3. APPROACH
MovieMagic offerings need to be analyzed with respect to usage
patterns, customer base and behavior, sales & product portfolio
and come up with strategies backed by customer understanding &
personalization.
DATA ARCHITECTURE MODEL DEVELOPMENT
DERIVING
STRATEGIES
- Databases are created
using various sources:
− Customer
− Media
− Usage
−Common linkages are
created using Primary
keys IDs
−New data definitions
are derived.eg.current
inventory, turn-around-
time etc.
Classification is done
based on Volume vs.
Value concept.
−Technique used:
− Cluster analysis
− RFM Technique
−Clusters/Groups
created based on
independent metrics
available for each of the
categories.
−Prediction models
developed for each
branch of fishbone.
Defining customers via
metrics in terms of
clusters created for each
branch.
–Personalization exercise
for each cluster.
–Strategy derived for each
segment:
– Financial
– Marketing
– Sales
– Supply Chain.
4. DATA ARCHITECTURE
Subscription ID Description
1 Online Yearly
2 Offline Yearly
3 Online Monthly
4 Offline Monthly
Custome
r ID
Name Address Email Joined
Date
*Custom
er Type
Subscrip
tion ID
Subscrip
tion
Date
Custome
r ID
Media
ID
Date of
Rent
Period Quantity Price
Media
ID
Media
Name
Type Categor
y ID
Inventor
y (Q)
Date
Time
Stamp
Rented
Quantit
y
[Movie/
Game
Name]
Movie/
Game
Cat. ID Descr. Class
1 Genre Movie
2 Starcast Movie
3 Released Movie
4 Language Movie
5 PC Game
6 PS Game
7 Genre Game
Customer Database
Rent Logbook DB
SubscriptionDB
Media Database
MediaCategoryDB
6. DETAILED PROCESS
Illustrated below is the detailed process of Clustering and RFM
RFM stands for Recency, Frequency & Monetary Analysis
Recency: When did the customer make their last purchase?
Frequency: How often does the customer make a purchase?
Monetary: How much money does the customer spend?
The following step by step process will be followed for the Modeling:
DATA:
It has the fields : (i) ID, (ii) Area, (iii) Country, (iv) Recency (Rcen),
(v) Frequency (Freq), (vi)Monetary (Money)
Only the 4 attributes, area, recency, frequency, monetary, and one class (output),loyalty, are used to
build the decision table.
STEP1:
Cluster customer value by K-means algorithm. This step the scaling of R–F–M attributes and yield
quantitative value of RFM attributes as input attributes, then cluster customer value by using K-means
algorithm. The detail process of this step is expressed into two sub-steps as follows: (..contd)
7. STEP1a:
Defining the scaling of R–F–M attributes.
This sub-step process is mainly divided into five parts introduced in the following:
(1) The R–F–M attributes are equal weight (i.e. 1:1:1).
(2) We define the scaling of three R–F–M attributes, which are 5, 4, 3, 2 and 1.
(3) Sort the data of three R–F–M attributes by descendant order.
(4) Partition the real data of R–F–M attributes respectively into 5 scaling in MovieMagic dataset
(5) Yield quantitative value of R–F–M attributes based on input attributes for each customer
Sample Table:
STEP1b:
Cluster customer value by K-means algorithm.
According to quantitative value of R–F–M attributes for each customer, partition data
into n clusters using K-means algorithm for clustering customer value. (contd..)
9. METHODOLOGIES:CLUSTERING/PREDICT
RentRent SubscriptionSubscription Non-SubscribersNon-SubscribersSubscriptionSubscription
One Time Repeaters Yearly/Monthly Yearly/Monthly One Time Repeaters
For each of the Fishbone branch>>Subset of data obtained>Clustering/RFM Technique Used>Model developed
Cluster Function
of:
-DatetimeStamp
-Geography
-Media Category
-Media Type
Cluster Function
Of:
-DatetimeStamp
-No. of Times
Rented
-Geography
-Media Category
-Media Type
Cluster Function of:
-DatetimeStamp
-Geography
-Media Category
-Media Type
-Subscription type
-No. of times
subscribed
-No. of times
discontinued
Cluster Function of:
-DatetimeStamp
-Geography
-Media Category
-Media Type
-Subscription type
-No. of times
subscribed
-No. of times
Discontinued
-Browsing history
Cluster Function
of:
-DatetimeStamp
-Geography
-Media Category
Media Type
-Browsing history
Cluster Function
of:
-DatetimeStamp
-Geography
-Media Category
-Media Type
-No. of times
bought
-Browsing history
10. IMPLEMENTATION
Sales Strategy
•USAGE PATTERN: Within
X days of release, Y% of
extra streaming over base
and thereafter Z% of extra
rent over base value after X
days.
•COMBO PACKS: Creating
**combos of DVDs to push
sales of Slow Mover DVDs
Financial Strategy
•DISCOUNTING:
Discounting to clusters of
users based on profit
generation, less penetration,
opportunity index. Like Hike
Prices when demand more in
streamline for a
period,pattern of usage.
•SUPPLIER
NEGOTIATION: Based on
usage pattern, demand
forecasting, days of payable
outstanding can be
negotiated with the suppliers
Marketing Strategy
•GEO SPECIFIC
ADVERTISING:
More advertising in less
penetrating areas with
respect to the usage index,
competitive scenarios.
•TARGET MARKETING:
based on usage pattern &
specific demands, Customer
Lifetime Value.
Supply Chain Strategy
•STOCKOUT/BACKLOG:
predict inventory to avoid
stockouts, Calculate Adjusted
Turn Around Time based on
Consumption Pattern for each
branch of Fishbone.
•RED FLAGS: Flagging Users
which are probable unsubscriber/
discontinuation of usage based
on patterns of subscription
packages they used.
•DISTRIBUTION NETWORK:
local network at more
demanding areas.
**DVD Types:1 movie/game pack, N in 1 pack ,Combo Packs , Star Packs, Vintage packs etc.
Personalization: Creating portfolio of users at Individual level and implementing above strategies.