The Why Factor in Speech Analytics
Speech analytics (SA), the process of analyzing recorded calls to glean information, brings structure to customer interactions and sheds light on information buried in customer telephone interactions with an enterprise. But the value of SA is truly realized when recorded and analyzed data is reapplied to performance pockets within a business. Operations, marketing, and sales all stand to profit, and SA may even help the contact center evolve from a costly liability to a strategic asset for the enterprise.
According to Mariann McDonagh, vice president of global marketing at Verint Systems Research, only 20 percent of customer complaints are agent-related, and the remaining 80 percent stem from product or process issues and misaligned expectations. "Speech analytics solutions that go beyond keyword spotting to provide automatic categorization and root cause analytics automatically surface these issues."
So, if you have seen sales and marketing campaigns fall short of expectations, SA can help pinpoint why. Are incoming calls up? Again, speech analytics can guide businesses to clues. Are customers jumping ship at a higher rate? SA and its root cause analysis capabilities can offer reasons for it. Do you poop out at parties? No, wait...that's for a different topic.
Rooting around in the contact center
The process of identifying why customers are calling the contact center begins with SA applications, which are primarily used to perform root-cause analysis. Getting to the underlying source of contact center misfires, instead of tackling obvious symptoms, helps companies develop corrective or preventive processes for issues. Some of these misfires, say, customer dissatisfaction with a current product, may not have been associated with the contact center prior to an SA review. "A lot of times [companies] knew what issues were, but once it's on a report like this it's hard for other departments to argue against it, and it's hard for your own department to argue against it," says Donna Fluss, principal of DMG Consulting and author of DMG's "2006 Speech Analytics Market Report."
Naturally, identifying and fixing the problems that prod customers to make a call to a company can reduce incoming calls and cut costs. And customer experience and retention rates improve, which can lead to increased sales. A good example is 93-year-old CCH, a Wolters Kluwer business that provides tax and business law information and software. The firm produces about 700 publications in print and electronic form, and CCH had been using rough-edged call-analysis methods to track interactions, such as occasionally surveying CSRs for their impressions of calls and attempting to bucket calls using paper forms and a proprietary CRM application. With no formal speech analytics process in place, the company found itself with restricted ability to analyze recorded content. CCH uses Witness Systems' call recording and quality monitoring capabilities to evaluate its CSRs' performance and call handling.
"Customer service representatives cannot possibly be expected to remember the details of all of their calls every day, so surveying them only resulted in a limited amount of information," says Benjamin Gettinger, former business analyst of customer service and operations at CCH, now senior analyst at Accenture. "It proved difficult trying to bucket the calls as they came in, because often there would be multiple issues on one call. It became a difficult decision as to which bucket to put the call in."
CCH selected Utopy SpeechMiner as its speech analytics platform. The deployment went live in April 2005 with the help of CCH's internal resources and Utopy's Client Solutions Group. Using data culled from customer interactions, CCH can now identify trends and address operational and strategic issues like which customers are dissatisfied and why, which agents would benefit the most from additional training, and which customer issues are taking up most of the agents' time.
"The speech analytics software does an excellent job of measuring hold times by looking for silence, as opposed to normal telephone measurements that monitor the hold function of the phone," Gettinger says, citing an example in which CCH used the SA application to see where hold times were slightly increasing. "The team leaders were then able to take action and coach their team members to reduce the hold times to their previous levels."
Another opportunity for improvement took shape when CCH learned that agents were struggling with selling products to callers who wanted product information. That insight allowed CCH to craft a corrective plan. It worked--selling efforts shot up 400 percent within just three weeks, resulting in a 91 percent sales increase for that campaign.
Not Just For the Contact Center
"Before the advent of speech analytics, [customer interactions] had been not only the domain of the call center--mostly for quality monitoring--but had been barely tapped," says Cliff LaCoursiere, cofounder and senior vice president of business development at CallMiner. Speech analytics "gives the enterprise access to a really important company asset--agent-customer conversations." SA can help legal and compliance departments, for example, use agent-customer contacts to pinpoint how well CSRs are adhering to scripts for things like greetings and closings. Speech analytics can help determine if agents are properly verifying callers.
