The document is a summary of Wyndham's divisions and brands from a 2014 budget review. It states that Wyndham is the world's largest hotel company based on number of hotels, with approximately 7,500 hotels and 646,900 rooms. It is also the world's largest lodging loyalty program, vacation ownership developer and marketer, vacation exchange network, and professionally managed vacation rentals business. Wyndham has over 900,000 owners of vacation ownership interests, 3.7 million exchange members, and sends approximately 4 million consumers on vacation through vacation rentals annually.
2. 2014 Budget Review
Wyndham divisions and family of brands
World’s largest hotel company, based on
number of hotels
World’s largest lodging loyalty program,
based on participating hotels
Approximately 7,500 hotels and 646,900
rooms
More than 121 million room-nights sold in
2013
More than 9% of U.S. hotel room supply
World’s largest vacation ownership developer
and marketer
Approximately 185 vacation ownership
resorts with approx. 23,000 units throughout
North America, the Caribbean and South
Pacific
More than 900,000 owners of vacation
ownership interests
World’s largest vacation exchange network
World’s largest professionally managed
vacation rentals business
Approximately 107,000 properties in nearly
100 countries
More than 3.7 million exchange members
Send approximately 4 million consumers on
vacation through vacation rentals
jonathan@isernhagen.us @jon_isernhagen
5. 2014 Budget Review
Session agenda
Using data to target prospects and retarget customers:
1) Analyzing data to re-target new customers
1) search data
2) customer data
2) How CRM databases can be used to improve site
retargeting
jonathan@isernhagen.us @jon_isernhagen
6. 2014 Budget Review
Discussion agenda
1) Marketing 101
2) Targeting tasks
3) Extracting insights from data
a) Data assembly
b) Data mining
4) Targeting and personalization examples
a) Email
b) Display retargeting
c) Site
jonathan@isernhagen.us @jon_isernhagen
10. 2014 Budget Review
Discussion agenda
1) Marketing 101
2) Targeting tasks
3) Extracting insights from data
a) Data assembly
b) Data mining
4) Targeting and personalization examples
a) Email
b) Display retargeting
c) Site
jonathan.isernhagen@wyn.com @jon_isernhagen
12. 2014 Budget Review
Customer Experience Maturity Model stages
Source: “Connect: How to Use Data and Experience Marketing to Create Lifetime Consumers”
Stage Description
Initiate Establish an initial web presence. Push brochure content online. Spam everyone.
Radiate Reach customers through appropriate channels. Visitor/conversion focus. Start
making content consumer-relevant. Use personas.
Align Measure impact of marketing efforts (attribution). Articulate how marketing
supports strategic goals. Rate campaigns. Communicate across departments.
Optimize Personalize website experience using all signals and data available and as much
analytical horsepower as possible. A/B test and iterate.
Nurture Develop single customer profile. Listen for intent signals in all communications
via all channels. Improve relationship through automated trigger-based dialog.
Engage Establish unified customer database to bridge between online and offline.
Generate advocacy among your customers.
Cement Unify all departments to create great customer experiences fostering lifetime
customer relationships. Optimize customer experience with real-time predictive
analytics.
jonathan@isernhagen.us @jon_isernhagen
13. 2014 Budget Review
Customer Experience Maturity Model actions
Source: “Connect: How to Use Data and Experience Marketing to Create Lifetime Consumers”
jonathan@isernhagen.us @jon_isernhagen
14. 2014 Budget Review
Category Data types
Digital fingerprint
When visitors arrive on website: marketing campaign, keywords,
referring domain, location, device type, IP address.
On-site behavior
Observed while on site: landing page, site areas, product/service
areas, internal search keywords, content type.
Situation Weather, season/holiday, trending topics, time of day.
History Transactions, email response, website behavior, call center contact
Demographics
Gender, age, status, job role, acquired from forms, data vendors
and/or social data miners.
Psychographics
Interests, activities, values, lifestyle collected from surveyors, social
networks, and onsite behavior. E.g. spontaneous vs. methodical.
