How to use OLAP for interpreting Cluster Results

How to Intepret Clustering Results / K-means Cluster output using Data Warehouse

OLAP and Clustering ? Strange combination isn't it? OLAP - we know what to do, what fields to aggregate, slice and  dice, what report to show and what fields to show. Clustering on the other hand is looking for groups (with or without a particular goal ). How OLAP can then help Clustering. Please Read on. You might be convinced to try

K-means (or any other method) clustering partitions the data into groups that are homogeneous with respected to the cluster variables.

The output mostly would be the primary variable, and its associated cluster along with other cluster / segmentation variables

Cluster Profiling / Interpretation

Though cluster output and their scatter plot provides some interesting findings, it's the cluster profiling that helps the user in a better way

OLAP and Data Warehousing

OLAP  is an BI (Business Intelligence) approach to answering multi-dimensional analytical queries in real or near-real time

OLAP consists of Facts, Fact Tables, Dimension tables, Cubes that contains the precise data for a given requirement etc.

OLAP and Clustering

If the cluster output along with the interpretation variables and the cluster score (index) is fed to the OLAP and analyzed, it will provide a faster way of Cluster Interpretation and through many interesting insights

In similar way decision tree results can also be fed back to OLAP (+Oracle , +SQL Server , +Terradata AG )

How to Segment Social Media Users (Influencers / Detractors / Recommenders)

Segmentation of Social Media Influencers

We have seen in the past about the Retail Specific Segmentation Standards and also some Segmentation Tools that help to achieve that.

Here we would delve bit deep into segmenting the social media users.

Behavioral Segmentation of Social Media Users

Social media users are classified as shown by the Klout Influencer Matrix given below

Depending on the Need / Goal, one can select the specific segment of users. IBM's Social Analytics Tool (SPSS) helps to identify the following from Social Data

• Influencers
• Recommenders
• Detractors

The above combined with customer and transaction data would give necessary insights into customer behavior

Merging Social Data with CRM / Customer Data

The social users can also be segmented by demographics, geographics, which would provide information that might not be captured in CRM (for example, +Microsoft Dynamics ERP )

The 7 elements of social data

Sandile Mayambala a digital analyst as come up with the following combinations

• demographic
• product
• psychographic
• behavioral
• referrals
• location
• intention

It may not be possible to get all data points for every customer. In that case the data is normalized and analyzed

See also :How to Find Social Influencers using analytics,
Five Types of Social Media Influencers

Link-Selling Analytics (Web Analytics)

How does Analytics help efficient Link-Selling

Link-Selling is a form of cross-selling ,
which occurs when a product is selected. Then the shopper taken through a series of ‘yes/no’ questions regarding ‘additional extras’ that can be added to the cart.

The catch here is there would be series of questions put up and the the subsequent questions should be modified/updated according to the previous one.

Though mostly the logic is rule-based, use of Predictive modeling increases accuracy

Customer Analytics - Understand Mood of the Day

How to Analyze Customer Behavioral Changes on Time (Day)






 Netflix - the world's leading Internet television network with more than 33 million members in 40 countries has developed an algorithm (Cinematch recommender tool) that can predict the movies that a customer would watch based on

• Past behavior (view history)
• Product rating
• Recent comparison

Cinematch uses Nearest Neighbor method for predicting a movie

The company decided to have an Open contest to fine tune the recommendations (or the Next Best Action if you wanted it to be called so) - the winner would be the one who beat the recommendation of Cinematch by 10%.

AT&T with the combination of nearest neighbor , singular value decomposition (SVD) methods were able to win the one


Please have a look at http://www.research.att.com/articles/featured_stories/2010_01/2010_02_netflix_article.html?fbid=8rnpOtsYzRv for the full text.

Widely used Segmentation Products / Solutions

Free Segmentation Frameworks / Products

We had an indepth analysis of various segmentation frameworks for profiling customers in Segmentation Standards / Framework. Here we will look in more detail on some readily available Segmentation products that will accelerate segmentation/clustering.



Prizm's market segmentation comes-up with beautiful segments for a particular Zipcode. If you want to open a Store in the locality ZIP Code 65231, Auxvasse, MO. Here are the results from the search (http://www.claritas.com/MyBestSegments/Default.jsp?ID=20#)

PersonicX is a household-level segmentation system including 70 clusters and 21 life stages. The PersonicX suite helps to know and anticipate customers’ demographics and buying behaviors, conduct market analysis, plan customer acquisition strategies, and create cross-sell, up-sell and retention campaigns that are truly targeted, personalized and powerful.





Forrester's Social Technographics data classifies consumers into seven overlapping levels of social technology participation. Based on their proprietary Consumer Technographics survey data, they can
share with how social participation varies among your consumers globally and help plan a targeted social technology strategy.

+Monetate's LivePredict is an automated segment discovery product that automatically identifies valuable customer segments and the attributes that define them. This gives marketers the ability to
instantly take action by targeting offers and content specifically to those segments

+OfficeMax is using LivePredict to identify highest and lowest performing segments at both brand and campaign levels

Creating Valuable Data Columns - Aggregate Data

How to create insightful summary data from Operational Data

Analytics is not a magic wand. It produces useful insights when fed with good data, insightful information. The primary source of information is from Operational Data.

The operational data, however, is not quite useful as it is . It needs to be rolled-up to form a good amount of data points.

Some days back we were looking at the importance of Relative Pricing in Customer Analytics. Another real world example would be the Bizocity scores used by AT&T

Bizocity score takes into account the originating call number, the destination, the duration etc to arrive at the score.

See also: Case Study: Customer Analytics for Telecom Operator to Cross-Sell and Up-Sell

Customer Analytics - Relative Price as Data Point

Relative Pricing in Customer Behaviour Analytics

We had some discussion on Price Sensitivity earlier. This was used for determining the Right Price of offer for that customer. However, the price sensitivity is more related to Demand Elasticity

Relative Price is the price of a product/service with respect to another product/service. The latter one is usually termed as the base product/service.

How to use Relative Price to understand Customer Behaviour

Relative price provides very much useful insight about the customer's behaviour. The following would be the typical values for Relative price

• High
• Medium
• Low

The relative price is applied to each item based on a base item from that category. For example a segregating a customer based on the
different price preferences (Twinnings Tea might be a High valued one, while Tetley might be Medium priced; depending on the purchase behavior of each individual across category the following Relative Pricing Table is created)

Cust ID	Category	Premium	High	Medium	Low
RF0000123	OVGT120980	0	6	16	0
RF0000227	OVGT120980	11	3	2	0

Relative Pricing Customer Table

)Looking at the above table, one could distinguish between the customers - (for a particular category) - the second customer mostly shops for Premium items, while the former is a bit conservative.

The above methodology can be used and the following can be used as Segmentation variables

• No of High Value Items
• No of Medium Value items
• No of Low Value items

Please try this out and let us know the difference it brought to your segments

How to measure effectiveness of Switcher Campaigns in Retail 

Switcher campaigns are the one that disrupt brand loyalty and tease an otherwise loyal customer to switch over to the competitor. Tetley and Brooke Bond are competitors - what would be the impact of providing an offer on Brooke Bond tea for a Tetley fan.

dunnhumby's analysis on offers finds Switcher campaigns are always the worst performing. According to the reporthouseholds tend to buy more of their favorites, such as Coke products, or a new Coke product, if offered an incentive rather moving to Pepsi for example