Next-best offer refers to the use of predictive analytics solutions to identify the products or services your customers are most likely to be interested in for their next purchase.
Facing this topic I have made a personal research, and realize a synthesis, which has helped me to clarify some ideas. This presentation does not intend to be exhaustive on the subject, but could perhaps bring you some useful insights.
2. Next Best Offer Batch Use case
Smart Outbound Personal Banker Calls example
Situation
Opportunity to analyze customer banking
activity to detect opportunities for personal
banker to cross- and up-sell.
Problem
Information in transactional systems needed
to be pulled together and analyzed.
Solution
All customer activity is loaded into the AEI
Warehouse. 300 business rule queries scan
the customer database every night to direct
significant customer events to trigger out the
best opportunities. Information is driven to
banker desktops for outbound calls.
www.decideo.fr/bruley
Impact
• Scan 2.7M daily
customer events
• 3M annual opportunities
• 500,000 relevant calls
• >40% response rate
3. Personalized Offers via The Call Center?
Personalized Offers
Customer
X
Cindy Bifano
Renewals: 07/02/09
Affinities: e-Nest3
Product links
Trigger
1168 Barroilhet Dr.
Savings
Hillsborough, CA, 94010
555-954-5929
Customer Value score: 87
Attrition score: 32
Accounts
708009838228
Email
Lending
LB@gmail.com
Household
Joint account
Summary
Date
Call Ctr
Inbound
03/02/07
Call Ctr
www.decideo.fr/bruley
Inbound
X
I see you made a large deposit
4/13/07. Do you have any plans
for this? Can I suggest a high yield
bond?
Did you know you are near your
overdraft limit? Would you like to
consolidate this into a term loan?
04/18/07
04/21/07
My Sales Targets & Scores
Offers Made
Target
75
Actual
63
Sales
$ Target
81%
X
Hand
offs
>
<
Personalized offers
X
Contact
Outbound !
Acct Age: 7
Last order: 01/15/07
Last offer: B707
!
Customer History
email
<
Customer View
>
21
4. WHAT IS A RECOMMENDATION
ENGINE?
Recommendation engines form a
specific type of information filtering
system technique that attempts to
present information items that are likely
of interest to the user.
www.decideo.fr/bruley
9. SHORT SCIENCE RECOMMENDATION
ALGORITHMS
Recommendation in general:
•Possible to use a wide palette of recommendation algorithms
•The best fitting algorithms are selected – after careful analysis of the data – to the given
recommendation problem and the corresponding optimization task
Overview of recommendation algorithms:
•Collaborative filtering (CF): Based on events generated in your service (Vod purchase,
Live channel watching event), finds similar behavior on users, and similarity on items
(VoD content, live schedule, etc.)
•Content based-filtering (CBF): Using only user/item metadata. Recommendations are
based on matching keywords.
Measuring Recommendation Quality:
•Average Relative Position (ARP): The distance between the prediction and the user’s
choice
•Top 10 Recall: the probability of hitting the chosen item from the top 10 items of the
personalized list
www.decideo.fr/bruley
10. Early generation recommendation
solutions…
… Did not offer really personalized recommendations for each and every user…
Not personalized
Only based on part of
the available information
Low customer retention
(if any)
www.decideo.fr/bruley
Minimal revenue
increase
Lower conversion rate
Increase of customer
satisfaction is
questionable
12. Teradata Solutions
Applications that utilize the data
and insight to address key
business functions
BUSINESS
APPLICATIONS
Integrated data
foundation
for competing
on analytics
www.decideo.fr/bruley
DATA
WAREHOUSING
BIG DATA
ANALYTICS
Technology and
solutions to drive
greater insights
from new forms of
data (exploding
volumes and
largely
untapped)
13. Next Best Offer: customer centric
marketing
•
•
Action can take multiple forms
- Purchase recommendation
- Pricing recommendation
- Advertising recommendation
- Promotion recommendation
- …
Recommendations can be based on multiple
factors
- Product affinity
- Pricing affinity
- Behavior affinity
- Lifecycle affinity
- Attribution analysis
- …
Ability to customize actions to get more favorable outcomes
www.decideo.fr/bruley
14. Understand Affinity between
Departments
Drive Sales by Cross-selling Products
Home & Garden,
Home & Garden,
Bedding and Bath &
Bedding and Bath &
Furniture have high
Furniture have high
affinity
affinity
Low Affinity
Low Affinity
between certain
between certain
departments
departments
www.decideo.fr/bruley
15. Overview of Cross-Basket Affinity
Challenge
•
Difficult to do in a relational DB due to
the sheer size of the combinatorial
permutations of the various purchasing
sequences.