Connections
Social activity, connections, network properties (e.g. influencer or
connector)
- 14 -
Model inputs
Source: “Connect: How to Use Data and Experience Marketing to Create Lifetime Consumers”
jonathan@isernhagen.us @jon_isernhagen
15. 2014 Budget Review
Discussion Agenda
1) Marketing 101
2) Targeting tasks
3) Extracting insights from data
a) Data assembly
b) Data mining
4) Targeting and personalization examples
a) Email
b) Display retargeting
c) Site
jonathan@isernhagen.us @jon_isernhagen
16. 2014 Budget Review
SQL: Visual QuickStart Guide = easy SQL onramp
• Simple, English-like
language
• Enables you to play with
the data and understand
its possibilities
e.g.
Select Name_first, Name_last
From tblCustomers
Where State = “AK”
jonathan@isernhagen.us @jon_isernhagen
17. 2014 Budget Review
Pulling profile data together: back office transactions
Customer/Visitor Records
• Customer #1, Mike Johnson, ...
• Customer #2, Amy Morris,…
• Customer #3, Frieda Zimmerman…
• .
• .
Transaction data:
• Customer #1: 3/18/12, Ramada Yonkers, $119.00
• Customer #1: 11/22/13, Best Western Inn Ramsey, $551.18
• Customer #1: 2/14/14, Days Inn Nanuet, $93.81
• Customer #2:
• .
Transaction summarized data:
• Customer #1: 209 days ago, 3 stays, $763.99 total spend
• Customer #2:
• .
jonathan@isernhagen.us @jon_isernhagen
18. 2014 Budget Review
Pulling profile data together: web site behavior
Customer/Visitor Records
• Customer #1, Mike Johnson, ...
• Customer #2, Amy Morris,…
• Customer #3, Frieda Zimmerman…
• .
• .
Transaction data:
• Customer #1: 3/18/12, Ramada Yonkers, $119.00
• Customer #1: 11/22/13, Best Western Inn Ramsey, $551.18
• Customer #1: 2/14/14, Days Inn Nanuet, $93.81
• Customer #2:
• .
Transaction summarized data:
• Customer #1: 209 days ago, 3 stays, $763.99 total spend
• Customer #2:
• .
Site visit data:
• Customer #1: 2/1/14 13:40:00 Days Inn Home Page
• Customer #1: 2/1/14 13:40:10 Days Inn Results Page
• Customer #1: 2/1/14 13:40:25 Days Inn Property Detail Page
• .
Site data:
• Customer #1: 225 days ago, 12 page viewed, 5 minutes on site
• Customer #2:
• .
jonathan@isernhagen.us @jon_isernhagen
19. 2014 Budget Review
Extracting web data from Google/Adobe Analytics
Google Analytics
BigQuery
Google Analytics
Premium
Your database
Live Stream
Adobe Analytics
Premium
Your database
Data feeds
Adobe Analytics
Your database
jonathan@isernhagen.us @jon_isernhagen
20. 2014 Budget Review
Pulling profile data together: email data
Customer/Visitor Records
• Customer #1, Mike Johnson, ...
• Customer #2, Amy Morris,…
• Customer #3, Frieda Zimmerman…
• .
• .
Transaction data:
• Customer #1: 3/18/12, Ramada Yonkers, $119.00
• Customer #1: 11/22/13, Best Western Inn Ramsey, $551.18
• Customer #1: 2/14/14, Days Inn Nanuet, $93.81
• Customer #2:
• .
Transaction summarized data:
• Customer #1: 209 days ago, 3 stays, $763.99 total spend
• Customer #2:
• .
Site visit data:
• Customer #1: 2/1/14 13:40:00 Days Inn Home Page
• Customer #1: 2/1/14 13:40:10 Days Inn Results Page
• Customer #1: 2/1/14 13:40:25 Days Inn Property Detail Page
• .
Site data:
• Customer #1: 225 days ago, 12 page viewed, 5 minutes on site
• Customer #2:
• .
Email records (Sends, bounces,
opens, clicks, bookings)
jonathan@isernhagen.us @jon_isernhagen
21. 2014 Budget Review
Pulling profile data together: vendor data
Customer/Visitor Records
• Customer #1, Mike Johnson, ...
