Sifting Through Text to Find What Customers Want

Can text analytics find stale buns?

Apparently so, according to customer experience vendor InMoment. For an unnamed national restaurant chain, the Salt Lake City-based company analyzed unstructured text feedback from customers and surfaced frequent reports of stale buns in some locations. The data led the chain to a specific bakery vendor, which assumedly was taken to task.

In June, InMoment emerged as the rebranded name of Voice-of-the-Customer vendor Mindshare Technologies, enlarged by its acquisition last September of customer experience management provider Empathica.

Digesting Data

InMoment's main offering is its Experience Hub, a cloud-based customer experience optimization platform that inhales and then processes the many structured and unstructured data trails that customers leave behind – trails generated from browsing websites, talking to call centers, receiving SMS coupons, filling out online questionnaires, making comments on social networks, and the like. After digesting the data, the hub is designed to deliver actionable insights to managers about what customers really want.

"Depending on the type and location of the data," InMoment Chief Product Officer Kurt Williams told CMSWire, "we [either] collect it ourselves or build APIs [or] form partnerships."

He added that data is processed using "sophisticated text analytics and other technologies, as well as expert analysis to generate actionable insights." The platform's text analytics technologies are built on the same Natural Language Processing (NLP) engine used by IBM's famed Jeopardy winner, the super-computer Watson. IBM has a major effort underway to offer Watson as remotely-available, distributed super-intelligence in a wide variety of fields.

Williams described the IBM service as "a sophisticated linguistic NLP engine that effectively allows us to teach the computer to read." He added that "other approaches are more statistical in nature and are inflexible, require vast sets of training data, and cannot model the subtle nuances of human language as effectively." Additionally, he said, non-developers such as subject matter experts are able to create the linguistic rules.

2014-1-July-InMoment Channels-Images.jpg

From the InMoment website

Linguistics vs. Machine Language

One question about Watson's distributed intelligence is whether subscribers to IBM's services will simply become lookalike vendors of Big Blue's computer intelligence..

"Right now," Williams told us, "most customers think text analytics is a commodity and it's not." He noted that different approaches yield different results. "Linguistic approaches are much better for high quality analysis of short comments, while statistical/machine language might be better for vast quantities of text."

InMoment, Williams said, tunes the Watson engine to its needs – customer experience optimization in the verticals of retail, food services, financial services and other areas, plus "an additional level of calibration to each client."

"We are able to capture the knowledge and build our collective intelligence into the product to the point that even another Watson provider would not be able to match our results," he said.

'Less Salt on the Fries'

Williams noted that the kind of numerical scores establishments get from customers can be "helpful," but, without textual analysis to locate specific issues, "managers are left to guess at how to address problems, and which behaviors to reinforce or reward."

2014-1-July-InMoment Comment Poster-Images.jpg

A screen from InMoment's Comment Poster

It's the difference, he indicated, between knowing that an establishment got a score of four from a customer in food quality, and knowing that the same customer wrote there "could have been a little less salt on the fries."

In the case of Chinese fast food vendor Panda Express, Williams said InMoment's text analytics allowed the chain to "support their hunches with hard evidence." Text analytics, for instance, reinforced the importance to Panda customers of "'small' things like smiling when greeting guests, food texture [and] fortune cookies." 

In addition to textual analysis of customer feedback, InMoment customer sentiment detection offers several widgets, such as a Comment Poster that brings forward what Williams calls "the most actionable and interesting comments."

The days of detecting what your customers want by talking to a few of them or checking a handful of response cards are over. The customer's increasing power is now based not only what they do or buy, not on the basis of a few comments, but directly from everything they state about your company.

Title image by salajean /