Adam Mosseri - product designer, Facebook mossier@facebook.com
Data Informed Design (Not Data-Driven) Adam: 2 years at Facebook Who makes the decisions: small teams Structure of core team - how FB works
Product designers: product strategy, ix design, visual design
Researchers
Engineers
Product manager- also responsibility for quality of the product, mini CEOs Record 4 terabytes a day
Data helps us understand how users use the product, how to improve Photo uploading story: 200m? uploaded a day, largest photo site on the web How we use data - Use data to optimize a workflow
Hypothesis generation
Hypothesis evaluation *** Waterfall analysis ***
Of users who try, only 87% reach steady state
57 52 up,oaf
4% successful 85% only selected one photo How we use data - Sanity check How we use data - Evaluate designs Healthy skepticism of being overly data driven:
Very difficult for any set of measures to fully capture what you think is important Attempts at quantifying engagement
Raw reads and writes - but not all writes are created equal
85% of writes generated by 20% of users. If we optimize for heavy users, sub optimizes light users Overreacting to data can lead to micro-optimizations. (but missing the whole)
Optimizing for a local maximum, but missing a bigger structural change. Miss disruptive opportunities.
Dangers of chasing local maximum Why we're wary of data: Real innovation invariably involves disruption.
The greatest risk is taking no risk Group with 20,000 members: I automatically hate the new Facebook homepage
Data Informed Design (Not Data-Driven) Adam: 2 years at Facebook Who makes the decisions: small teams Structure of core team - how FB works
Product designers: product strategy, ix design, visual design
Researchers
Engineers
Product manager- also responsibility for quality of the product, mini CEOs Record 4 terabytes a day
Data helps us understand how users use the product, how to improve Photo uploading story: 200m? uploaded a day, largest photo site on the web How we use data - Use data to optimize a workflow
Hypothesis generation
Hypothesis evaluation *** Waterfall analysis ***
Of users who try, only 87% reach steady state
57 52 up,oaf
4% successful 85% only selected one photo How we use data - Sanity check How we use data - Evaluate designs Healthy skepticism of being overly data driven:
Very difficult for any set of measures to fully capture what you think is important Attempts at quantifying engagement
Raw reads and writes - but not all writes are created equal
85% of writes generated by 20% of users. If we optimize for heavy users, sub optimizes light users Overreacting to data can lead to micro-optimizations. (but missing the whole)
Optimizing for a local maximum, but missing a bigger structural change. Miss disruptive opportunities.
Dangers of chasing local maximum Why we're wary of data: Real innovation invariably involves disruption.
The greatest risk is taking no risk Group with 20,000 members: I automatically hate the new Facebook homepage
No comments:
Post a Comment