• Customer #2, Amy Morris,…
• Customer #3, Frieda Zimmerman…
• .
• .
Transaction data:
• Customer #1: 3/18/12, Ramada Yonkers, $119.00
• Customer #1: 11/22/13, Best Western Inn Ramsey, $551.18
• Customer #1: 2/14/14, Days Inn Nanuet, $93.81
• Customer #2:
• .
Transaction summarized data:
• Customer #1: 209 days ago, 3 stays, $763.99 total spend
• Customer #2:
• .
Site visit data:
• Customer #1: 2/1/14 13:40:00 Days Inn Home Page
• Customer #1: 2/1/14 13:40:10 Days Inn Results Page
• Customer #1: 2/1/14 13:40:25 Days Inn Property Detail Page
• .
Site data:
• Customer #1: 225 days ago, 12 page viewed, 5 minutes on site
• Customer #2:
• .Vendor-provided
demographics/psychographics
• Customer #1, retired construction
foreman, $485K net worth, 3 children,
13 grandchildren, 2 Pomeranians….
Email records (Sends, bounces,
opens, clicks, bookings)
jonathan@isernhagen.us @jon_isernhagen
22. 2014 Budget Review
Demographic/Psychographic data appends
1) Age/Sex/Race/Marital status/# and age of kids/Life stage
2) House value/type/residency length
3) Income/net worth/affluence/financial stress
4) Consumer-saver type/Coupon user
5) Web consumer type/ISP domain
6) Category bucket/Portrait
7) Politics/Religion/Environmental concern/Veteran status
8) Auto Make/Type/Fuel
9) Hobbies/Interests/Fashion segment/Pets
10)Medical interests
jonathan@isernhagen.us @jon_isernhagen
23. 2014 Budget Review
Social data appends, DIY
Hands-on data mining
text, using (free) Python
• Introduces social sites
• Describes the sites’
uniqueness and
unique data
• Explains how to pull
and analyze
jonathan@isernhagen.us @jon_isernhagen
24. 2014 Budget Review
Social data collection DIY: Twitter
Teaches you how to:
• Discover trending topics
• Identify retweeters of a status
• Identify all followers of a Twitter user
• Analyze a user’s friends and followers
• Perform tweet frequency analyses
• Find the most popular tweets
• Search for individual tweets
• Harvest a user’s tweets
• Crawl a Friendship Graph
• Analyze a user’s favorite tweets.
jonathan@isernhagen.us @jon_isernhagen
25. 2014 Budget Review
Social data collection DIY: Facebook
Teaches you how to:
• Analyze social graph connections.
• Analyze Facebook pages
• Analyze things your company’s friends like
• Analyze mutual friendships
• Visualize directed graphs of mutual
relationships.
jonathan@isernhagen.us @jon_isernhagen
26. 2014 Budget Review
Social data collection challenges: record matching
http://mashable.com/2011/02/25/data-mining-social-marketing/
Methods include:
• Company participation: “On Facebook…businesses can
gain access to the profiles of anyone who clicks the
“Like” button on the company’s business site…”
• Mining + Algorithms: If a company has one or two key
pieces of information about its customers — e-mail
address is often the most important — that company
can accurately identify them on a social site and extract
a substantial amount of data
jonathan@isernhagen.us @jon_isernhagen
27. 2014 Budget Review
Discussion agenda
1) Marketing 101
2) Targeting tasks
3) Extracting insights from data
a) Data assembly
b) Data mining
4) Targeting and personalization examples
a) Email
b) Display retargeting
c) Site
jonathan@isernhagen.us @jon_isernhagen
28. 2014 Budget Review
Definitions: Data Mining
“The computational process of discovering patterns in large
data sets … the automatic or semi-automatic analysis of
large quantities of data to extract previously unknown
interesting patterns such as:
• groups of data records (cluster analysis), and;
• dependencies (association rule mining).
http://en.wikipedia.org/wiki/Data_mining
jonathan@isernhagen.us @jon_isernhagen
29. 2014 Budget Review
Data mining by Clustering: flower categorization
http://www.mathworks.com/help/stats/examples/cluster-analysis.html
Fisher’s
iris data
30. 2014 Budget Review
Category Data types
Digital fingerprint
When visitors arrive on website: marketing campaign, keywords,
referring domain, location, device type, IP address.
