Cygron DataScope provides organizations with a powerful data visualization and data mining environment to drive their Customer Relationship Management (CRM) activities to success. With DataScope, organizations can analyse their customer data and quickly discover what their customers want and predict how they will behave.

Data Visualization: the key to successful CRM

Visual data exploration & customer profiling

Cygron DataScope's award winning interactive data analysis and visualization can transform customer relationship management. Easy-to-read 3D visualizations, visual queries and relationship charts allow patterns, trends and hidden relationships to be identified and customer profiles quickly built. Powerful data mining algorithms perform customer segmentation, identify new customers, model loyalty factors, estimate lifetime value, predict buying patterns and customer churn, and more.

Direct Mailing case study

In this case study, DataScope helps run a direct mailing campaign for a retail bank. The goal is to target valuable bank customers who are potential defectors. To support the mailing campaign DataScope is used to build predictive models that:

1. Predict potential defectors
2. Predict their reason for defection - this determines the content of the mailing.
3. Estimate customer value - this helps prioritise the mailings and obtain the optimal ROI

The data available for analysis is the result of a survey that asked one thousand bank customers to rate the bank service level and products and specify what factor would most cause them to defect to another bank. This data has been combined with personal and demographic information.

Customer profiling through visualization

DataScope's powerful interactive visualization makes customer profiling easy. Synchronised 2D and 3D plots reveal that customers citing poor bank rates as reason to defect have high monthly account balances but low transactions. An automatic relationship finder helps identify relevant correlations, for example, a link between monthly transactions and monthly account balance and account type.

Visual data exploration & customer profiling

Identifying the profile of customers likely to defect due to poor bank rates. Filtered and unfiltered 3D bar charts are compared side by side. The left bar chart displays demographics for all customers while the right bar chart shows only customers concerned with poor rates.

Identifying the profile of customers likely to defect due to poor bank rates. Filtered and unfiltered 3D bar charts are compared side by side. The left bar chart displays demographics for all customers while the right bar chart shows only customers concerned with poor rates.

To investigate the profile of defectors further we use a DataScope visual query. Identical 3D bar charts are created to display customer demographic data. A visual query is created to select only customers citing poor bank rates in one chart. By comparing the two charts we see the differentiating characteristics of customers concerned about poor rates. For example, the ratio of small town customers who are male is drastically different.

Predicting customers likely to defect

Prediction model rules

A wizard guides prediction model building. Different modeling techniques are provided to handle different situations, in this case a rule-based model is selected.

We use a visual query to quickly identify surveyed customers who rate both bank products and services badly - these form the examples of defectors needed by the rule-building algorithm. Examples of the generated prediction rules are shown below. A second model to predict the likely reason to defect is built in a similar manner.

Data Visualization improves model evaluation

2D confusion matrix showing model performance

Before using a predictive model it must be thoroughly validated. Validation is commonly undertaken by applying the model to test data and reporting the number of errors made. However, this cannot differentiate good and bad parts of the model. DataScope's interactive visualization provides more comprehensive model evaluation. Regular performance statistics can be displayed in an intuitive visual way - as a 2D or 3D confusion matrix for example.

In addition, visualisation can also show detailed model behaviour on specific customer segments. By plotting the model predictions onto color in a 3D scatter-plot for example, we see that most model errors are for low income customers.

Red points are predicted loyal customers and green points predicted defectors. The plot is filtered to display only loyal customers - hence green points also indicate model errors.

Red points are predicted loyal customers and green points predicted defectors. The plot is filtered to display only loyal customers - hence green points also indicate model errors.

Modelling customer value

The last stage in preparing the direct mailing list is to encode some of the bank's business rules- in this case their definition of a valuable customer. Using DataScope's multi-criteria decision tool we build a model of customer value. When applied to the whole database, the model assigns to each customer an aggregated score and ranking.

Assembling the final mailing list

The list of predicted defectors with their account details, likely reason for defection and estimated value to the bank can be exported to database, spreadsheet or html file. To generate the final mailing list we simply import the complete customer database into DataScope. DataScope automatically reapplies the prediction models to the new data and generates the final mailing list.