Speech recognition and speech analytics are separate yet related technologies, and each is having its own influence on the contact center. As we discussed in part one of this two-part series (see the November 2007 issue of this magazine or visit www.tmcnet.com/1428.1), speech recognition has evolved rapidly in recent years. Generally, these automated systems are used for handling informational or low-value transactions, however, as they become more advanced they are being used to handle more complex transactions, as well as to convey the company brand, philosophy and culture through the “personas” they sometimes assume.

Considering the rapid adoption rate of speech-enabled self-service solutions across the industry, one might come to suspect that most enterprises would prefer to just automate every transaction and do away with their agents completely. Although many centers have realized staffing efficiencies from the implementation of self-service systems, most will tell you that the real purpose is to boost agent productivity and improve customer service, not to reduce head count. Organizations aren’t buying speech solutions and then saying “Great, now we don’t have to talk to our customers anymore,” rather they are saying, “Good, now we can free our agents to focus on the interactions that mean the most.” Most people in the industry agree that regardless of how sophisticated these solutions become, there will always be a need for the human touch in the
contact center.

“What’s interesting — and where we spend a lot of time working with prospective and current customers — is this perspective that you need to automate 100 percent of the calls,” said Tim Pearce, Global Solutions Manager for Speech at Dimension Data, a global IT services company. “People get very emotional and attached to the idea of 'What’s the automated call completion rate going to be?' Yes, that is important and yes, that is a business driver — but that needs to be balanced against customer experience. It wouldn’t be too difficult to create an application with no dropout to an agent, so you are entirely beholden to the self-service application — and you will sit in that application until you complete your transaction or until you hang up. But that clearly is counter-productive. You need to bring the customer service element into it.”
For years there has been a perception among consumers that organizations that use speech-enabled self-service solutions don’t really care about the customer experience — that all they’re trying to do is save a buck by implementing an automated system rather than pay someone to help customers in person. This perception has been perpetuated by the bad experiences consumers had with the early IVR systems, which generally didn’t work well and merely frustrated people. Now, the only way the industry can reverse that is by demonstrating the accuracy and reliability of today’s self-service solutions and showing that they actually deliver an improved customer experience. With today’s speech-enabled self-service systems, callers can experience shorter hold queues and faster call handling time — the two main drivers of customer satisfaction — and perhaps best of all, they can be in full control of the interaction as it progresses through the various stages.

Leveraging Speech Analytics
Speech analytics is software that analyzes or “mines” recorded interactions for the purpose of revealing trends in customer or agent behavior. Unlike the speech analytics solutions of a few years ago, which were mainly used to perform simple “word spotting” in recorded conversations, today’s solutions can identify the context of what is being said, giving organizations the ability to more accurately categorize calls and gain new insights into their customers and agents.

With today’s speech analytics solutions, call center managers can mine or search all call data recorded during a particular period and look for trends and the root causes of those trends. This is a huge advantage compared to traditional quality monitoring, where only a small sample of calls are manually listened to post-transaction to make sure agents are in compliance with business objectives. It can be argued that today’s speech analytics solutions are the “new QM,” however, it should be noted that most centers that have speech analytics solutions in place typically continue to monitor calls manually, simply because there are things that can be discovered through manual monitoring, such as quality of service, that can’t be uncovered by speech analytics.

“With traditional call monitoring, you really only get a small sampling of the calls — maybe five calls per agent per week — but the agent might have taken 500 or 1,000 calls that week,” explained Daniel Ziv Director of Business Analytics, Verint Systems. “Usually, with traditional QM, you’re looking for only a few specific things like, did the agent do the opening correctly, or did they read the script on those samples. Now, that’s valuable because if you have an agent who doesn’t really know anything, then that will surface fairly quickly. But most agents do know something. They can handle, say, 80 percent of the calls pretty well. And that random sample might not surface their areas of weakness — like a repeat call or a call about a new product or a call that is challenging or emotional. Those situations are actually more important as they have more impact on the customer and more impact on the organization. It could be someone calling to say they’re going to leave and the agent doesn’t know how to retain them. Those can sometimes be missed using a traditional call monitoring program. So speech analytics can surface areas that the general agent population has problems with, and that particular agents have challenges with compared to other agents. To find these trends, you really need to look at a larger sample, and speech analytics looks at all of the call data, not just the one percent covered by traditional QM. With today’s speech analytics, you can quickly search 10, 20 or even 100 percent of the calls. So it gives you a much larger sample — and can statistically identify real areas where agents need help.”