On-site behavior
Observed while on site: landing page, site areas, product/service
areas, internal search keywords, content type.
Situation Weather, trending topics, time of day.
History Transactions, email response, website behavior, call center contact
Demographics
Gender, age, status, job role, acquired from forms, data vendors
and/or social data miners.
Psychographics
Interests, activities, values, lifestyle collected from surveyors, social
networks, and onsite behavior. E.g. spontaneous vs. methodical.
Connections
Social activity, connections, network properties (e.g. influencer or
connector)
- 30 -
Model inputs
Source: “Connect: How to Use Data and Experience Marketing to Create Lifetime Consumers”
jonathan@isernhagen.us @jon_isernhagen
31. 2014 Budget Review
Data mining by Association Rules: politics v. beers
http://www.marketplace.org/topics/life/final-note/what-your-beer-says-about-your-politics
32. 2014 Budget Review
Data Science on the cheap: Coursera and R
jonathan@isernhagen.us @jon_isernhagen
33. 2014 Budget Review
Discussion agenda
1) Marketing 101
2) Targeting tasks
3) Extracting insights from data
a) Data assembly
b) Data mining
4) Targeting and personalization examples
a) Email
b) Display retargeting
c) Site
jonathan@isernhagen.us @jon_isernhagen
34. 2014 Budget Review
Advantages to segmenting/personalizing e-mail
1) Technically simple and cheap
1) No A/B test tool required
2) No architectural changes needed
2) Asynchronous: time to analyze results instead of
responding real-time
3) Email address is ready-made primary key for
combination with other data sources
Source: TheEmailGuide.com
jonathan@isernhagen.us @jon_isernhagen
35. 2014 Budget Review
Personalized email best practice: Slingshot
• Not highly subdivided
• Softened #Fname#
• Top-of-funnel offer (for
re-engagement
campaign)
• Sent only to people
who hadn’t already
downloaded this app.
Source:
http://blog.hubspot.com/blog/tabid/6307/bid/341
46/7-Excellent-Examples-of-Email-Personalization-
in-Action.aspx
jonathan@isernhagen.us @jon_isernhagen
36. 2014 Budget Review
Personalized email best practice: Dropbox
• Behaviorally triggered
• Provides education on
how best to use their
product.
• Increases “stickiness”
Source:
http://blog.hubspot.com/blog/tabid/
6307/bid/34146/7-Excellent-
Examples-of-Email-Personalization-in-
Action.aspx
jonathan@isernhagen.us @jon_isernhagen
37. 2014 Budget Review
Personalized email best practice: Twitter
• Association mining
• Favorite restaurants
and people of other
washsquaretavern
followers turn out to
be good
recommendations.
Source:
http://blog.hubspot.com/blog/tabid/
6307/bid/34146/7-Excellent-
Examples-of-Email-Personalization-
in-Action.aspx
jonathan@isernhagen.us @jon_isernhagen
38. 2014 Budget Review
Resources: The Retargeting Playbook
Articulates complete retargeting
strategy and set of tactics:
• Setting up your campaign
• Segmenting your customers
• Optimizing your ads
• Meeting specific objectives
• Optimizing for Social and
Mobile
• Adhering to privacy laws
jonathan@isernhagen.us @jon_isernhagen
39. 2014 Budget Review
The “re-” is important, or else it’s just “targeting”
“It was already blue before.”
jonathan@isernhagen.us @jon_isernhagen
40. 2014 Budget Review
Definitions: Site Retargeting is…
Someone arrives at your site
(often from search)…
…then leaves without buying
(or buying enough).
“The Retargeting Playbook,” Berke
jonathan@isernhagen.us @jon_isernhagen
41. 2014 Budget Review
Definitions: various “Retargetings”
Term Actually describes
Search
retargeting
Targeting display ads based on Google search terms
Email
retargeting
Sending e-mail to people who visit your site,
or
Using in-message retargeting pixel to dynamically
adjust e-mail
Social
retargeting
Targeting a consumer based on Facebook “like,”
or
Targeting site visitors on Facebook Exchange.
