This document provides an overview of a presentation by Jed C. Jones on big data and advertising. It discusses defining big data, exploring common myths and truths, and provides tips on how to use big data in advertising. It also presents a case study of a national diamond chain store that used big data to understand online prospect characteristics and increase sales of high-end jewelry. The key lessons from the presentation are that big data is more hype than reality, smaller amounts of data can still utilize big data thinking, the focus should be on strategies rather than technologies, there are two types of customer data, and four ways big data can improve advertising through monitoring campaigns, testing hypotheses, building models, and finding new patterns.
Decoding Patterns: Customer Churn Prediction Data Analysis Project
Big Data and Advertising
1. Cutting through the Hype, Getting to What Matters
PRESENTED BY:
Jed C. Jones,
Ph.D.
Co-Founder of
MindEcology
Big Data and Advertising:
Mindecology.com
2. Who Am I?
But first, a word about me:
Jed C. Jones, Ph.D.:
• Co-founder and chief data scientist at MindEcology, LLC, a data-
driven marketing firm based in Austin,TX
• Heavily engaged in database marketing and digital marketing
for 15+ years
• Worked as a marketer for 20+ years on 3+ continents
• Picked up a few fancy academic degrees along the way
• Fascinated by the place where data, customer needs and
business capabilities intersect
3. Business Challenge: A national diamond chain store was interested in
understanding which online prospect characteristics were the most
significant predictors of likelihood to purchase a high-end products
within one of their more lucrative product lines
Complication #1: The shopper and the ultimate purchaser were not
always the same person or household
Complication #2: The company had hundreds of data points about
shopper behavior on hand and was not sure where to start with the
analysis
(to be continued . . . )
Case Study
National Diamond Chain Store
4. Agenda
Together, we’ll:
• Define Big Data
• Explore some common Big Data myths
• Explore some Big Data truths
• Share some tips with you on how to use Big Data in
advertising
• Explore a brief Big Data case study
• Share some resources for getting started
• Have some Q&A time
5. Questions forYou:
How wouldYOU define Big Data?
What are some of its KEY characteristics?
What is the PROMISE of Big Data (real or imagined)?
What is it?
What is Big Data?
6. The 3Vs….
What is it?
Defining Big Data
Variety: Lots of different types of data being potentially used in a
combined way to solve the same problem or set of problems
Velocity: Data being created, updated and even made obsolete
very fast – often in real-time
Volume: Terabytes or petabytes
7. Key Big Data technologies:
What is it?
Defining Big Data
• Distributed storage and processing used to crunch lots and lots
of data very, very quickly
• Database management solutions to manage and access the
data
• Analytics platforms to draw conclusions and make decisions
10. The Hype
BIG
DATA.
A
promising,
yet
somehow
a
bit
imposing,
phrase.
Some
things
you
may
have
heard
about
it
include:
• There
will
be
40
zeCabytes
of
data
in
the
world
by
2020
(1
ze0abyte
is
a
billion
terabytes)
(Hortonworks,
2014)
• New
technologies
are
revoluEonizing
what
we
know
and
can
know
about
our
world
(and
our
customers)
• The
Internet
of
Things
(IoT)
is
coming,
which
means
we
will
have
access
to
even
more
types
of
customer
data
to
analyze
• You
beCer
get
on
board
or
you
will
be
losing
out
(right???)
11. 5 Things
Today We Will Explore 5 Things about
Big Data that Can HelpYou Do
Better Advertising
13. 4 Big Data Myths:
Myth #1: More data is always better
Myth #2: It’s a cure-all for your advertising and marketing
problems
Myth #3: The technologies are smart enough to run themselves
Myth #4: Big Data is always better than small (traditional) data
Big Data Myths and Truths
Lesson 01
14. Truths about Big Data:
Truth #1: Allows you to combine different types of data sets easily
Truth #2: Enables you to do find things you didn’t know were there
Truth #3: Allows small firms access to processing power that only
Fortune 50 companies had access to just a decade ago
Big Data Myths and Truths
Lesson 01
16. Don’t have access to true Big Data in your organization? That’s okay
because:
1. Even when dealing with much smaller volumes of data (like
megabytes instead of terabytes ), you can still leverage the capabilities
inherent in Big Data, and
2.You can buy it: leverage third-party providers’ Big Data-driven insights
by combining it with your in-house data
Lesson 02
You Don't Need Tons of Data to Use Big Data Thinking
18. Technologies will come and go.You will benefit from thinking like an advertising
strategist before spending time on the specific technologies. Start by asking:
