Data Mining with Sitecore xDB

CelebrateYour
Inner Geek with
Sitecore xDB
Presented by:
Amanda Shiga
Courtney Brownell

© 2001-2016 Sitecore Corporation A/S. All rights reserved. Sitecore® and Own the Experience® are registered trademarks of Sitecore Corporation A/S. All other brand and product names are the property of their respective owners.
#Sitecore SYM
3
Who is Rotman?

#Sitecore SYM
4
Who is Rotman
John Hull
“wrote the book on derivatives.”
Richard Florida
Best-selling author/Top50Thinker
Anita McGahan
Head of the Academy of Management

#Sitecore SYM
5
Who is Rotman?

6
A full service digital agency.
Specializing in the multi-billion dollar markets of content
management & digital marketing
TORONTO | OTTAWA | NEW YORK | LONDON | SAO PAULO | FLORIANOPOLIS
Over 200 Sitecore deployments
From basic sites to complex
enterprise engagements across a
range of vertical markets.
40+ trained, experienced
Sitecore developers
Ready to meet and exceed
your all Sitecore needs.
Full services teams
on three continents
With offices in Canada,
the United States, the UK,
and Brazil.

#Sitecore SYM
7
Nonlinear + Rotman: Since 2011

Rules-Based Personalization Example

Rotman – BehaviourVaries byVisit
10.11%
8.65%
6.94%
4.03%
New Visitors
Second time visitors
Third time visitors
Fourth+ visit
Home page visitors going to
fees and expenses - MBA - by
visit number
4.42%
5.47%
5.67%
6.54%
New Visitors
Third time visitors
Fourth+ visit
Application deadlines from MBA
Fulltime page
13.49%
11.09%
10.53%
7.16%
New Visitors
Third time visitors
Fourth+ visit
Admissions criteria from MBA
Fulltime page

Personalization byVisit Number
(V = 1)
Can I get
in and
how much
does it
cost?

Personalization byVisit Number (V=2)
Why
should I
choose
Rotman?

Rotman Geographic Personalization Results

#Sitecore SYM
14
• With Sitecore 8, we have access to rich, granular interaction data
• How can Sitecore Analytics and the xDB further inform our web
optimization strategy?
• Also – data science is cool.
• What can we explore and learn?
Exploring a new phase of website optimization

#Sitecore SYM
15
Quick refresher on xDB
• Collects all customer interactions
• Consists of a few different pieces
• Our focus: the raw data collection

#Sitecore SYM
16
Business Context – 2016
We had this notion that our audience was
compelled by ideas
Sitecore Analytics and xDB let us test it!

#Sitecore SYM
18
We wanted to answer these 3 questions:
Business context
We tend to look at our audience in binary terms
Finance vs Creative types, the Poets and Quants
— how is this characterization playing out?

#Sitecore SYM
19
Business context
We assume that thought leadership plays an important
role in our students deciding to attend Rotman
— is this true?

#Sitecore SYM
20
“What drew me to Rotman was its world-class faculty &
reputation as theTop MBA in Canada.”
-Lakshmi Ra, FTMBA ‘15
Business Context

#Sitecore SYM
21
Andrew’s finest moments were when he felt himself transformed by an idea.
“I wish each interaction at Rotman could inspire you
with a rush of new ideas.”
-Andrew Peters, Morning MBA Student ’10,Management Consultant, 15 years of work experience
Business Context

#Sitecore SYM
22
“You get set in your ways. Being here really challenged
me to do things differently.”
-Executive MBA focus group participant
Business Context

#Sitecore SYM
23
Business context
How do you take an academic research idea and
make it broadly relevant?

#Sitecore SYM
24
Organizational Need

#Sitecore SYM
25
Dean’sVision
• Research Excellence
• Globally Engaged Alumni
• Excellent Student Experience
Organizational Need

#Sitecore SYM
26
Big image slide with beaker >> Entering the data science lab…
Entering the data science lab

#Sitecore SYM
27
• The perfect ingredients for taking a small step
• An organizational culture open to experimentation and plenty of curiosity
• A base level of organizational analytics maturity
• Interesting hypotheses about audience behaviour and drivers
• Limited time, data, resources and processing power
• Fantastic data in Sitecore Analytics & xDB!
Shaping our experiment roadmap

#Sitecore SYM
28
• Lead scoring - understanding the factors that predict conversion
• Can this help us test our theories on poets/quants and thought leadership as a driving
force of engagement?
• Audience segmentation - clustering into groups showing similar behaviour
• Can this help us understand better how poets and quants behave, or whether there are
other behavioural clusters we may not be aware of?
• Data Mining for the Masses by Matthew North – applied to business scenarios
• RapidMiner Lead Scoring pilot – a 2-month pilot using Salesforce, Pardot
Opportunities for marketing teams

#Sitecore SYM
29
Our process

#Sitecore SYM
30
Courtney B.
MarComms /Web / Design
Sitecore Power User
YuanYuan H.
PhD, Applied Math
Data scientist
Amanda S.
Comp Sci / MarTech
Sitecore MVP
Our team
L to R

#Sitecore SYM
31
Our Data & Engagement Model
Interactions =Value
Tip:
Make sure your goals are
measureable by a pageview!

#Sitecore SYM
32
Content Profiling
Think about your audience.
Who are they? How do you segment them?

#Sitecore SYM
33
Content Profiling

#Sitecore SYM
34
Content Profiling
The pages we tagged were
given a value from 1-10 in each
of the segments.