“The Retargeting Playbook,” Berke
jonathan@isernhagen.us @jon_isernhagen
42. 2014 Budget Review
Campaign Setup
1) Post your privacy policy on all data-collection pages.
2) Tag your website to start building lists
a) Each ad marketplace has a JavaScript tag for each page header
b) Verify that the tags are working
c) Accumulate at least 500 visitors before impressions start serving.
3) Create and upload ads
a) There are ten total ad types used in retargeting
b) Five major types (300x250, 160x600, 728x90, 100x72, 200x200)
4) Launch your campaign
5) Segment
“The Retargeting Playbook,” Berke
jonathan@isernhagen.us @jon_isernhagen
43. 2014 Budget Review
Segmentation: Basic
“The Retargeting Playbook,” Berke
Basic retargeting segmentation is driven by intent signals.
1) Funnel-based segmentation
a) Number of visits to the site
b) Time on site
c) # of pages viewed (and funnel depth)
d) Items added to cart
2) Possible segmentation scheme
a) All site visitors
b) Viewers of at least one product page
c) Shopping cart users
d) Purchasers
jonathan@isernhagen.us @jon_isernhagen
44. 2014 Budget Review
Facebook targeting parameters
1) Location (Country, State, City, Zip)
2) Age (13-65 or 65+)
3) Gender and relationship status
4) Precise interests (liked “The Biggest Loser”)
5) Broad categories (e.g. small biz owners, Hispanics)
6) Connections (target/exclude fans)
7) Friends of connections
8) Education level
9) Likes and Shares
http://socialfresh.com/facebook-ad-options/
jonathan@isernhagen.us @jon_isernhagen
45. 2014 Budget Review
Throttling
1) Use Conversion charts to determine:
a) Frequency cap: max impressions a user can see/day
b) Audience duration: how long to keep targeting? (30 days?)
2) Cadence modification: bidding less on successive
impressions
3) Segment prioritization: e.g. exclude purchasers
4) Inventory management: drop out of non-performing
spaces
“The Retargeting Playbook,” Berke
jonathan@isernhagen.us @jon_isernhagen
46. 2014 Budget Review
Personalization using SiteSpect A/B testing tool
Browser Web server /
Application server
Algorithm engine
Personalization engine /
A/B testing tool
Cookie
Cookie data
Page request
w/cookie data
Personalized
Page response
Request and
Cookie data
Recommended
Content
Recommendation
request
Recommendation
Response
jonathan@isernhagen.us @jon_isernhagen
47. 2014 Budget Review
Site personalization: Guardian Royal Baby toggle
jonathan@isernhagen.us @jon_isernhagen
50. 2014 Budget Review
Summary take-aways
1) Do the hard segmentation work, targeting will take care
of itself
a) Gather all available data
b) Slice creatively
2) Understand where you are on the Customer Experience
Maturity Model and next actions to level up.
3) Even if you use vendors and/or an agency to do all the
technical heavy lifting, learn what’s happening behind
the curtain.
a) Ask the dumb questions until you can explain the processes.
b) Have at least a vague idea of how hard it would be to DIY
jonathan@isernhagen.us @jon_isernhagen
51. 2014 Budget Review
Test length for statistical significance
Sample size = 2 * Z^2 * Conversion * (1 - Conversion)
(Conversion * Change)^2
• Change: ….the smaller the lift you want to detect
• Confidence: …the greater the confidence you want to have
• Conversion:…the closer the page’s conversion is to 50%
• Contamination: …the purer you want the results to be.
If you let experiments re-use each others’ traffic, you can get
more data faster.
You have to test longer…
jonathan.isernhagen@wyn.com @jon_isernhagen
Hinweis der Redaktion
Hi, my name is Jonathan Isernhagen and I provide analytical support to the eCommerce team of….
…the Wyndham Hotel Group, which has all the brands in the box on the left….
Promotion of which is the major focus of….
