If you’ve listened to recent webcasts I’ve led or have read some of my previous blogs, you’ve probably heard me talk about the misconceptions associated with text analytics – particularly that it’s our silver bullet, or our knight in shining armor, sent to rescue CX programs and guide us through our CX journey.
Of course, we all know that one data source and one analytical method aren’t enough to fully understand a customer base. We know that we must look across a wide variety of information to make any conclusions. However, for some reason, we tend to think of text analytics differently – it’s not treated as simply the data source that it is. It’s treated as something separate and mystical.
So how are we CX professionals supposed to fight this misunderstanding, both within our own companies or with our clients?
One idea that we’re tossing around here at Walker is that text analytics isn’t always an appropriate solution. It’s not meant for every qualitative data set, regardless of what certain marketing/sales folks would have you believe. Sometimes, we’re trying to simply manage a data set, rather than analyze it.
That differentiation might seem like semantics, and I concede that the approaches are similar. They both deal with turning unruly qualitative data sets into understandable themes or topics. They both reduce the need to read an entire data set to understand what’s being said.
Thus, the main difference lies in what the ultimate outcome of the text processing is, and how those results are used. For example, if you’re simply managing text, then you might be pleased with seeing what main ideas pop out of the data – that Products is talked about in 60% of the comments, and 45% of that set is negative; or that 90% of Service comments mention Timeliness. You’re simply quantifying the results to understand main themes.
To analyze the text, you might want to take the results a few steps further. Who is mentioning the negative Products comments? Are they more likely to talk about Product Features or Product Performance when speaking negatively? How does their feedback impact other behaviors? Are these results in-line with other metrics we’ve collected? Tet analytics asks us to fit the qualitative results into a broader CX approach or understanding; this is where human intervention and expertise come into play.
Again, I admit that these approaches are very closely related to each other, and we're still hashing out some of the details and explanations. However, I think identifying that there are different methods of handling automated text processing will help us to alleviate some of the misunderstandings we’ve encountered in our CX programs.