Grammarly interviews
United States
Experiments
“When deciding on the size of these sets, one needs to balance the benefit of additional statistical power with the opportunity cost and business disruption that a large “go-dark” experiment can cause.”
Arik Motskin, Michael Haugan and Parth AidasaniGrammarly Data Science
Power against disruption
Decide whether one winner among 20 AI-written headlines should roll out.
ExampleMetrics and funnels
“top-of-funnel channels that drive views instead of clicks, such as digital video and TV, tend to be under-credited.”
Arik Motskin, Michael Haugan and Parth AidasaniGrammarly Data Science
Top-of-funnel channels are under-credited
Decide whether AI shopping assistants bring growth or only trials.
Example“we can improve our understanding of our media efficiency by considering incremental cost per new user instead of just cost per new user.”
Arik Motskin, Michael Haugan and Parth AidasaniGrammarly Data Science
The work itself
“It helps acquisition leaders better address the question “Where should my next dollar go?””
Arik Motskin, Michael Haugan and Parth AidasaniGrammarly Data Science
Where should the next dollar go?
Decide whether AI shopping assistants bring growth or only trials.
Example“Needing to coordinate across teams to make a change tends to introduce delays for delivering new features to users.”
Sasha SuhakGrammarly Engineering Blog author
Coordination slows delivery
Decide whether AI share cards that doubled sharing should roll out.
ExampleLast reviewed
