When the "Voice of the Customer" is Actually the Voice of the Customer
The subject of big data is burning up the airwaves. What is it? How do you define it? Why is it different than business intelligence? What value does it bring?
These are all valid questions. But when it comes to big data, perhaps the most important question is, "How do I analyze all this stuff?" Because the nature and history of information management has created, as we all know, a ridiculous number of non-integrated systems (popularly called "silos") that nether recognize one another nor contribute to a single version of the truth, as they say. In other words, the variety of user interfaces and content repositories make it nearly impossible to benefit from the promise of big data integration.
And that's just text and data. In an effort to "simplify the complexity" of big data analytics, I sought out a person who I thought would help. And he did, but he also had the audacity to make it even more complicated. Tom Goodmanson is the president and CEO of Calabrio, Inc. His company wants to add yet another layer of complexity to the big data analytics challenge; specifically he is interested in the challenges of understanding voice. Yep. Recorded voice. Whew. It's getting messy out there.
"A lot of people think of big data as simply a large amount of bits and bytes being held together in rows and tables. But in fact the idea of big data has more to do with unstructured data—large amounts of unstructured data. The trick is to categorize that unstructured data.
"We have another spin on it," Tom continues. "For us the focus is often on voice. We start with a recording of a transaction with a customer. Then we make sense of it via analytics. But we also pull all the multiple channels into the information cycle, whether it be chat, email, everything, and do correlation analysis."
Wait. Did you say voice? Are you kidding me? "Sure, the only true voice of the customer is the actual voice of the customer," says Tom.
When you think about it, it makes sense. Many of the interactions between your customers and your customer-care representatives take place in the form of telephone conversations. There is gold in those interchanges. But without a way to capture and analyze the words being exchanged, it's sort of not much help.
The only way to get a complete picture, Tom insists, is to look at every data point, including voice, and then correlate them all together. It's a head-spinning task, but these guys are pretty far ahead in achieving it, I am learning. "Plus we have some technologies that help. We have compression techniques that allow us to either use the cloud for storage, or on-premises. It comes down to where you're storing, what you're storing, etc.," Tom explains. "What's important is connecting the content—the voice recording—with the metadata around the call. The metadata tells you that the call exists, and then we can store the actual voice call anywhere. You can store terabytes of voice in the cloud, but we recall it using 1K or 2K of metadata. So it's an efficient way to handle large amounts of data."
Another factor in this is the ability to clean out the garbage. What do you want for lunch? Honey, bring home a loaf of bread. That kind of thing. You clean that stuff up by applying metadata to business transactional calls. Using phonetics and other tools you can largely avoid hitting those non-essential conversations. "You can't get to 100% of removing the false positives," admits Tom. "But there are several methodologies for sorting out the good from the trash, including manually—the agent can go in and say ‘don't bother recording that. It's not relevant to anything.' And we can also automatically scan for the incoming number, and know ahead of time that anything coming in from that number is garbage."
Racing Into the Future
It's right on the sharp edge. I wondered whether he thought this technology qualified as "artificial intelligence?" It's one thing to record a conversation and listen back; it's quite another to correlate terms together, organize the conversations by subject, categorize the conversations into "file folders," if you will, in order to apply some structure to some of the most unstructured elements on Earth-human voice and written text. How do you do that?
"It's very much that. It's taking processes that used to be very manual. Before, the best you could do is listen to a sample of calls. Now we can listen to 100% of contacts, whether they are voice, social media, text, whatever, and correlate those in a fashion that gives you results you're looking from a results-oriented analytic output," Tom says.
OK. Tom says it's like looking for a needle in a haystack. Like I always say, that's easy—it's much harder to find a needle in a needlestack. In other words, differentiating content objects (in this case including recorded conversations). "I like that. It's not unlike voice. Think about all the terabytes of data in a contact center. Finding a word among all the thousands of words in a contact center is like finding a needle in a needlestack."
It's not a leap of imagination to assume these guys work for some of the intelligence community. After all, listening to telephone calls, associating who talks to whom... that's pretty spooky (and in-the-news) stuff. "No. We don't have customers like that. We want to listen to the voice of the customer. We never spy on citizens," he insists. I believe him.
For Tom, the "voice of the customer" is the prime directive. "Simply, how that's done is this: A voice call is just a number of phonemes that come together. There are great engines out there that put those phonemes together," he says. I have to take his word for it. This is above my pay grade.
There is some Star Trek in here, admittedly. There are currently a few dozen Calabrio customers who are taking strong advantage of this solution, says Tom. "But it's real, working, in-the-field at multi-billion-dollar companies." I asked for examples. "It's being used in compliance, where certain words aren't allowed to be used, in case agents are going off-script and creating a compliance risk. And in sales tools, where they are learning more about what their customers and salespeople are saying that result in either a ‘no-sale' or a ‘sale,' to make them better at what they do," Tom says.