Having recently completed a home renovation, my wife and I have been doing some shopping through Joss & Main, an online retailer with outlet-like selections of home furnishings. Like many other businesses, Joss & Main sent me a survey to measure my post-purchase experience. Unlike other businesses though, Joss & Main’s survey was only two questions (paraphrased as follows):
1. How likely would you be to recommend Joss & Main to a friend/colleague? (10-point scale)
2. Please elaborate on why you chose your answer to the previous question. (Open text)
If I told you that Joss & Main could use those two questions to create some highly informative quantitative models, you might be suspicious. After all, the survey includes only one question with numerical data.
In fact, modern text analytic methods have become so advanced, I have no doubt that Joss & Main is converting its unstructured data (i.e. text) into structured data (i.e. numbers). Text analytics has become critical to truly understanding the voice of the customer, and figuring out which customers will buy more and which will defect.
Text analytic software, combined with instruction and “training” provided by researchers, combs through verbatim comments in search of common themes. It can categorize data into themes, even attaching a sentiment to each theme. Going back to the Joss & Main example, in my comments I mentioned how frustrating it is to me that I cannot call their customer service team (Joss & Main only accepts customer service queries through email). The software will see me mention “customer service”, and then attach a negative sentiment to it because of my use of “frustrating”. Similarly, if a different person states how helpful Joss & Main customer service was for her, that comment gets coded separately under “customer service” with positive sentiment.
Here is where it gets really good: the software is simultaneously tagging people across all different themes. The result becomes a database of categories, along with data on which customers flagged each category. One can now create models that identify the key drivers of advocacy (or Net Promoter Score) by seeing which categories have more predictive value. If Joss & Main is really savvy, they can overlay data about those customers sourced from Acxiom (or the like) and then create extremely robust predictive models based on lifestyle segment, demographics, and other fields. The result is a “big data” approach to customer experience measurement, all driven by a basic, minimally intrusive two question survey.
So, next time you think about sending out a 40-question survey to hear the voice of the customer, think about whether simply structuring your unstructured data will be more efficient. Joss & Main seems to have it figured out. Now, if only my wife could find more online coupons for them…
What are your thoughts on the use of text analytics for marketing purposes? Continue the conversation with blog author Steve Seiferheld on twitter!