…the Wyndham Wyzard, not to be confused with….
Tormund Giantsbane, to whom he bears a passing resemblance.
Our discussion agenda focuses on the use of data to target prospects and retarget customers.
In the interest of clarity I’d like to start with a few definitions, then discuss
Which targeting tasks we need to accomplish
What data we’ll need for that purpose
How we’ll assemble and mine the data, then;
Some examples of online targeting.
From time to time as we cover this topic it may seem as though I’m equating Segmentation, Targeting and Positioning. This is not the case, but they are entertwined.
Segmentation involves subdividing the members of your market by every available characteristic and seeing which ones create useful differentiation.
Targeting is the act of choosing attractive segments, after which you;
Use the four P’s to position your product as persuasively and profitably as possible.
It all starts from segmentation.
Charles Kettering, the American inventor who gave us Freon and leaded gasoline, once said that a problem well stated is a problem half-solved.
To which I would add the corollary that a market well-segmented is a market half-targeted.
So which specific targeting tasks do we need to accomplish?
The best resource I’ve found on the intersection of segmentation and data is called “Connect, how to use data and experience marketing to create lifetime customers.”
It categorizes every online retailer into one of seven levels of Customer Experience Maturity, which they call:
Initiate
Radiate
Align
Optimize
Nurture
Engage and;
Lifetime Customer, which I replaced with “Cement” because there has to be a verb there.
and tells us how to level up.
http://www.amazon.com/Connect-Experience-Marketing-Lifetime-Customers-ebook/dp/B00MFPZ9YU/ref=sr_1_3?s=books&ie=UTF8&qid=1429869865&sr=1-3&keywords=Personalization
Initiate is where we all started back in 1993, translating printed brochures into HTML and spamming random offers out to everyone in our address list.
Radiate is when you’re marketing through all digital channels and reporting basic website metrics.
I don’t think the authors understand algorithmic attribution, but that’s what Align tries to address.
Optimize and Nurture are where we start using data to segment, target and personalize the target consumer’s experience.
This view—which is an eye chart—shows the build-up of activities we are doing or doing better, by stage.
By the “Optimize,” stage among many other things, we are:
Targeting email to relevant segments;
Doing rules-based personalization of site visitors’ experiences, and;
Retargeting some site visitors via display ad sites.
What kind of data do we need to drive these personalization activities?
The authors split data into two big buckets:
What we know about you without a customer profile: including:
the digital fingerprint from how you got onsite;
onsite behavior is observed during the visit, and;
everything happening in the wider world at this time, and;
Stuff you can learn if we know your identity, including:
The history of all your interactions with the company;
Hard facts about you like gender, age, status, and job role.
Soft facts about your attitudes and interests, and;
Your degree of social connectedness and place in the social networks’ order.
So how do we pull this together?
Structured Query Language, or “SQL,” or “Sequel” is a language for manipulating data in one or more different tables of a relational database.
Even though you have someone who’s doing this for you, it’s still good to understand so you can drive conversations about which variables you can and can’t use to target consumers.
One way to think about this exercise is to imaging a gigantic, rectangular Excel spreadsheet of the kind you would use to do a multivariate regression, but with hundreds of columns and possibly millions of rows. You’re trying to set yourself up to find useful correlations.
The easiest data to start with are transaction data;
At the very minimum, this should be summarized into recency, frequency and monetary value. If you have “low,” “medium,” and “high” buckets for each of these dimensions you’ll have 3 x 3 x 27 segments to work with right away.
As you get more sophisticated, you can include brands, products and categories of products.
Depending on your web monitoring tool you can often export your click data and then roll it up into sessions.
You can then bolt the summarized data to your customer records to give you a richer understanding of your consumers’ behavior.
People shop a lot more than they buy, and the content and timing of their searches are very revealing.
Your ability to access raw site click, visit and visitor data is a function of the tool you use:
If you’re using straight vanilla GA, you’re out of luck. Google owns the data and provides no means of exporting it at the record level.