The value of SA is real; the technology itself isn't new. Government agencies were among the early adopters of the technology, and use speech analytics to sift through vast volumes of conversation to identity potential threats. SA vendors like CallMiner, Nexidia, and Utopy develop solutions that help the public and private sectors mine information from recorded interactions to gain valuable insight.
SA apps can also be used to glean intelligence from operational metrics. Anna Convery, senior vice president of marketing and product management at Nexidia, describes the company's recent managed-service engagement with a large Internet-service provider, which included taking and analyzing data from the provider's customer retention branch. Agents often live by average call-handle time, but in this case, Convery says, speech analytics uncovered that when the agent "spent a little bit more time on the call with the customer, he or she actually had a significantly higher [propensity] of retaining the customer."
Convery uses one of Nexidia's telecommunications customers to illustrate another example of how marketing departments can benefit from speech analytics. The telecom, like many of them, lets customers find on its Web site how many minutes they've used. But when the marketing department made a change to its Web site, its contact center was flooded with customers calling to request minute usage. An analysis of the recorded calls revealed that the Web-site change made the usage section of the site difficult to find. "Overnight we were able to audit about 1,300 hours of audio and find out what was going on," Convery says. "If they were doing what they had normally done, it would have taken them as long as six weeks to get into those calls manually and find out what was going on."
Just another customer service call can, also, be converted into an opportunity to capture data that's useful to other units, like sales. Paul Stockford, chief analyst at Saddletree Research, says, "The opportunity to mine customer data and listen to it with a more analytical ear as enabled by speech analytics is going to open up worlds of opportunity throughout the enterprise." Stockford describes an example of an insurance company that gets a call from a customer who wants to change her son's address on her car insurance policy, because he's leaving for college. The agent enters the new address and ends the call. But, Stockford posits, had the company been able to use an SA system to analyze the interaction, information that might have been used for possible sales opportunities may have been revealed: The address change indicates that the son will have a separate residence, which represents an opportunity to add a renter's policy to the family's coverage.
In fact, if implemented properly and used just by the contact center, organizations will realize a payback on their speech analytics deployments in about nine months, according to Fluss. But if the contact center, sales, and marketing departments all use speech analytics the payback will be in less than five months.
Take Your Pick
Speech analytics as a method or system can vary in execution, depending on which vendor you speak to, but most industry pundits agree that there are generally two ways to automatically audit recorded calls: speech to text and phonetic. Speech-to-text offerings are based on a large vocabulary continuous speech recognition (LVCSR) engine and translate recorded audio into searchable text. Phonetic-based products scan the recorded call itself. These offerings identify the original audio as a string of phonemes (the component parts of language) and match queries against it to return audio files that match query criteria, according to Ri Pierce-Grove, associate analyst of technology at Datamonitor. CallMiner, for example, uses the speech-to-text method, while Nexidia applications are phonetic-based.
The CallMiner Analytics Suite uses speech recognition, data mining, and trend mining techniques to convert spoken word to text and data (including elements like silence, stress, and tone) to analyze call content and intent, and classify calls. It also allows users to create ad-hoc and automated reports. CallMiner partners with recording and monitoring providers Mercom Systems, Voice Print International, and Witness Systems, but according to CallMiner's LaCoursiere, the company can read and mine data from pretty much any recorder.
Of course, each has its limitations. "You're risking losing information in the process of translating from speech to text," Stockford says. But "in evaluating speech directly it's a time issue. It takes a lot longer to get where you might want to get to," he adds. Speech to text "is a longer implementation at this point than phonetic," Fluss says, but with phonetic "you have to specify what you're looking for. You can't just say here's a trend that I found that's beyond what you asked about."