1. At which point in the marketing and sales cycle have I identified a problem?
2. What data can I directly acquire from related to this problem?
3. What additional data could I add to my existing data?
4. Which technologies should I apply to my problem?
5. Which analytical techniques should I use?
Lesson 03
Focus on the Thinking, Not the Specific Technologies
19. Don’t have your own data science team? Hire one full-time – or outsource it.
Here are the advantages of each option:
Lesson 03
Focus on the Thinking, Not the Specific Technologies
Advantages of hiring in-house:
• Anytime access to your data
scientist
• Predictable ongoing costs
• Perfect for recurring type of
analyses
Advantages of outsourcing:
• You are not tied to any single set of tools or
analytical techniques
• Ideal for periodic or custom analyses
• No need to pay for a regular salaried
position or software
21. Directly-Acquired:
Examples:
• Revenue data from purchases
• Geographic data
• Website visits
• Products reviewed or ordered
• Survey data
Lesson 04
Two Kinds of Data: Directly-Acquired vs. Sourced
Sourced:
Examples:
• Demographic data
• Psychographic data
• Media consumption data
23. 1. Monitor Existing Campaigns: Thanks to data gathering software such as Web
analytic, POS systems, and coupon codes, acquiring performance data about
your campaigns is (usually) pretty easy.
The process:
a. Collect data directly or from media partners
b. View data points as they vary over time or compare against pre-set targets
c. Make better decisions on how to spend your budget
Benefits:
Improved efficiencies in marketing expenditures makes it easier for you to gain
client and/or C-level executive buy-in for future campaigns.
Lesson 05
4WaysYou Can Use Big Data to do Better Advertising
24. 1. Monitor Existing Campaigns
Example: Compare cost-per-click data for different Adwords campaigns
Lesson 05
4WaysYou Can Use Big Data to do Better Advertising
25. 2.Test Hypotheses: Test your assumptions about your campaigns to find out what
really works. For example, you can run an A|B split test whether your call to action
should be “Buy Now” or “Learn More.”
The process:
a. Form a hypothesis by altering a single variable and guessing at the
outcome (i.e., if I do X, then Y will happen)
b. Test that hypothesis – dozens or hundreds of times, if possible
c. Based on result, alter hypothesis and/or update your campaign
Benefits:
Helps guide creatives (web designers, graphic designers, and writers) to build
effective content. Helps marketing directors fine-tune strategies using facts.
Lesson 05
4WaysYou Can Use Big Data to do Better Advertising
26. 2.Test Hypotheses
Example: Send web traffic to both pages and determine which layout actually
converts best
Lesson 05
4WaysYou Can Use Big Data to do Better Advertising
27. 3. Build Models: Models can be used to predict the likelihood of future events
based upon known data about past events.You can isolate the relevant
characteristics of your best customers and then design ad campaigns that push
their “buy buttons.”
The process:
a. Isolate the salient facts about your best customers that sets them apart
from the rest of your market
b. Create advertising that speaks to the motivations and “buy buttons” of
your best prospects.
Benefits:
Provides the basis for generating creative campaigns that are focused on the right
type of prospects. Helps creatives, writers, and media buyers target more
effectively.
Lesson 05
4WaysYou Can Use Big Data to do Better Advertising
28. 3. Build Models
Example: Build customer personas on your best-converting, highest-revenue
customers and rank them by historical conversion rates
Lesson 05
4WaysYou Can Use Big Data to do Better Advertising
29. 4. Find New Patterns: Clustering – also called unsupervised learning – involves
applying machine learning techniques to large set of variables to find patterns,
without starting with a hypothesis.
