Getting from raw data to deploying data-driven solutions requires technology, data, and people. All of which exist. So why aren’t we seeing more truly data-driven companies: what's missing and why? During Strata Hadoop World Singapore 2015, Pauline Brown, Director of Marketing at Dataiku, explains how lack of collaboration is what is keeping companies from building and deploying data products effectively. Learn more about Dataiku and Data Science Studio: www.dataiku.com

1. The 3 Key Barriers Keeping Companies
from Acting Upon the Possibilities
That Big Data has to Offer

2. A little bit about me…
• Born & raised in Palo Alto, California
• BA in European History From Columbia University
• Masters in Marketing & Communication from Sciences
Po Paris
• Director of Marketing at Dataiku
• Currently living in Paris

3. Shift from
Uomo Universale
« A man can do all things if he will. »
-Leon Battista Alberti (1404-72)
Excel at all things:
• Intellect
• Mathematics
• Science
• Art
• Social
• Physical

4. Shift from
Uomo Universale To Expert

5. Required Assets:
• Hacker mindset
• Logic
• Statistics
• Polyglot Programmer
• Mathematics
• Algorithmics
• Engineering
• Databases
• Machine Learning
• Strong creativity
• Strategical thinker
• Business understanding
• Strong communication skills
• Project management
Data Science Superstar

7. Are we in some sort
of Renaissance Era
of Big Data?

8. If so, what’s next?

9. Investigation Part 1:
What is a Data Product?

10. Data Products

11. Data Products

12. Data Products

13. Data Products

14. Data Products

15. Data Products

16. Data Products

17. Data Products

18. Data Products

19. Data Products

20. Data Products

21. Data Products
=
Data + Technology + Data Scientist? + End User

22. Investigation Part 2:
What goes on behind the scenes?

23. Building a Data Product
User
Interface
Stream / Real-time
Query
Data
Preprocessing

24. Building a (Predictive) Data Product
Predicted Data
Historical Data
Machine Learning
Model
Preprocessing

25. Building a (Predictive) Data Product
Predicted Data
Historical Data
Machine Learning
Model
Preprocessing
Pre-processing & cleaning data alone can take
up to 80% of the time spent on a data project

26. Building a (Predictive) Data Product
Predicted Data
Historical Data
Machine Learning
Model
Preprocessing

27. Building a (Predictive) Data Product
Predicted Data
Historical Data
Machine Learning
Model
Preprocessing

28. Running a (Predictive) Data Product
Deployment
Real-time / Stream
Model
Preprocessing
Predicted Data

29. Investigation Part 3:
Who does what?

30. Customer Data!
Machine Data!
System Data!
Graph Data!
Structured Data!
Unstructured Data!
Transactional Data!
Catalogue Data!
Web Log Data!
RAW DATA

31. System Architect /
IT Team / Data Engineer
RAWDATA
Data Product
=
Business Incentive

32. Mathematics / statistics
Data Product
=
Business Incentive

33. Data Product
=
Business Incentive
Mathematics / statistics / Business

34. « Data Scientist »
Data Product
=
Business Incentive

35. « Data Scientist »?

36. Data Engineers
« Data Scientist »

37. What I’ve Learned

38. Fact 1: The Skill Sets Exist
Business Statistics Math
Data Engineering
Build
Maintain

39. Fact 1: The Skill Sets Exist (& your company probably already has them)
Business Analyst
Data Engineer
Build
Maintain
Mathematician / Statistician

40. Fact 2: The Technologies Exist
(and some are free!)

41. Fact 3: The Data Exists

42. Why is Production & Industrialisation of
(Predictive) Data Products Important?

43. Those who win are those who deliver
new data products continuously

44. Those who win are those
who deliver data products
Data products are supposed to deliver
business value… if you don’t deploy them,
where’s the long term value?

45. No Industrialisation =
Limited ROI
Those who win are those
who deliver data products
It’s like building your dream house but
never moving in.

46. No Industrialisation =
Limited ROI
Those who win are those
who deliver data products
It’s like building your dream house but
never moving in. Absurd!

47. So Why Aren’t More Companies Deploying
(Predictive) Data Products?

48. A Data Product must be
business focused (ROI) & mathematically accurate (RELIABLE)

49. 1° Business Analytic & Algorithmic Minds Are Different…

50. 1° Business Analytic & Algorithmic Minds Are Different…
The Business Analysts Brain
Patterns. Patterns. Patterns.