Requires good customer recognition via a
credit card database or a customer loyalty
card program.
Cross-Channel Transactions
X Customers X Marketing Campaigns
Transactional DB
Customer Loyalty
With Teradata Aster
•
•
Use nPath/Sessionization to identify
“super” baskets within a time window.
Tighter time window implies higher
affinity.
Run Basket Generator to identify the
most frequent affinity items &
subcategories.
TransID
UserId
Date/Time
Item
UPC
874143
10001
11/12/24
83321
543422
20001
11/12/28
73910
632735
30002
11/12/24
39503
452834
10001
11/12/30
49019
•
Enables more accurate targeting of
customer needs; reduce direct marketing
spend, increase revenue yield.
www.decideo.fr/bruley
Address
Phone
10001
10 Main St
555-3421
20001
24 Elm st
232-5451
534 Rich
232-5465
Retail EDW
Product/Item Hierachy
Item UPC
Category
Dept
83321
Heels
Shoes-Womens
73910
Impact
UserId
30002
•
Handbags
Accessories
39503
Dresses
ApparelWomens
49019
Perfumes
Cosmetics
Marketing/Promotions
Date
CampaignID
UserId
11/12/24
3241
10001
11/12/28
2352
20001
11/12/24
3241
30002
11/12/30
2352
10001
16. Barnes & Noble: Using Aster SQLMapReduce
Dynamic Consumer Personalized Recommendations
How to increase relevancy of cross-category offers ?
Analyze Cross-Channel Consumer Data
• Both “known” members and non-Members
• Purchases and browsing behavior online, in-store, and mobile
• Rapidly change targeting strategies & models
Drive personalized recommendations across products
and categories through any in-bound or out-bound
delivery
•Co-purchase analysis and category affinity scoring
•Customer recommendations:186 million product pairs
•Keep scoring models updated across changes in both customer and
aggregate actions
•Ensure that model output is available to all consumer communication
channels: in-bound and out-bound
www.decideo.fr/bruley
17. Increased Conversions from
Personalized Recommendation Engine
Aster Data Business Impact and ROI
•
•
•
Increase conversions from recommendations; analyze patterns across eBook
(Nook) customers; 360 degree view of customer across in-store
and .com behavior
Build revenue attribution models to link every purchase to a site feature
Analytics Efficiencies:
- Payment processing and analytics; from 1 day to 1 minute processing with SQL-MR
- eBook analysis (downloads, reader preferences…); from 4-5 hours to 1-3 minutes
- Web log data processing: from 7 hours to 20 minutes
- Web Analytics data loading from Coremetrics: from 4 hours to 30 minutes including
geographical IP look-up
www.decideo.fr/bruley
18. Advanced Site Behavior and
Personalization
Personalization
How to increase purchase size with personalized recommendations?
Interpret individual user site visit behavior
•Customer example: Growing from 10TB to 20TB of
semi-structured clickstream data
•Capture behavior patterns in a site visit using Aster
Data Sessionization operator
•Determine who put what in their cart and if they
checked out
Deeper, personalized recommendations cross-product
and cross-category with graph analysis
•Improve recommendations beyond “people like you”
•Identifies relationships between pairs of product
types, association and direction of relationship
Behavioral pattern analysis for site optimization
•Discover order in which customers add/remove items
to/from carts
www.decideo.fr/bruley
19. Global Architecture Solution In Detail
…
1. Observed patterns pushed to Channel
2.
Inbound
Channel
Customer Interacts
with a Channel
Prioritized / Personalized
Content, Message, Offer
4. Returns offer
3. Begin
Processing
5. Continuous learning
and updated models
Dynamic
Profiling
360 degree view
Demographics
Transaction data
Contextual
No data
replication
www.decideo.fr/bruley
Multidimensional
Analytics
Business
Rules
Campaigns activation
and qualification
Offers governance
Offers history
Automatic real-time
targeting
Likelihood estimation
Response prediction
Message
Strategies
Aligns customer
interests and
organization objectives
Balances channel and
marketing
Using Aster Discovery Platform, you can identify when customers are transitioning from one department to other and use this insight to better understand the affinity between departments. In the visual, lines represent number of visits going from one node to another across unique sessions.
Home & Garden, Bedding and Bath & Furniture have high affinity as indicated by the thicker line connecting these departments.
Low Affinity between certain departments – e.g. customers are not moving from Crafts to Luggage or vice versa.
Such a affinity analysis can be used to:
Strategically place ads in one department to drive cross sell of products from another department with high affinity
As a retailer with a brick and mortar presence, you can also look into whether you should change the layout of the physical store. E.g. place Furniture and Home and Garden together.