Mining for Meaning
But the biggest goals in speech analytics (increased sales, smartermarketing, and improved business operations) will be hard to come bywithout partnerships. Vendors with partnerships bring to the table notonly their own capabilities, but their partners’ services as well.Stockford notes that from a competitive perspective, "any companythat’s not going to either partner or acquire technologies will notremain competitive."
The Debate Continues
The argument for speech analytics is, in some ways, far from over.Though case studies with large companies like Comcast have shown howextracted data can improve business outside the contact center,executive buy-in and sales to businesses with less than 100 seats arenot hitting the mark. Vendors and analysts agree that speech analyticsworks, but only if a company works speech analytics. That means morethan just pressing a button and walking away; to get the problemsolved, time, money, and manpower must be allocated. And, as morecompanies begin to view the contact center as a place for sales and notjust help, analytics could deliver on its promise of trulyunderstanding a customer.
In coming years, the market could also morph to more closely resembleother facets of the contact center technologies space. "There are lotsof players in the market, but there are different niches now," Nuance’sBorton says. "This market is going to differentiate into probably fiveor six submarkets in five years because that’s the typical pattern oftechnology."
CallMiner’s Gallino says the future of speech analytics lies primarilyin the hands of the customer. "Deep down, people are a bit intimidatedby the concept of speech analytics, the idea of automatinghuman-to-human interaction," he states. "If you look at the hype cyclefor speech analytics, we’re coming up the other side of the trough.Some of the promise is now starting to be delivered."
LVCSR versus PhoneticsThe two methods used for audio mining in speech analytics, Large Vocabulary Conversational Speech Recognition (LVCSR) and phonetic search, rely on different types of technology to extract data from recorded conversations. While both have proved reliable, vendors and analysts agree that sometimes one method can trump the other, depending on a company’s goals. A breakdown of the two methods follows.
LVCSR: Used in extracting specific words preprogrammed by a customer or application developer, LVCSR not only finds words, but also helps users grasp the context in which the word or phrase was uttered. While more CPU-intensive than phonetics, LVCSR is often more accurate.
"LVCSR is better than phonetics at detecting subtle word differences," says Larry Mark, chief technology officer at SER Solutions. "If you’re talking about a disclaimer where it matters—if I say, ‘Past performance is no guarantee of future returns’—that kind of difference in a longer phrase you won’t pick up in phonetics."
Phonetics: Rather than create grammars before implementation, as required by LVCSR applications, phonetic searches translate conversations and allow those analyzing data to pick up key words, phrases, or topics. Often less expensive to implement, phonetic search is typically used as a "first tier" form of data analysis—users mine data with phonetics, and then go back with LVCSR to refine the search. "Phonetics is still a great technology for lots of apps," Mark says. "It’s good for finding competitive mentions and understanding what agents are saying."
The Verdict: Vendors and analysts interviewed agree that speech analytics companies will need to offer a combination of both search methods to stay competitive. Rather than competing over which method is better, Donna Fluss, president of DMG Consulting, says vendors and users should see the two as complementary. "It’s unlikely that an application would run all conversations through both technologies," she states. "It’s more likely that an application would use one of the technologies and then refine the analysis with the other one.
Is Real Time Really Necessary?Vendors like CallMiner offer audio-mined data results on a real-time basis, giving users instant insight into problems that could have arisen within a contact center or on an enterprise-wide level. Of course, quick access means more money. We asked analyst Paul Stockford from Saddletree Research to weigh in.
Speech Technology: Is it compulsory for companies to have data produced in real time? When is it not necessary?
Stockford: We’re talking about a call center, not national security issues. Time is relevant, but it’s not the defining factor. There are advantages [to real time], but you take a service scenario where you take a five-day chunk, it gets processed into a text format, you can scan it, and that’s where you start gleaning things and asking questions you didn’t know you should have asked before you did the scan. You can go back and dig deeper and deeper into that single factor or group of factors as opposed to having to keep processing all the time. But if you need [the information] really fast, then real-time makes sense, but I don’t think there are that many call centers out there that need it immediately. I think it’s way too early to say whether one will be preferred over the other