A major membership organization approached Market Strategies to undertake a multi-study customer satisfaction program comprising 200,000 verbatim responses. Processing these comments was an unusually complex task with a taxonomy that had evolved into thousands of categorizations to capture every possible aspect of customer experience. The categorization process demanded a high level of skill, and, compared with other projects, took three times as long. The method was too restrictive because the client needed to interrogate the data interactively and use it to respond to urgent questions raised in its business. The goal was to find a tool that the client’s analysts could use, while reducing the effort needed to prepare the data.
Market Strategies used three text analytic technologies: Natural Language Processing (NLP), machine learning and semi-automated coding. We deliberately adopted a hybrid approach to automate as much of the work as possible. We developed codeframes or taxonomies using NLP and then applied them to the data in a machine learning phase, which classified each new batch of data automatically. Then human coders used semi-automated coding to look for exceptions. Market Strategies developed a master taxonomy that applies to all four of the client’s tracking studies, and the client’s analysts can log into a web-based portal to apply the taxonomy to their own unstructured data.
The client is now coding more open data and analyzing more text than ever before. In combining the three analytical methods, Market Strategies brought the voice of the customer into the heart of the client’s analysis of both market research and enterprise feedback management data.
- We reduced labor by 95%--from 2,900 hours to 150 hours.
- We increased productivity by a factor of 21 (just 5% of the effort previously required to achieve an equivalent result in preparing the data for analysis).
- We enhanced quality control.
- We created a new service for the client allowing it to interact directly with its customers’ feedback.