#Sitecore SYM
35
Content Profiling Example

#Sitecore SYM
36
Content Profiling Example

#Sitecore SYM
37
Our tools and infrastructure - keeping costs low

#Sitecore SYM
38
Step 1:What can SitecoreAnalytics tell us?

#Sitecore SYM
39
SitecoreAnalytics – comparing engagement

#Sitecore SYM
40
Step 2:Where do we get the data?
• Experience Extractor vs. Reporting API vs. Mongo

#Sitecore SYM
41
• Time was spent translating the xDB into a dataset that could best
answer our business questions and was easily consumable by our models
Step 2: Making the xDB as consumable as possible

#Sitecore SYM
42
• The “fun art” of feature engineering: thoughtful creation of new attributes
based on domain knowledge and end business goal
• Extract as many helpful columns as possible from your raw data
• Keep in mind what’s accessible in Sitecore Rules Engine – you’ll want to
make use of your strong predictors!
Step 2: Making the xDB as consumable as possible

#Sitecore SYM
43
• 2 months of data = ~200,000 interaction records
• Approx. 60-90 minutes to run processing batches
• Ended up with 47 attributes (columns)
Step 2: Final ExampleSet

#Sitecore SYM
44
We ran binominal classification to predict two attributes:
1) hasConversion – indicating whether they converted on any of the 12
goals on the website
2) highValue – indicating whether the session achieved an engagement
value higher than 75 points
Step 3: Running a supervised ML classification

#Sitecore SYM
45
Supervised machine learning
The model learns… And then we test how good it is!

#Sitecore SYM
46
Supervised machine learning

#Sitecore SYM
47
1) Conversion is predicted by:
1) Visit to Research area of the website
2) Matched toThought Leader pattern
3) Matched to Full-Time MBA Prospect pattern
4) Visit to “IdeasYou Can Use” area of the website
* Model performance accuracy 88%
Step 4: Results and interpretation

#Sitecore SYM
48
1) High engagement is predicted by:
1) Matched to Full-Time MBA Prospect pattern
2) Matched toThought Leader pattern
3) Visit to Academic/Faculty area of website
4) Visit to Research Areas of website
* Model performance accuracy 85%
Step 4: Results and interpretation

#Sitecore SYM
49
Let’s inform marketing strategy for Homepage refresh initiative
1) Push the highly engaged Creatives towards a higher value conversion like Apply
Now
2) Lift the engagement of Quants by presenting very targeted financial thought
leadership
3) Bring forward academic research, “ideas you can use” and professor bios more
prominently on high traffic areas
The power of the results

#Sitecore SYM
50
Tie back to the Rules Engine

#Sitecore SYM
51
Second application: A predictor model for conversion that gets stronger
with more training
We can ultimately start a lead scoring feedback loop at the xDB Contact
level – likely to convert?
The power of the results
Retrain
model
with new
data
Auto
scoring of
customer
records
Update
Contact
with lead
score

#Sitecore SYM
52
1) More feature engineering to drive wider set of predictors
1) Pull out more behaviour data from pages visited
2) Add GeoIP data
2) Audience clustering
1) Deeper insight into finance vs. creativity theory
2) Refine understanding of thought leadership consumption
Step 5:What’s next? (Short term)

#Sitecore SYM
53
• Analyzing behaviour patterns over a years’ time
• Seasonal variations in the higher ed space
• Adding in Customer Data
• Looking at data governance and privacy
• Considering Student Data to append: events, enrollment, courses, profs, etc.
• Folding back into Marketing Operations
• For lead scoring and clustering, retraining models regularly and feeding back into
the xDB at the Contact level
• Making the information available in the Rules Engine and other marketing
platforms
Step 6: Long term possibilities

#Sitecore SYM
54
More data: CRM Connect / Data Exchange Framework

CRM
Social
Media
Customer
Support
Website
POS AppsxDB
IoT

#Sitecore SYM
56
• Incorporate data enrichment early in your IA and content strategy
• QuickWins for Richer Sitecore xDB
• If you’ve got a solid measurement program in place, mining the xDB for
repeatable insight is worth it (for your team skillset too)
• Be confident in your domain knowledge and the support of a data scientist;
find pockets of enthusiasm in your team
• Treat your first foray as a low-pressure pilot project, and allow for ~3-6 weeks
of data shaping and feature engineering
• If you’re extending xDB or xConnect, keep prediction and clustering in mind
for your data model
Key takeaways and lessons learned

Next steps/Resources
57
What to do
• Apply a quick data enrichment strategy
• Boldly enter the world of data science
• Give us a call! (And see back of room )
Resources available
• Slide deck / recorded presentation
• Data enrichment blog post
• Upcoming article series

#Sitecore SYM
58
http://youcaring.com/SitecoreSymGives

FOR DISCUSSION PURPOSES ONLY. Sitecore
Confidential and Proprietary. © 2016 Sitecore
Corporation A/S. All rights reserved. Sitecore®
and Own the Experience® are registered
trademarks of Sitecore Corporation A/S. All
other brand and product names are the
property of their respective owners.

Data Mining with Sitecore xDB

Recommended

Recommended

More Related Content

Similar to Data Mining with Sitecore xDB

Similar to Data Mining with Sitecore xDB (20)

Recently uploaded

Recently uploaded (20)

Data Mining with Sitecore xDB