Tuesday, January 13, 2009

Customers Are Talking: Reading Between The Lines

One of the important insights in looking for meaningful stories in customer interactions is the following: you can't read a story by looking at metrics. That is to say, how long someone talked, what time of day it occurred, etc., has no relationship to the content itself. In my work, I listen to lots and lots of customer stories, and I have experienced this very thing. If you want to understand the story, you have to read, or listen to, the whole thing.

It's unfortunate that this is so, because the quickest way to absorb information is to read it in summary. It's also the easiest way for computers to process information. Computers are excellent at counting, measuring, etc., but terrible at reading and interpreting.

I hear you already: what about semantic analysis? Good: doable by computers. Bad: doesn't provide much insight. Here's an example: evaluate all customer service calls longer than 8 minutes and containing the word "unhappy." Let the computer pull out two sentences before and after that word. Won't that sort out all the unhappy customer calls and allow us to analyze a manageable data set? [If you think this is difficult to do, I can point you to a slew of vendors who are dying to talk to you about their products.]

The problem is, "unhappy" is context-dependent. The caller may be unhappy with the quality of her service. She may also be unhappy she forgot to pack her son's lunch that morning, Someone else may be unhappy for a completely unrelated event.

[As an experiment, I've been monitoring Tweets referring to the Blackberry Storm using the happy :) or unhappy :( emoticons--easy to do with Twitter Search. With more than 100 tweets examined, very few of the emoticons represented satisfaction or dissatisfaction with the device itself--they were related to wanting the device and not getting it, or hoping to get it, for example.]

In a recent discussion, a friend talked about word clouds as very useful summaries of social media data. I pointed out to him that the appearance of a word in a story doesn't create significance. Similarly, the absence of a word doesn't mean that word is insignificant. (What's unsaid may, in fact, be the most important words in the entire dialogue. Harold Pinter won a Nobel Prize for his mastery of this truism.)

In sum, at present, the intervention of a person close to the customer interaction at the time it occurs is the best way to determine if a communication is significant or not. If it's someone looking at it after the fact, that person will have to read the entire story, not a summary. I wish there were a shortcut, but there's not.

Are keyword searches or word clouds useless? No. If you are a cable company, searching for specific, unambiguous words like "DVR" in your customer communication is likely to be useful. Searching for context-dependent items like "unhappy" or "delighted" is not.

blog comments powered by Disqus