When a company goes through a hyper-growth phase, things move faster than you can control or even sometimes comprehend. This pushes teams to take shortcuts to facilitate faster data-driven decisions. If not prudently managed, this can result in a high “data debt,” which inhibits future growth opportunities.
During Pinterest’s hypergrowth journey over the past six years, the data team made conscious decisions about technology, infrastructure, and processes to ensure that the data infrastructure scaled with the user growth, teams had the right tools to derive insights from the data, decision-making velocity was not impacted, and data debt was kept under control.
Romit Jadhwani (BI Engineering Manager) deep dives into the different phases of Pinterest’s journey, focusing on questions asked, analytics and data solutions designed, and value added during each phase, offering a high-level blueprint for data teams to consider for their own hypergrowth journeys.
This presentation was delivered at Strata + Hadoop World Conference in San Jose, CA on March 16, 2017.
20. Actors
?
I am the ASKER.
I ask questions to help
make decisions.
I am the ANALYST.
I analyze the data
to identify insights
and trends.
21. Actors
?
I am the ASKER.
I ask questions to help
make decisions.
I am the ANALYST.
I analyze the data
to identify insights
and trends.
I am the ENGINEER.
I build systems and
tools.
26. Bigbang
What is happening?
What or “Awareness”
Why or “Understanding”
Constant need for data availability
Heroic effort by engineer/analyst
… high value, high debt
1
27. Bigbang
What is happening? What will help?
What or “Awareness”
Why or “Understanding”
Constant need for data availability
Heroic effort by engineer/analyst
… high value, high debt
KPIs
Hire data engineer/s
Be scrappy—launch and iterate
1
28. Bigbang
What is happening? What will help?
What or “Awareness”
Why or “Understanding”
Constant need for data availability
Heroic effort by engineer/analyst
… high value, high debt
Diminishing returns!
KPIs
Hire data engineer/s
Be scrappy—launch and iterate
1
35. We don’t have a good
idea. How can we find
out, Engineer?
Pedaltothemetal2
?
ASKER ANALYST
What is our most
profitable customer
segment?
36. We need to more data
signals
Pedaltothemetal2
?
ASKER ANALYST ENGINEER
What is our most
profitable customer
segment?
We don’t have a good
idea. How can we find
out, Engineer?
37. Pedaltothemetal
What is happening?
What or “Awareness”
Why or “Understanding”
How or “Process”
Constant need for data breadth and depth
Different “slices” of the company
… still high value, still high debt
2
38. Pedaltothemetal
What is happening? What will help?
What or “Awareness”
Why or “Understanding”
How or “Process”
Constant need for data breadth and depth
Different “slices” of the company
… still high value, still high debt
Standards
Data automation & infrastructure
Wide vs deep focus
2
39. Pedaltothemetal
What is happening? What will help?
What or “Awareness”
Why or “Understanding”
How or “Process”
Constant need for data breadth and depth
Different “slices” of the company
… still high value, still high debt
Silos!
Standards
Data automation & infrastructure
Wide vs deep focus
2
44. The dashboards are
owned by different
teams
?
Why doesn’t my
revenue match with
what CFO presented?
Payingitdown3
ASKER ANALYST
45. #Facepalm
Let’s fix this
The dashboards are
owned by different
teams
?
Payingitdown3
ASKER ANALYST ENGINEER
Why doesn’t my
revenue match with
what CFO presented?
46. Payingitdown
What is happening?
What or “Awareness”
Why or “Understanding”
How or “Process”
How well or “Efficiency”
Constant need for data coverage
“Slices” give way to the full pie
3
47. Payingitdown
What is happening? What will help?
What or “Awareness”
Why or “Understanding”
How or “Process”
How well or “Efficiency”
Constant need for data coverage
“Slices” give way to the full pie
Metric governance
Golden data democracy
Self service and up-skilling
3
48. Payingitdown
What is happening? What will help?
What or “Awareness”
Why or “Understanding”
How or “Process”
How well or “Efficiency”
Constant need for data coverage
“Slices” give way to the full pie
Entropy!
Metric governance
Golden data democracy
Self service and up-skilling
3
53. Payingitdown
Week over Week (WoW Change Breakdown)
Component 1 Component 2 Component 3 Metric 1 Component 4 Metric 2 Component 5 KPI
p
% of Total Trending
High
Time
Low
Historical Trending
High
Time
Low
3