Develop data-driven hypotheses for new A/B tests
Learn what was the data-driven hypothesis behind FB stories, Netflix free trial, Youtube Premium, etc.
Design actionable metrics
When to use average-based vs threshold-based metrics? How to identify those thresholds in metrics such as percentage of users with 7 friends in 10 days?
Evaluate metric trade-offs
How to evaluate changes that positively affect some key metrics, but negatively affect others? Which strategy to use to decide whether a product change is good or not in those cases?
How to design statistically sound A/B tests? How to choose test traffic percentage? How to deal with novelty effect, connected users, randomization bugs, or other typical issues in tech A/B tests?
How to deal with actual tech data, such as sudden spikes or drops, bugs, unbalanced metrics, non-random missing data, A/B test results, conversion data, etc.
How to use machine learning to fight frauds, build data products, or develop product insights? How to evaluate the impact of new machine learning models into production?
12 Sections, >130 lessons, exercises, and product case studies
Real product data science problems using tech company tables, code in R and Python
Bahar Bazargan – got a job as a Data Scientist @ Facebook – US
Product Case Study – Insights
Should a free trial require credit card info? Where in the funnel should the highest friction point be?See the lesson
When to choose a logistic regression, how to interpret it, and how to use its output to come up with product ideasSee the lesson
A/B tests: Sample Size
How to estimate for how many days you should run an A/B test, from both a statistical and business perspectiveSee the lesson
A/B tests: Novelty Effect
How to understand if a test is affected by novelty effect and how to deal with it in practiceSee the lesson
A/B Tests: Randomization
The crucial step of making sure that test and control are properly randomizedSee the lesson
Kenny Tang – Data Scientist @ SmarTone Telecom Lim – Hong Kong
1) Product Sense via Machine Learning
Practical examples on to use regressions, ML, partial dependence plots, and rulefit to drive product development and come up with ideas for new product features
2) Product and Metrics – Case Studies
18 case studies on how to design actionable metrics, understand what drives them, and figure out how to improve them via new product features
In depth practical exercise on how to use machine learning to build a data product personalized at the user level. This is the framework used to optimize almost all data products
4) Unbalanced Classes
Almost all tech company data have unbalanced classes, i.e. fraud, ad clicks, conversation rate, email clicks, etc. These exercise explain how to deal with that
5) Missing data in tech
Most of missing data in tech are non-random, i.e. users choose to not provide certain information about themselves.These lessons explain how to deal with biased missing data. Include Uber and Airbnb case studies
6) Fraud – Case Studies
Fraud is one of the most common data science application. These case studies explain how to set up the problem from a ML standpoint as well the how to build a product around it
7) A/B Testing – Practice
A series of lessons covering all that’s needed to know about A/B testing. Includes statistical inference relevant theory as well as very practical tech problems (novelty effect, randomization, sample size, testing by market, etc.)
8) A/B testing – Case Studies
12 case studies describing how top tech companies design their A/B tests. They focus on the most common issues tech companies face, like how to test in social networks or marketplaces, how to estimate long term effects, when A/B tests fail, etc.
9) New Product Case Studies
A new list of product case studies. They focus on trade-offs when evaluating multiple metrics at the same time (i.e. what to do if some key metrics are up and some down) as well as explaining the thought process behind using data to choose whether testing a totally new feature or product
10) Collection of tech company blog posts/case studies
This is a collection of company write-ups, tutorials, and blog posts. Includes Airbnb, FB, Linkedin, Google, Netflix and many more other companies describing how they design A/B tests and use DS to drive product development
11) Data challenges with solutions
They come from the “Collection of data science takehome challenges” book. They touch all the topics taught in the course. All challenges come with full solution in R and Python
12) Metrics via SQL
SQL exercises to create some of the most common metrics used by tech companies. I.e., identify power users or group by users based on their cross-device history. Queries rely heavily on window functions
$ 1625 2 monthly payments
☑ Lifetime access to course curriculum
Curriculum includes a mix of theoretical lessons, product case studies, and challenges with solution
☑ Unlimited 1:1 support from course author for 1 year
Any questions you have about the course material or anything related to product data science, you can send an email, skype chat, or share a Google doc with all the questions. You will get a prompt reply
☑ Personalized feedback
Send your solution for all the exercises in the course. You’ll get a detailed feedback on your work
Doudou Tang – Data Scientist @ Booking.com – UK
I am buying this with my employee training budget, do you provide a certificate or invoice that I can show to my employer?
Yeah, definitely. Can provide certificate, invoice, or anything else you need.
Yeah, all those challenges and product questions are here too. However, this course includes much more.
Its main focus is on teaching product data science via a combination of theoretical lessons and practical examples. The challenges then come at the end to make sure things were learned properly.
Also, the course includes many new product questions like, for instance, the first one in the samples.
And all the challenges have a solution, not just 4.
What’s the main difference between this course and the other million data science courses available?
Frankly speaking, most data science courses are not particularly useful because they have little/nothing to do with what data scientists actually do at work.
Firstly, data scientists don’t spend their time over-tuning a fancy model to marginally improve its performance. In fact, most of data science work is about looking at the data to come up with product ideas and properly designing A/B tests.
And the few applied courses tend to be so simple to the point of being highly unrealistic. I.e., most A/B test courses will teach you to randomly split users, run the two versions of the site, and check the results. But that strategy almost never works for several reasons (test and control would never be independent in social networks or marketplaces, can’t estimate novelty effect, etc.).
This course will teach you how product data science is actually done at top tech companies. The course uses data that look exactly like tech company tables (including wrong entries, non-random missing values, A/B tests, etc.). And then actual data science projects are built on top of those data.