TransOrg Analytics explains how you can find look-alikes in real-time using “Big-data” driven algorithms by analyzing your customers’ behavioral signatures to:
• Integrate near real-time predictions
• Optimize campaign targeting
• Run simultaneous campaigns
• Significantly boost conversion rates and marketing ROI
Contact us at clonizo@transorg.com to learn more
1. This document contains confidential information
“Find similar customers, Boost ROI”- Presented by
TransOrg Analytics
Top 50 Best Companies to work for 2016,
Silicon Review magazine
Predictive Analytics Company of the Year
2014 – CIO Review Magazine
Top 20 Company in India - TiE Lumis
Entrepreneurial Excellence Awards 2013
Top 50 Big Data Analytics Companies in
India 2013 – CIO Review Magazine
2. This document contains confidential information
Introduction To Transorg Analytics
• Niche analytics services and
products company
• Founded in 2009
• Global presence in US, UK,
Singapore, Middle East & India
• 60+ scientists from top schools*
Who are we?
• Provide Descriptive, Prescriptive
and Predictive analytics services
and products**
• Turn data into intelligence and
actionable insights
• Institutionalize results-oriented
data-centric culture
What do we do?
• Industry-centric solutions
• Expertise in analyzing
unstructured data
• Ability to work with open source
& big data technologies
• Partnerships with global
platforms
• Help organizations adopt user
centric solutions
How are we different?
• Top 50 Best Companies to work
for 2016, Silicon Review
magazine
• Predictive Analytics Company of
the Year 2014, CIO Review
magazine
• Top 20 Companies in India, TiE
Lumis Entrepreneurial Excellence
Awards 2013
• Top 50 Big Data Analytics
Companies in India 2013 – CIO
Review Magazine
Recognition
1
3. This document contains confidential information
Descriptive Analytics
Real Time dashboards, Root cause analysis
Customer Segmentation
RFM, Statistical Clustering
Predictive Modeling
Retention, Cross-sell, Response, Market Mix, Pricing, Risk, Fraud,
Collections
“Mobisights” – Interactive Mobile Dashboards
Drill down, Visualize, Annotate, Connect, Notifications
Learn more: http://y2u.be/8R5ox_XlqR4
“Clonizo” – Cloning Customers
Identification of similar customers based on behavioral attributes
Big Data Analytics
Real Time Analytics, Stream Computing, Unstructured Data
Web & Social Media Analytics
Click stream, Retargeting, Advertisement Effectiveness, Lead Gen,
Campaigns
Our analytics offerings range from Descriptive Analytics to Big Data Analytics 2
4. This document contains confidential information
What we will be covering today
• How the current campaign targeting system works
• What are the problems with current campaign targeting system
• How look alike modeling can solve this
• How look alike modelling is currently being used in multiple industries
• Walk through a success story where Transorg optimized campaign targeting for a telecom giant
3
5. This document contains confidential information
How current campaign targeting systems work
• Multiple campaigns are run simultaneously wherein target
base is selected based on either some business logic or
through campaign specific predictive algorithms
• Customer behavior, habits, past spending trends, etc. are not
considered
• Business Logic
• Customer
Behavior
• Habits
• Spending Trends
• Preferences
Target Base 1
Target Base 2
Target Base 3
Target Base 4
• Same customers end up getting bombarded with multiple
campaigns
• This causes increase in campaign cost, customer agitation and
decrease in campaign conversion rate
4
6. This document contains confidential information
Problem with current campaign targeting systems
• Targeting right customer for right product
• Increasing conversion rate of campaigns
• Decreasing cost of campaigns
• Reducing customer agitation
• Eliminating need of creating different predictive models for
different campaigns
5
7. This document contains confidential information
What are look-alike models? How can they solve this problem?
• Use of behavioral and look-alike machine learning algorithms to continuously learn and adapt to changing behavioral
signatures in near real-time to target prospective customers
• Forming high quality target pool and increase the number of customers
• With decrease in target base, the conversion ratio increases exponentially
• Ensures reduction in cost and effort to run campaigns
6
8. This document contains confidential information
Look-Alike models in practice
An online clothing retailer increased
their Click through Ratio (CTR) by 120%
and conversion rate by 350%
Digital Marketing
0.04
2
0.11
12
0
2
4
6
8
10
12
14
CTR Conversion Rate
Before After
Telecom
A telecom company increased their
campaign conversion rate by 46%
1.00%
1.46%
0%
1%
1%
2%
2%
Before After
Retail
A software retailer increases ROI by
112% using Look-Alike Modelling
$40.00
$80.00
$0
$20
$40
$60
$80
$100
Before After
7
9. This document contains confidential information
What does Clonizo do?