The process:
a. Collect and arrange variables about customers, products, etc.
b. Run models (e.g., K-means) to cluster them together into natural groups
c. Form hypotheses about why the data clustered that way
Benefits:
Helps you spot patterns among customer or product groupings that you never
thought were there or could not adequately sketch out.You can then tailor your
marketing to each of those segments, separately.
Lesson 05
4WaysYou Can Use Big Data to do Better Advertising
30. 4. Find New Patterns
Example: Cluster customers in terms of historical purchases of products or groups
of products, then look for patterns as to what makes one group distinct from the
next.
Lesson 05
4WaysYou Can Use Big Data to do Better Advertising
32. Business Challenge: A.national diamond chain store was interested in
understanding which online prospect characteristics were the most
significant predictors of likelihood to purchase a high-end products
within one of their more lucrative product lines
Complication #1: The shopper and the ultimate purchaser were not
always the same person or household
Complication #2: The company had hundreds of data points about
shopper behavior on hand and was not sure where to start with the
analysis
Case Study
National Diamond Chain Store
33. Solution Architecture:
1. Data preparation: We pulled the raw data out of the data warehouse.We
engaged in “binning” - or grouping - of the data pertaining to each data
point into ranges, resulting in fewer variables required to be processed later.
2. Neural network machine learning: Arranged the data into input variables
and ran them through a machine learning algorithm to find the ideal
combination of independent variables that would predict sale vs. no sale.
3. Tactical marketing implementation: Once we knew which variables were
most predictive of a future sale, we implemented new marketing tactics that
helped the client leverage those insights.
Case Study
National Diamond Chain Store
34. Result: The process resulted in the isolation of 8 variables that, together, were
highly-significant indicators of likelihood to purchase a high-end piece of
jewelry.
To apply these findings to marketing tactics, we did the following:
1. On-site Behaviors: we developed special banner ads that could be shown
only to those individuals who were more likely to buy a high-end product; we
also simplified the path to the purchase page for all users
2. Characteristics: we built customer personas that could be hyper-targeted
via online and offline messaging
Case Study
National Diamond Chain Store
35. Business Outcomes:
• Increased sales revenue significantly* for this product line,
year-on-year.This was the result of a combination of
increased unit sales AND higher average transaction value.
* increase % protected due to client NDA
Case study
National Diamond Chain Store
36. Summary
5 Take-Aways from Today’s Session
1. Big Data is Mostly Hype, with Some Very Actionable Stuff, as Well
2. You Don’t Need to Have Tons of Data to Use Big Data Thinking
3. Focus on the Thinking, Not the Specific Technologies
4. Two Kinds of Customer Data: Directly-Acquired and Sourced
5. Four Ways You Can Use Big Data to do Better Advertising
42. Appendix
More applications of Big Data thinking to advertising:
1. Perform text analytics on social media mentions (Tweets, FB posts, etc.)
in order to calculate the % of positive vs. % of negative sentiments
2. Auto-classify inbound lead form submits into “strong” and “weak” lead
categories, respectively for more efficient salesperson time allocation
3. Perform a cross-sell analysis to calculate which products available on
your website you should bundle together as special promotions
4. Build a recommendation engine for your website, in the spirit of:“If you
liked reading ‘Adventures of Huckleberry Finn’, you also might like
‘Treasure Island.’”
5. Build an optimization model to determine the best marketing mix given
your available budget, media availability, and past conversion rates.
43. Appendix
More applications of Big Data thinking to advertising (continued):
6. Split test which billboard creative drives more in-store traffic by running
each creative type for one month at the same location.
7. Build a site selection model for a restaurant that caters to a nearby
business district.To understand passers-by, leverage 100s of gigabytes
of mobile phone geo-location data to extrapolate demographic and
psychographic info based upon where they go home to at night.
8. Build a color-coded heat map of where your best prospects live in the
highest concentrations within your trade area.
9. Build a model that predicts which shoppers are college-age males based
upon purchasing habits.