It is important to note that within the contact center, the adoption of speech analytics is still in its infancy. Most companies today are either counting on managers to manually evaluate a portion of recorded calls, relying on individual agents to analyze calls, or aren't performing any analysis at all. Informal processes are inefficient, ineffective, impractical, and a waste if the information isn't used in a timely fashion. This becomes even more disturbing considering that during interactions with contact center reps, customers reveal valuable information such as their satisfaction (or lack thereof) with a company and its products and what they'd like improved (or discontinued).
While SA can help lasso, corral, and tame unruly customer info, Fluss notes that there aren't a lot of best practices in the marketplace, a typical achilles' heel for budding markets. Stockford adds that while some customers are "trying to figure out how they might be able to use speech analytics and what this technology is all about, in the R&D labs of speech analytics companies they're already talking about the third and fourth generation."
"If you're going to make business decisions you need analysis and that analysis has to be accurate," says Ted Lubowsky, executive vice president of Utopy. "Otherwise you're going to be making decisions and changes based on things that maybe aren't true." Fluss says that most of the innovation centering on accuracy improvement spreads across three areas: speech engine, queries, and data aggregation and reporting. She also expects to see improvements in ease of use and implementation.
Another huge innovation area will be performing analysis on a real-time basis. "Speech analytics and real-time analytics go together," Fluss says. "Once we structure it we can analyze it, and once we analyze it our imaginations are our only limits." On a broader analytical scale, expect to see more SA using traditional BI to draw more intelligence out of calls. "Even though we're providing good reporting on what's happening in the calls, there's a lot more value that standard data-mining techniques can [bring] to the call center data to yield even more insight," LaCoursiere says.
There's no question that SA is gaining substantial traction. In 2005, contact center implementations grew from 25 to about 200, and contact center speech analytics applications will realize a growth rate of 120 percent in 2006 and 100 percent in 2007, according to DMG Consulting. "There is so much low-hanging fruit that we can learn from these engines where they don't need to be 100 percent accurate right now," Fluss says. But "by the time we get to that point where the market really needs them to be even more accurate, they will be." Current apps certainly are robust enough to generate considerable value, but there are several improvements on the horizon.
"Customers should keep in mind the technology will only improve in terms of accuracy and functionality, but enterprise customers that record large volumes of calls today can benefit from this incredible audio mining tool," says Roger Wooley, vice president of marketing at etalk. Companies should think hard about the people in the organization who could benefit from speech analytics, and how to plan to train them, Pierce-Grove says. "Like other business intelligence tools, speech analytics runs a risk of being siloed and not being exploited to its full extent. Good organizational planning can mitigate that risk." For companies on the fence about SA, remember this: "It's not a disruptive technology, but an extension of global trends in organizing and mining data," Pierce-Grove says. "In many ways, you could argue that given the customer-centric enterprise, speech analytics was an inevitable development."
Contact Associate Editor Coreen Bailor at cbailor@destinationCRM.com
Talkin' Technology: A Vendor Pocket Primer
Following are functionality facts on two leading SA sellers:
Nexidia searches and analyzes the actual recorded audio, and with its Enterprise Speech Intelligence (ESI) 6.0 release debuted Forensic Search, a feature that allows end users to quickly ad-hoc search large sets of audio. Nexidia isn't strategically aligned with any of the quality monitoring players, but can interface with a variety of recording devices from vendors like etalk, NICE Systems, Witness, and Verint, and homegrown devices.
Utopy analyzes voice data directly and can link phrases together to determine exact meaning within the spoken context. Through its SpeechMiner suite the company can analyze calls and identify events; provide stats on agent performance and customer satisfaction; and allow users to create custom business events or categories. Utopy partners with quality monitoring provider Envision Telephony, but can integrate with recording products from other recording providers.
Other vendors with SA capabilities include quality monitoring players etalk (an Autonomy company), NICE, and Verint. --C.B.