Scribd
Upload
154 views

Customer Analytics - Course Notes

This course provides an introduction to customer analytics and focuses on segmentation, targeting, and predictive modeling techniques. The first part covers segmentation using clustering and dimensionality reduction. The second part explores descriptive and predictive models for customer purchase behavior, including the application of deep learning.

Uploaded by

Deepak Purohit
Copyright
© © All Rights Reserved
Available Formats
Download as PDF, TXT or read online on Scribd
154 views

Customer Analytics - Course Notes

This course provides an introduction to customer analytics and focuses on segmentation, targeting, and predictive modeling techniques. The first part covers segmentation using clustering and dimensionality reduction. The second part explores descriptive and predictive models for customer purchase behavior, including the application of deep learning.

Uploaded by

Deepak Purohit
Copyright
© © All Rights Reserved
Available Formats
Download as PDF, TXT or read online on Scribd
You are on page 1/ 19

Course Notes

You are here because you want to learn the fundamentals of Customer analytics. The course provides
insight into the fundamental marketing modeling theory: segmentation, targeting, positioning,
marketing mix, price elasticity, and neural networks.

Customer Analytics: The first part of the course focuses on how to perform customer segmentation,
using a hands-on approach. It involves the application of hierarchical and flat clustering techniques for
dividing customers into groups. It also features applying the Principal Components Analysis (PCA) to
reduce the dimensionality of the problem, as well as combining PCA and K-means for an even more
professional customer segmentation.

Purchase Analytics: The second part of the course explores both the descriptive and predictive
analysis of the purchase behaviour of customers, including models for purchase incidence, brand
choice, and purchase quantity. Not only that, but it also covers the application of state-of-the-art deep
learning techniques to make predictions using real-world data.
Segmentation
STP is a fundamental marketing
Targeting framework. It can be applied to all areas
of business and marketing activities.

Positioning

B2C The clients of our business are individuals rather than organizations.
The advantage of a B2C model in terms of data science is that we have much
(business-to-customer) more data points.
The process of dividing a population
of customers into groups that share
Segmentation similar characteristics.

Observations within the same group


would have comparable purchasing
behavior.

Observations within the same group


would respond similarly to different
marketing activities.
The process of evaluating potential
profits from each segment and
Targeting
Segmentation deciding which segments to focus on.

Selecting ways to promote your


products. You can target one
segment on TV and another online.

Examining customers’ perception.


(Involves psychology and usually
budget constraints).
What product characteristics do the
Positioning
Segmentation customers from a certain segment need?

Shows how a product should be


presented to the customers and through
what channel.

Marketing
Mix In fact, this process is so important, that it
has a framework of its own called:
Marketing Mix.
Develop the best product or service and offer it at the right price through the right channels.
Product
Product features;
Branding; Packaging

Price
Product cost;
Long term price changes

Promotion
26% Price reduction,
display and feature

Place
Distribution:
intensive, selective, exclusive
Retailers
Physical Online
Characteristics
Location Many locations, returning customers. One location, many customers.
Data -- Fewer customers at a particular store, ++ More data points and more diverse
due to the physical restriction. customer information.
Returns ++ In physical stores, customers can see -- Products are returned more often, as
the product itself. Returns are less likely. customers cannot see and test an item.
Purchase history Gather information through loyalty cards. Database with all past purchases of customers.
Brand choice -- Unavailable in physical stores. We ++ We may have data for all products that the
assume the customer has considered all customer has looked at and which competing
competitor brands. products a customer has considered.
Ratings and reviews -- Unavailable in physical stores. ++ Different items could be reviewed and rated
(significant features for predictive modeling).
Price elasticity measures how a variable of interest changes when the price changes.

Price Elasticity Own Price Elasticity


Price elasticity with respect to the same product

Y: economic variable of interest


P: price

Supply and Demand Cross Price Elasticity


*The cheaper the product Price elasticity with respect to another product
the higher the demand

Q: quantity
P: price
01 Choose the number of clusters

02 Specify cluster seeds

03 Calculate the centroid (geometrical center)

04 Repeat until the centroids stop changing


The goal of PCA is to find the best possible subspace which explain most of the variance.
Most commonly it is used to reduce the dimensionality (number of features) of a problem.

Dimensionality reduction

technique

2D Plane
3D Plane
01 Reduce Dimensionality with PCA

02 Perform K-means with PCA scores as features

03 Visualize and interpret clusters


Quantifies the change in probability of purchase of a product with a given change in its price

Own-price elasticity of purchase probability


Quantifies the change in probability of purchase of a product with a given change in a competitor brand’s price

Cross-price elasticity of brand choice


Quantifies the change in purchase quantity of a product with a given change in its price

Price elasticity of purchase quantity


*Closest to price elasticity of demand
This is a deep neural network (deep net) with 5 layers.
Hidden layer 1 Hidden layer 2 Hidden layer 3
Input layer

How to read this diagram:


Output layer

A layer

A unit (a neuron)

Arrows represent mathematical


transformations
Hidden layer 1 Hidden layer 2 Hidden layer 3
The width of a layer is the Input layer
number of units in that layer.

The width of the net is the


number of units of the Output layer
biggest layer.

The depth of the net is equal


to the number of layers or
the number of hidden layers.
The term has different Width
definitions. More often than
not, we are interested in the
number of hidden layers (as
there are always input and
output layers).

The width and the depth of


the net are called
hyperparameters. They are
values we manually chose
when creating the net.

Depth
Our neural network has 10 features; therefore the input layer size is 10.
The hidden layer size is a hyperparameter. We can adjust it during the
learning process. In the lectures we start from 50 nodes, but its size hidden_layer_size hidden_layer_size
could be any integer number.
The output layer contains the two possibilities for the targets (0 and
1), therefore it has a size of 2.

How to approach similar problems? . .


1. Preprocess the data . .
a. Balance the dataset . .
. .
b. Create train, validation and test sets .
. .
c. Save the data in a tensor-friendly format .
. .
2. Train the model .
. .
a. Outline the model (create or envision a diagram like the one
on the right) . .
b. Create the actual network and choose appropriate starting
hyperparameters
c. Optimize the model by fiddling with the hyperparameters
d. Test the model
3. Save the model and deploy it where needed

You might also like