2. 2
Patrick Deglon Bio
After a PhD in Particle Physics and ten years at the University of
Geneva studying the creation of the Universe, Patrick spent the next
decades driving business insights at eBay, Motorola Mobility, and Teradata.
At eBay, he led significant improvements in marketing effectiveness by
developing methods to measure incremental sales, and by running large scale
experiments on Internet marketing channels.
At Google’s Motorola Mobility, he raised the bar in Analytics and on-boarded
open Google tools and technologies.
In Dec 2016, he joined Teradata as the Vice President of Advanced Analytics
driving the strategy, direction, investment and realization of Teradata’s advanced
analytics portfolio, including the Teradata Database, Aster Analytics, and Open
Source Software.
He is married with two kids and moved to San Diego, California in Dec 2016.
3. The Role of Advanced Analytics
in the Modern Enterprise
21. 21
60%
Number of failed
Big Data projects
90%
Number of useless
data lakes by 2018
6 months
Static data projects
value duration
50%
of data in any project
is an exact repeat of
>5 other projects
80%
Of all projects is spent
preparing data rather
than creating value
5 months
Average to develop, test,
validate, deploy and scale
new analytical models
“Lakes are overwhelmed
with information assets
captured for uncertain
use cases”
“Not establishing
data governance and
management is
underlining value”
“We have institutionalized repetition
and redundancy
because of the way we
manage data”
“We lack discipline in
data management to generate
long term
value”
“Acting like a Fintech
is a lot easier said
than done”
“We keep buying promises,
and are not cynical enough about
the time it take to
realize them”
Endemic Challenges That Must Be Solved
29. 29
1st Example: How much is worth a human life?
1982: New chemical labeling in workplace (cost of labeling vs cost of life)
Occupational Safety and
Health Administration
Yes
Office of Management
and Budget
No
George H.W. Bush
Vice-President
?
30. 30
1st Example: How much is worth a human life?
https://www.nprillinois.org/post/how-value-life-statistically-speaking
Kip Viscusi
1982: New chemical labeling in workplace (cost of labeling vs cost of life)
Occupational Safety and
Health Administration
Yes
Office of Management
and Budget
No
George H.W. Bush
Vice-President
?
• US Worker risk of death: 1 in 25,000
• Dangerous jobs (arctic fishermen, oil rig workers, loggers) have higher risk
• By normalizing 200,000+ job profiles for education and skills, we can estimate that for $1,000 per
year more, worker are willing to take a extra 1 in 10,000 chance of dying on the job
• 10,000 workers = 1 estimated death
• so 10,000 * $1,000 = $10 millions (value of statistical life)
• Yet each life is priceless, especially for the love ones
31. 31
2nd Example: How much should you pay for the
keyword “red dress” on Google?
Google Shopping
(SKU-based, pay per impression, per click,
per sale, or for Return on Ad Spending)
Google Ads
(Keyword-based, pay per click)
Google Search
(Content-based, free)
32. 32
Experimental Design
Test Group
• Switch off Google AdWords
• 30% of USA
Control Group
• Keep Google AdWords
• 30% of USA
• Similar buying pattern/seasonality
than Test Group
US DMA – Designated Market Area
Google AdWords Locations Targetting
eBay Marketing
Experiment
36. 36 Don’t Do Marketing Do Marketing
No Purchase
Purchase
37. 37 Don’t Do Marketing Do Marketing
No Purchase
Purchase
L L
38. 38 Don’t Do Marketing Do Marketing
No Purchase
Purchase
L L
D D
39. 39 Don’t Do Marketing Do Marketing
No Purchase
Purchase
L L
D D
C
C
40. 40 Don’t Do Marketing Do Marketing
No Purchase
Purchase
L L
D D
C
C
?
?
41. 41 Don’t Do Marketing Do Marketing
No Purchase
Purchase
L L
D D
C
C
?
?
Cost
Direct Return
Incr Return
42. 42 Don’t Do Marketing Do Marketing
No Purchase
Purchase
L L
D D
C
C
?
?
Cost
Direct Return
Incr Return
Rule #1: Never, ever, spend money
unless you really-really have to
49. 49
Operational Simplicity
• Only SQL used
• One command to train the model
• One command to score
Verizon Results
GOAL RESULT
Avoid Data Movement / Duplication Met
Initial Accuracy of 64% or better Goal Exceeded: 69.8%
Model Training with >1M records to be <20 min Goal Exceeded: <13min
Model Scoring >200M records to be <30 min
(scoring the entire US customer base)
Goal Exceeded: 22.5min
“I’ve done this for a
long time. I really
haven’t seen this
result ever.”
- Ksenija Draskovic
Operational Results
In less than 40 minutes, they can refresh their
model and score their entire customer base,
with results live in their Teradata system
56. 56
Discover the Possibilities with the Teradata Vantage
Prediction
• How much revenues will we
have next month?
Segmentation
• Which prospects are the more
likely to purchase our product?
Understanding Causality
• Which customer events are
the most important to drive a
sale?
$
Text Mining
• Which offers include non-
compliant terms?
Networking Hypothesis testing
• Which customers are likely to
be fraudsters?
• Does our new website
generate significantly more
leads?
?
Re: Investment question
I can guarantee you a return on investment
of 10%, if you open a new saving account
with ACME Bank Inc. before the end of the
month.