Abacus market analytics offers a full range of marketing analytics solutions in the telecom domain encompassing acquisition, account management, retention and operations optimization.

Marketing CRM solutions

Case study 1: Churn Solution

Objective

1. To develop a churn likelihood score to predict the probability of customer likely to churn in next 3 months
2. Identify customers from a database of users who have entered the grace period
3. A comprehensive indicator of Customers lifetime Value to bring profit based decision framework

Methodology

1. Data selection – Try and include all variables including sequential ones using customer usage and complaints data.
2. Data transformation – Identify the best predictors.
3. Binning – Construct new attributes or tranform attributes by binning.
4. Sampling – Based on the usage and complaints data take an effective sample size to maintain predictive accuracy taking equal number of non-churners as well.
5. Use of modeling technique – Use decision tree algorithms including – Random forests, boosting trees and Support vector machine.
6. Judging the technical results of different techniques using the confusion matrix and lift charts.

Case Study 2: Segmentation

Objective

1. Identify Telecom operators customer’s through segmentation based on attitudinal data;
2. Project identified segments to the entire customer base of the Telecom operator and
3. Acquiring customer knowledge.
4. Targeted marketing through appropriate segmentation.

Methodology

1. Primary research conducted amongst XXX mobile users and intenders (based on prepaid and post paid universe);
2. User segments identified with the objective to project and map these segments to the entire customer base using the usage profiles;
3. Sample distribution – X % pre-paid, Y % post-paid customers
4. Identify the key drivers of all the segments and important variables for the subscriber database
5. Run cluster algorithms to match primary clusters
6. Taking these clusters, build a predictive model using complex algorithms
7. Train and test the model on the same sample of customers and then test the model on same customers data (but different time series subscriber data) and find the predictive accuracy of the model

Results

1. Result of the model were tested on test data set, correctly classifying X % of the data into right segments
2. Results deployed on different time series data for same customers, correctly classifying 70% of the customers into right segments
3. These study finding are being deployed on customer data base