• Predicts the behavior of customers with matching characteristics and behavior timelines to customers having a high
probability of performing a specific action
• Polishes the model to get customers who are highly likely to respond to the campaign goal
• Continuously learns, modifies and adapts to changes in existing features or addition of new features to the data
Clonizo’s Artificial Intelligence engine works on 3 basic principles:
8
10. This document contains confidential information
How does Clonizo work?
Data Upload
Data is uploaded through
Clonizo UI and fed to Clonizo
AI engine
Prediction
Lookalikes of positive
customers in target base are
found by Clonizo AI engine
Scoring
Each lookalike is scored based
on probability to convert
9
11. This document contains confidential information
Predictions & scoring explained better
Target customers are scored on the basis of prediction & probability to respond to an offer. Such customers can be
targeted with communications, and marketing campaigns. Hence, focusing exclusively on highly qualified prospects.
Lookalike Cust ID Positive Cust ID Score
2131232 1231231 0.991
4321344 4563221 0.973
6443322 1231231 0.88
1324221 6537653 0.71
4321245 3457432 0.657
8664631 6543456 0.5
5357965 4563221 0.412
9864352 4575431 0.333
4567434 9847738 0.212
3545443 7643222 0.123
Clonizo segments lookalikes into deciles on the basis of their
probability score
Top decile contains customers with maximum potential to convert
Top 3-4 deciles can be targeted to reach high potential customers who
are most probable of performing the desired action that aligns with the
campaign goals
10
12. This document contains confidential information
End-user display
Using which user can
upload intended target
base, previous conversion
base along with their
respective features, select #
look-alikes to be found of
each converted customer
Structured data or CSVs
Intended target and conversion
base are stored on HDFS on the
Clonizo’s production server
consisting of peer networks
connecting high computing
machines
Data transformation &
Clonizo Engine
Data from Clonizo’s production
server is ingested into peer
connected Spark servers
Spark, which enables efficient
scalable machine learning, is used
for data processing, transformation
& cluster expansion algorithms for
finding look-alikes
Obtaining look-alike
results
Result from Clonizo Engine is
stored on production server as .csv
or .txt file
This file can be downloaded by user
using FTP or HTTP protocols which
can then be fed into campaign
execution systems
Clonizo’s production servers
are rented on Digital Ocean’s
cloud computing architecture
Clonizo Engine works on Spark’s
MLlib scalable machine learning
library and Spark Python API or
PySpark for finding look-alikes
What is Clonizo architecture? 11
14. This document contains confidential information
“App download” campaigns executed using Clonizo resulted in 46% increase in
CTRs from target Telecom subscriber base
Objective
• Find look-alikes of positives from entire target population to reduce the
base size and increase % clicks to target base.
• Developed and applied look-alike model for cluster expansion
• Model compared and scored each feature in positive base to respective
feature in target base and got a cumulative score in range 0-1, 1
representing maximum similarity
• Filtered out subscribers with score >= 0.75 from the target base
Approach
Impact
• Reduced the target base by 60-70%
• With a minimal reduction in # of clicks, overall %
clicks to target base increased by 46%
$80.00
$40.00
$0.00
$50.00
$100.00
Prospecting Campaign
Clonizo-optimized Campaign
2.6
0.9
0.00
1.00
2.00
3.00
Millions
Prospecting Campaign
Clonizo-optimized Campaign
1.00%
1.46%
0%
1%
2%
Prospecting Campaign
Clonizo-optimized Campaign
50%
decrease in cost
per acquisition
(CPA)
65%
decrease in target
base size
46%
Increase in clicks
to target base
13
15. This document contains confidential information
Contact
US
Sandhya Krishnamurthy
sandhya.krishnamurthy@transorg.com
M: +1 510 516 6443
India
Shuchita Jain
shuchita.jain@transorg.com
M: +91 98112 60911
Debjit Sen
Debjit.sen@transorg.com
M: +91 99532 46251
UK
Naresh Priyadarshi
naresh.priyadarshi@transorg.com
M: +44 740 481 6818
Singapore
Vijay Bajaj
vijay@transorg.com
M: +65 9752 9020
TransOrg Analytics
www.transorg.com
https://www.linkedin.com/company/transorg-solutions-&-services
https://twitter.com/TransOrg
https://www.facebook.com/transorganalytics
14