51. The Algorithmic Brain
Performance. Truth. Anomaly.
1° Business Analytic & Algorithmic Minds Are Different…

52. …But Your Data Product Needs Both
Patterns, patterns, patterns
Performance, Truth, Anomaly
BUSINESS
KNOWLEDGE
MATHEMATICAL
ACCURACY
1° Business Analytic & Algorithmic Minds Are Different…

53. MINDSET
• Project alignment from conception to execution – install team mindset with
common goal – even if the paths to get there are different
FRAMEWORK
• One common platform with enough flexibility for both mindsets to fully
exercise their individual skill and expertise on a common project
Resolving the Skill Gap

54. 2° So Many Technologies, Languages, and Needs

55. 2° So Many Technologies, Languages, and Needs
R
/
Python

56. 2° So Many Technologies, Languages, and Needs
Code-­‐free
R
/
Python

57. 2° So Many Technologies, Languages, and Needs
Code-­‐free
SQL
R
/
Python

58. 2° So Many Technologies, Languages, and Needs
Code-­‐free
R
/
Python
Hadoop
SQL

59. …Or As My Boss Calls It: Technoslavia
Florian Douetteau
Dataiku CEO

60. …Or As My Boss Calls It: Technoslavia
Florian Douetteau
Dataiku CEO

61. …Or As My Boss Calls It: Technoslavia
Florian Douetteau
Dataiku CEO

62. …Or As My Boss Calls It: Technoslavia
Florian Douetteau
Dataiku CEO

63. OPTION 1:
Enterprise dictatorship
Living in Harmony with Technoslavia

64. OPTION 2 (my personal favorite):
Accept a polyglot approach
Living in Harmony with Technoslavia

65. 3° Production and Industrialisation is Complex

66. Data reliability is hard to guarantee
Technological Complexity
3° Production and Industrialisation is Complex

67. Technological Complexity
The assumption that production will identically reproduce the analysis phase
is a hard promise to make
3° Production and Industrialisation is Complex

68. Technological Complexity
Monitoring a predictive model’s life cycle is a tedious and continuous task
3° Production and Industrialisation is Complex

69. Human & Organisational Complexity
BUILDING
MAINTAINING
3° Production and Industrialisation is Complex

70. Human & Organisational Complexity
BUILDING
MAINTAINING
Business Analyst
• patterns
3° Production and Industrialisation is Complex

71. Human & Organisational Complexity
BUILDING
MAINTAINING
Business Analyst / Algorithmic
• patterns
• performance
• truth
3° Production and Industrialisation is Complex

72. Business Analyst / Algorithmic
• patterns
• performance
• truth
Data engineers
• stability
• reliability
• cost of ownership
Human & Organisational Complexity
BUILDING
MAINTAINING
3° Production and Industrialisation is Complex

73. Business Analyst / Algorithmic
• patterns
• performance
• truth
Data engineers
• stability
• reliability
• cost of ownership
Human & Organisational Complexity
BUILDING
MAINTAINING
3° Production and Industrialisation is Complex

74. TIP #1: Invest in a platform where development and production are the same
DEVELOPMENT
TEST
PRODUCTION
Making Complexity Work for You

75. TIP #2: Invest in monitoring capabilities & strategies
Making Complexity Work for You

76. Data Engineers must have visibility and understanding of the key business metrics
TIP #3: Name your Data Engineer(s) Wisely & Define Responsibilities
Making Complexity Work for You

77. Data Engineers must know if (and when) a model is diverging
TIP #3: Name your Data Engineer(s) Wisely & Define Responsibilities
Making Complexity Work for You

78. Data Engineer must be responsible for quality of service
TIP #3: Name your Data Engineer(s) Wisely & Define Responsibilities
Making Complexity Work for You

79. Differentiate builders of new data products
from those that maintain them.
TIP #4: It’s Not a One Man Show
Making Complexity Work for You

80. What To Expect?

81. From the Renaissance of Big Data…
Where the Data Science Superstar is one person that excels at all skill sets…
…and where actual data products are rarely deployed and maintained

82. To the Enlightenment of (Big) Data
Where the Data Science Superstar is a team of complimentary skill sets…
… and where data products are designed, built, tested, and deployed by a
team of skilled individuals that each have a distinct role.
TEAM

83. SPOTLIGHT on the
Data Science Team Manager

84. The Rise of the
Data Science Team Manager
The Data Science Team Manager must understand the stakeholders’ needs,
translate them into a business need that can be answered with a data
product…

85. The Rise of the
Data Science Team Manager
The Data Science Team Manager must permit and enable collaboration
between business analysts, statisticians, & engineers…
Collaboratively
design, build, & deploy
Data Products

86. The Rise of the
Collaborative Data Science Team
…all the while maintaining the distinction between each individual role
and each individual skill set.
Business
Mathematics
Data
Engineering

87. The Secret to Building and Industrializing Data Products
is Collaboration.
Today, collaboration between different
skill sets, technologies, and data is finally possible.

88. Data Science Studio:
One Platform for Development and Industrialization

89. Thank You!
Pauline Brown
Dataiku, Director of Marketing
Pauline.brown@dataiku.com
@pauline8brown
www.dataiku.com