Voice Analytics to Predict Customer Behavior

Project Description

Our Project:

Taking our recorded customer service phone calls and analyzing each one to determine sentiment to create a value that can be married to other attitudinal and behavior data points as a predictive measure of future behavior (persist as a Client, tenuous state, attrition etc.).

About TASC (Total Administrative Services Corporation):

TASC is a leader in the industry as the nation's largest privately-held TPA. TASC now offers more than 21 innovative products and services. Customers in all 50 states are served by the TASC team, which boasts 8,000 field representatives and over 900 associates at the Madison campus and remote locations. TASC works hard for tens of thousands of businesses, and last year the company's annual revenue exceeded $100 million.

Our Business Problem: 

For several years we have been collecting attitudinal survey information and have found that we are not able to accurately predict customer behavior using this data alone. We would like to bend the organization and the decision making of leadership to be data driven decisions not based on attitudinal alone but with more of an emphasis on behavioral data. To that end we are looking at whether or not the phone call data that we currently capture (every interaction) can be used to determine predictive behavior. For example, if a customer calls and is "upset" does that translate into a termination when x, y, and z is also present? Can we answer for the following:

We want to be able to know when the customer state of mind is tenuous before they even know

We want to be able to "read" people based on certain things before they even know how they feel"

Expertise Needed: Voice Analytics

Data Sources:

Each unique call is stored as a .WAV and has a unique ID with meta data including the assigned agent, duration, product line, customer ID etc. We also have customer demographics available in MySQL along with attitudinal data in .XLSX (separate files for each deployment but each have the customer ID attached to each record).

Deliverable: 

At minimum would be a database back with the call meta data and the 'sentiment rating' or another label of what analytics are able to be performed. Furthermore we could commission same entity to do the full analysis of the data set but primary objective is to get the quantifiable data from the voice records. (Would be an ongoing engagement on an X frequency - vs. a one and done type of project).