If you’re using GA Premium, you can now export up to X million records once per day;
If you’re using Adobe Analytics or AA Premium you can set up a recurring FTP process for exporting your records;
If you’re using Adobe Analytics Premium specifically, you also have the option of using their live stream product …..
There are also vendors who can place their own pixels on your site and generate user site behavior data directly.
Another easy branch of low-hanging fruit is your email service provider
Responsys, ExactTarget and Strongmail are all capable of exporting tables that contain all of your sends, bounces, opens and clicks
Opens and clicks can be put over sends to give a measure of engagement either overall or by message type.
To round out your consumer picture, and give you more hooks to hang your model on, data vendors have an amazing variety of data for sale.
Two vendors we’re evaluating offer between 550 and 650 individual fields that they can to your consumer data records with an 85% or greater success rate.
I tried to summarize them into general categories but can barely do them justice here.
Infocommand and Mosaic (and probably all the others) are happy to provide you with a detailed catalogue that shows every field they offer.
If you send them your database, they can send it back to you with the data appended, ready for use.
Matthew A. Russell wrote the book on social data mining which is actually the instruction manual for a free data mining tool he wrote in the Python programming language. Each chapter:
Introduces a new social website (Twitter, Facebook, LinkedIn, Google+)
Explains what is unique about the site and the data it generates, then;
Walks you step-by-step through the process of gathering and analyzing the data in various ways that help your business.
For example: With Twitter:
all of the data are public, so there are no access constraints, but;
the volume can reach several hundred thousand tweets per minute, and
connections are asymmetrical, which makes some nodes much more powerful than others and “following” does not imply a bidirectional relationship, and;
the data are not available in convenient database fields, but rather in free text that needs to be analyzed.
Russell shows you how to: ….
With Facebook the challenge is access:
You can only access data of your friends/followers;
There are more different types of data, but more constraints on obtaining it;
Connections are symmetrical so they are more meaningful;
Russell teaches you how to….
There are two popular ways of getting these data and attaching them to customer profiles:
The slow, small, cheap and easy way is to get a corporate Facebook page and get access to your targets’ data by persuading them to follow you.
The fast, big, expensive and harder way is to purchase social data and match it to your customer profiles.
Once we’ve assembled the data, what do we do with it?
Wikipedia defines data mining as the computational process of discovering patterns in large data.
You can use algorithms to:
Bucket things together, and/or
Find associations among them.
For example:
These data show the sepal length and width of three species of iris flower, represented by red, green and black circles.
In this example of what’s called “Clustering,” there are three types of Iris and the algorithm tries to bucket the species according to their leaf dimensions.
This can be done by measuring the distance from center points in the middle of each mass, or it can be done by drawing boundary lines.
Once you’ve done this with a known set of specimens, you get a set of rules you can apply to new ones.
You do this when you have categories already in mind.
Going back to our list of model inputs: we could use clustering to try to bucket site visitors as high-value or low-value based on the data available when the visitor first hits the site, then pause display ads if we feel they risk distracting a high-value visitor from completing a purchase.
Another type of data mining is determining association rules.
This is when you dive into the data with no biases and let an algorithm tell you about associations it discovered.
This example charts left-wing vs. right-wing political bias vs. preferences for beer.
As marketers, we aren’t interested in causality. Drinking Shiner may make you conservative or the other way around.
These just happen to be characteristics which move together. We don’t have to know causality to find them useful.
If you’re interested in learning more about this stuff on the cheap, the online learning center Coursera offers excellent online courses in data mining and other data disciplines for free
In the interest of clarity I’d like to start with a few definitions, then discuss
Which targeting tasks we need to accomplish
What data we’ll need for that purpose
How we’ll assemble and mine the data, then;
Some examples of online targeting.
Email is the best channel to start with.
You don’t need an A/B test tool;
Unlike with site personalization, no architectural changes are needed.
Email address is a great unique identifier which easily combines with all kinds of other segmentation data.
I never actually receive great personalized email messages, so I had to look online for examples.
This message has a casual salutation;
Includes a top-of-funnel offer for campaign re-engagement, and;
Was sent only to people who hadn’t yet downloaded a certain app.
The sender pulled together purchase history and site navigation data to create a tightly-targeted list with a highly-relevant offer.
