Call Center - Customer Analytics

Why Customer level Analytics in Call Center?

Great customer-centric brands recognize that when employees and systems know everything about every customer interaction over time, they can provide better service and support.

In this complex and challenging contact center world, Customer level analytics holds the key to working smarter and having real impact across channels on customer experience.

During our exploratory analysis, we have observed stubbornly CCO anchored customers, who prefer the CCO channel by looking at their call frequency in a year and the time difference between every call. More than 35,000 customers who called for more than 10 times in a year and more than once in less than 30 days were identified.

The objective of this study is to understand customers’ calling behavior, cluster them into similar categories and develop a framework that enables better customer management at the Call Center

The methodology developed is based on a customer centric framework where the objective is to move the call center inbound system from servicing calls to servicing customers. This is achieved using a three pronged strategy as described below
1. Link analysis to predict next customer call
2. Developing customer clusters with similar call reasons
3. Reverse Scoring framework (C2S2)

Link Analysis:
To predict the next call reason from a customer based on his previous call and proactively communicate about the next call even before the customer calls.

An exploratory data analysis reveals that customers calling for specific reason with high linkage to another reason, have a high likelihood to call again for the linked reason. A proactive communication to customers would prevent these linked calls in the future. A detailed analysis to bring out the possible linkages between call reasons is done to define actions that will reduce future calls with proactive customer communication.

Link analysis is used to discover associations between the various call reasons. It is the process of discovering and examining the connections between items in a complex unorganized data system. Association discovery rules are based on frequency counts of the number of times that item occur alone and in combination in the database. The rules are expressed as “if item A is part of an event, then item B is also part of the event X% of the time.” These rules are denoted as A->B.

For instance, there were 7000+ cases where customers who called for enquiring about their Credit Card Payment, called again in the next 7 weeks to enquire for Credit Card Installment. This shows the high linkage between these two call reasons along with sufficient instances in the last 12 months. So next time when a customer calls for card payment enquiry, he will be proactively communicated about card installment products through alternate channels like SMS or mail post the call informing about relevant credit card installment plans.

Developed Customer Clusters using Unsupervised Learning Techniques:

To classify customers into different clusters with common calling behavior.

Customers are first segmented into 3 categories based on their call volumes in a year into:
1. High engagement (8+ calls in a year)
2. Medium engagement (3 to 8 calls in a year)
3. Low engagement (1 to 3 calls in a year)

The next step is to cluster customers within each of these three engagement categories into clusters such that customers within a cluster have similar calling traits. This is done using the K-means clustering technique on SAS Enterprise Miner.

K-means algorithm is an iterative algorithm that tries to partition the dataset into K pre-defined distinct non-overlapping subgroups (clusters) where each data point belongs to only one group. It tries to make the intra-cluster data points as similar as possible while also keeping the clusters as different as possible.

Top 5 call reasons capturing 43 percent of the calls are mostly around credit card transaction & statement enquiry, card installments, online & mobile banking general queries, general enquiries & credit card block & replace requests

All customers within each engagement group are quite different from each other. For instance, Cluster H3 in figure3 has all customers covering 14% of the call volumes among high engagement group calls calling mostly for card installment products.

All these 25 clusters based on calling behavior were further defined by various customer attributes including behavior, revenue, product holding, financial segment, digital activity and persona including nationality and income.

Customer Calls & Service Score (C2S2):
To build a framework for scoring customers which can be used to prioritize customers to speak to Call Center Agents when they call on the IVR.

A multi attribute value model (MAVT) is used to compute a score for each customer. Multi−attribute value theory is a method that transforms and aggregates available data into useful information. It makes the ‘incomparable’ attributes comparable, prioritizes them by assigning weights and finally reduces the amount of information by aggregating the weighted standardized scores.

The process to be followed to carry out MAVT is listed below:
1. Definition of parameters: identify parameters which are to be compared with each other.
2. Assessment of scores for each alternative: assign values to each effect or indicator.
3. Standardization of the scores in order to make the criteria comparable with each other.
4. Weighting of criteria, in order to assign priorities to them.

A total score is calculated by multiplying the standardized scores with its appropriate weight, followed by summing the weighted scores of all criteria.

The parameters used in the model are as described below.
1. Call behavior
2. Revenue
3. Persona
4. Financial segment
5. Customer behavior
6. Digital score
7. Call reason

Weights are assigned for each of these parameters and based on the value of each variable customer get a score – Customer Calls & Service Score (C2S2).

The model is built in a reverse scoring mechanism where higher C2S2 score implies that customer would be assigned a lower priority on agent call transfers. This score would help agents prioritize calls in times of call outages or limitation of ground staff services.