International Bank, The Challenge
A worldwide leader financial institution, faced a crucial challenge for its private banking business and especially for a specific category of investment products and personal loans: How to increase sales, reduce any risk and reinforce brand image by communicating the quality of its private banking department.
The Solution :
Align Knowledge Strategy to Business Objectives
After analysis of processes, available data and information systems, we created a knowledge strategy in accordance with business objectives, macroeconomic trends, social environment, banks' human resources and previous experiences in marketing activities for the same categories of products or similar ones.
A strategy based on a progressive transition (customers loyalty-acquisition and information consolidation) from the most profitable and loyal clients, who in parallel offer complete and high quality information, to prospects. Strategy that is represented by the following diagram
In order to implement our strategy we had to define the notion-axe “time”. Having this axe we can identify a history, understand its evolution and predict the future behavior.
We defined the Personal-Family Economy on 3 axes. Income, Savings (positive or negative [loans]) and Consumption; But consumption (using a credit card) includes time factor.
So we had to understand credit card usage (consumption) per group of clients (clusters) and relate those findings with holistic relation between the clients and the bank.
Consumption of an individual is strongly associated with his socio-economic profile. As a result of this association, the groups deriving by the clustering procedure can also be considered as groups of people with a similar socio-economic profile.
The next step was to discover the correlation between these groups with the bank's products (investments and loans). The conclusion was that Credit Card owners buying behaviors', allow us to monitor the banking attitude.
In order to transform the above methodology into data mining applications, we adapted and implemented DATACTIF platform, especially for the needs of The Bank.
DATACTIF embodies a number of different algorithms from the neural networks and computational intelligence domain, some designed from our team and some from the state of the art of scientific research in data mining and knowledge discovery areas.
Contrary to the high level of complexity of DATACTIFs' algorithms, the friendly user interface allowed its use by decision makers without prior knowledge or experience of computer science and statistics.
Credit Card Owners Clustering
DATA. Analytical transactions of credit cards owners for a 6 months' period.
CLUSTERING. Clusters are groups of clients or other business objects that exhibit a certain degree of similarity in respect to a number of features that describe these objects (e.g. transactions of a client). The discovery and analysis of such clusters leads to a better understanding of the clients base and offers an add-on tool for use by the business executives.
The following pictures show the result of Credit Card Owners clustering and their consuming behaviors. The grey dots on the map are the created groups of people (clusters). The size of the grey dots is indicative of a cluster’s population.
The numbers on the dots are the cluster index (Cluster ID). The red color on the surface indicates similarity between neighboring clusters and the blue the opposite.
Credit card owners of cluster 14. How this cluster was formed
Hyper Clusters
We obtained 625 clusters with a detailed approach allowing an in depth analysis and 12 Hyper Clusters (Figure 5) with common characteristics, a number that allows the creation of efficient marketing strategy, taking into consideration particularities at the same time.
By associating Hyper Clusters with financial products (investment products, loan, mortgage, deposits, etc...) we had a global, direct and immediate evaluation of the existing clients. As we see in the example below (Figure 6) the association result between credit card owners and mortgage, defines one major target group and two secondary for further analysis and marketing actions.
(Figure 6: Association Result)
Prediction using Supervised Machine Learning
Prediction using Supervised Machine Learning
We used DATACTIF's supervised learning modules (SVM and Fuzzy System in this case) as they give the possibility to incorporate expert knowledge in a form of rules or in a form of examples.
Prediction of investment products future clients with DATACTIF SVM Prediction
DATACTIF SVM System was applied to the credit card owners' database and trained to predict new clients for investments product following the below strategy :
As a result we had a Prediction Accuracy : 41%
Prediction of Loans future clients with DATACTIF FuzzyPrediction
By Credit Owners buying behavior we predicted new clients for Personal Loans (PIL) following the below strategy :
As a result we had a Prediction Accuracy : 78%.
Prediction using Unsupervised Machine Learning
