomplish all of my data-driven tasks.”
It’s everyone’s job to listen to the customer
It’s everyone’s job to look at the data
50% of Tinder employees have daily access to
data
Flexible
Problem: Many data tools are built to answer only one
question.
• Learning from data is an ongoing process. Not a one-off.
• Slice and dice across arbitrary dimensions. And beyond.
• Don’t decide what is important beforehand, in ETL, in
indexes, in schemas. Decide at read time.
Accessible
Problem: Many data tools require extensive training to use.
• Visual, simple, and interactive self-service solutions enable
broad adoption.
• Make it easy to know what data is available.
• Remove friction for the business user. Don’t rely on data
specialist to answer simple questions.
• Sharing example queries help spark curiosity.
Scalable
• Problem: Many data tools require data to be downsized
• RAW data analysis. ALL your data is available.
• Tools shouldn’t break as data volumes increase
• Questions should be answered at interactive speeds.
Transparent
Problem: Dashboards and reports can be misleading
• “See the math” – Where did this result come from?
• Data ingest processes, ETL, can hide calculations from
the end user
• Enable analysis down to row level detail. No aggregation
or summarization boundaries.
Summary - FAST
• Flexible: You shouldn’t have to know the question
beforehand. You should be empowered to ask “the next
question.”
• Accessible: Your data tools should need minimal
training
• Scalable: You need row level access to all your data to
paint the full picture.
• Transparent: Data consumers need to understand
where the numbers came from.