This offer was triggered by the behavior of installing Dropbox.
It drives no incremental revenue but will increase “stickiness” by pointing out a useful feature of which the user may be unaware.
This is a perfect real-world example of association rule modeling.
Twitter analyzes billions of “Follows” and sends a list of the accounts of probable greatest interest to the email recipient.
1) Retargeting is the next obvious channel to benefit from segmentation and targeting.
2) Adam Berke and two co-conspirators literally wrote “The Retargeting Playbook,” which instructs you how to set up your retargeting channel from scratch.
The first thing Berke does is explain what he means by “retargeting.”
The “re” is important, like with Deja Blue.
“Our water used to be blue, then it was some other color, and now it’s back to blue again. Enjoy.”
The retargeting Berke focuses on is sometimes called “site retargeting.”
Someone comes onto your site, into the crosshairs, then moves on without transacting.
Other versions of “retargeting” include:
Search “retargeting” which isn’t really retargeting at all, just targeting based on search terms.
Email retargeting, which may mean:
Sending e-mail to people who actually visited your site, or;
Changing the content in a sent message based on whether someone saw your ad elsewhere or how your ad units are converting.
Social retargeting is:
Incorrectly thought of as targeting someone based on social behavior.
Correctly thought of as retargeting a site visitor using Facebook Exchange
Post your retargeting data policy on your site on each page where you collect retargeting data.
Tag your site with the JavaScript pixel of the ad marketplace that you chose, and make sure that it’s started working
Create and upload at least one of each of the five major ad types
Launch your campaign.
To start with, you can address unique messages to:
Anyone who’s touched any part of your site
Those who have viewed at least one product
Those who got to your review and continue page, and;
Those who bought something.
Location always used: a minimum of one country. If you choose more than one, no other geography can be specified
Many advertisers vary creative by age group for maximum effectiveness.
Even with
Overtop of these segments you will throttle your spend:
Use frequency and time vs. conversion charts to figure out the point of diminishing returns with total count of impressions and how long after the visit to advertise.
If you find that successive impressions are decreasingly successful at converting, bid less on each successive impression.
Spend most freely on your top targeted segments
Always be testing different creatives, and;
If your ad marketplace exposes this information to you, drop out of spaces which consistently underperform for your brand.
The hardest thing to personalize may be your website.
When a visitor reaches a personalized site, the;
Browser sends a request to the server which includes digital fingerprint information and anything useful the cookies contain.
The web server passes the request and cookie data to the A/B test tool
The A/B test tool asks the algorithm engine which treatment to display.
The algorithm engine responds to the A/B testing tool with the preferred treatment
The A/B testing tool records this choice and sends the content to the application server/web server
The web server sends the personalized page back to the browser.
In 2009 Netflix offered, and paid, a $1M prize to the team which produced the best algorithm for predicting a given watcher’s star rating of a new movie based on his or her rating history.
One of the best personalization examples is also a cautionary tale.
Orbitz used data mining to discover that Mac users running the Safari browser tended to choose higher-priced accommodations, which should come as a surprise to no one who ever saw Justin Long’s “I am a Mac” commercial.
Orbitz re-sorted their hotel results screens appropriately and caught heck in the media.
I was at Travelocity at the time and laughing my head off, but in reality Orbitz did absolutely nothing wrong. They didn’t obscure any search results from either user group.
The formula that governs how long a test needs to run in order to get to statistical confidence looks scary but is actually very simple.
The mimimum number of site visitors you need is equal to:
two, times;
the square of the Z statistic that corresponds to the confidence you want, which is 1.96 for 95% confidence, times;
the current conversion rate for the tested page, which is the ratio of visitors to successful completions, times (1- the conversion rate);
the minimum amount of change you want to be able to detect. If you think the impact will be obvious, you can set this to 10% and have a much shorter test, but if the real impact is only 5%, this shorter test may miss it.
The number that results is the minimum number of visitors you need for the control and each test variant.
The conversion rate is a property of the site, but you can play around with the change and confidence variables (as well as the number of test variants) to see what impact they will have on the minimum test length.