Post on 15-Apr-2017
Product optimization through Big Data Analytics
and Machine Learning
Inmar Givoni,
Ph.D. (Machine Learning)
VP, Big Data
About Me
● Ph.D. Computer Science (Machine Learning)
from University of Toronto ● Microsoft Research-Cambridge, SV ● VP of Big Data @ Kobo ● Multiple publications, patents ● Public speaking, outreach ● inmarg.net for more
The importance of testing everything
Gut feelings are often wrong
But post-hoc analysis can lead to great insights
The importance of testing everything
● Personal market research is useful but...
● Sample biases ● Do users always ‘know’ what they want? ● Why ask when you can get the answer?
random splits
right test
statistical
significance
minimal
detection rate
sample size
Hypothesis (A/B) Testing
Hypothesis (A/B) Testing
• The right way to make business
decisions
• Put reliable numbers on products
– cannibalism, time of x effects
• Can be integrated via 3rd party
technologies
– but do A/A test!
Personal Recos
● Important driver of sales & engagement
● Continuous analysis of user behaviour patterns
● Use domain specific knowledge
● Continuously improve
From Analytics to Data Products
● Email time of day
analysis →
trigger individually
● BI on best website assets
→
automatically optimizing
platform