Ziv pointed out that after identifying the agents who are having problems, “you can immediately drive additional training or coaching to them. We really see this as impacting the way centers do their quality monitoring, coaching and training,” he said. “We call it focused quality.”

Ziv cautioned, however, that organizations need to be careful not to overreact to one-time events that are not representative of an overall trend. For example, if it’s just one agent who didn’t handle one call particularly well, simply due to some noise on the line, “you might not want to send that training module right away, because it could have been just a bad call, or it could be that the recognition [software] didn’t categorize something properly.”

Most of today’s speech analytics solutions are capable of searching recorded interactions in near real time, and thus can be used to uncover agent performance problems almost as they are occurring. With these “real-time” capabilities, contact center managers have the means to detect when an interaction is going “sour” and then take appropriate action — sometimes while the customer is still on the line. If a problem is detected early enough, a manager can send a screen pop or IM to the agent, directing them how to bring the call to a positive outcome, or they can intervene on the call, either by barging in or using whisper coaching, for the purpose of preventing a customer from defecting.

However, there is some debate in the industry as to whether real-time speech analytics is really “real” at this point. There are those who say that the technology has advanced to the point where it can now be used for this purpose, but there are others who say real-time speech analytics is the “Holy Grail” of speech analytics and is still a number of years off.

“There is this buzz going around that you can do analytics in real time, while the call is taking place — that you can see, in this interaction, that the customer is getting upset, and you’re going to buzz the agent with some popup or extend the call to the supervisor — I don’t really see this as happening, yet,” Ziv said. “It could be that one day we’ll get there, but it’s hard to say when we’ll reach that point.”

On the other hand, some vendors see tremendous value in using speech analytics to detect problems in near real time and then intervene and guide a call to a successful conclusion. Those who are already jumping onto the real-time analytics bandwagon will tell you that once a customer is lost, he or she is gone for good, so it's in an organization’s best interest to try to salvage each interaction gone awry, using the technology that is currently available.

Tim Kraskey, VP of marketing and business development at Calabrio Software, said although he agrees that real-time speech analytics is the “Holy Grail” of the contact center industry, it is possible to use today’s “near real-time” solutions to salvage calls that have taken a turn for the worse.

“On one level, there’s analytics that has nothing to do with phonetics — the analytics that understands periods of silence, inflections of voice, amplitude or speech modulations, things that have nothing to do with phonetics,” Kraskey explained. “Then you add phonetics on top of that, too, such as George Carlin’s ‘seven words you can’t say on television,’ a competitor’s name, and statements such as ‘I want to speak to a supervisor,’ and ‘I want to cancel my account.' So it’s a combination of both the underlying analytics along with the phonetics. Then, once you have that in place, you can do some sort of thresholding, so that when a certain threshold has been breached, the supervisor can get an alert and then silently monitor the call and intervene if necessary.”

Kraskey said Calabrio has thousands of contact center customers that have had success using speech analytics as a quality monitoring tool to improve agent performance. Some of these customers, he said, are also using it with some success to detect problems in real time and then intervene on the call.

Grant Sainsbury, practice director of customer interaction solutions at Dimension Data, on the other hand, said he has “yet to see effective analytics in a real-time environment.”

“It is still primarily offline batched analysis — and that continues to hold its value,” Stainsbury said. “It gives you a chance to identify problem transactions — customers who are at risk of churn ― and try and make things right before they do churn. Right now that is the Holy Grail of the industry: real-time analysis.”

Sainsbury said when it comes to call monitoring and intervention, the majority of his customers “tend to focus more on the screen and coaching the agent visually, in real time, through the call,” rather than using more intrusive methods such as barging in on the call or using whisper coaching.

If there is one thing for certain, it is that speech solutions for the contact center will only continue to evolve and improve in the years to come. As organizations continue to realize the operational efficiencies which can be gained through both speech recognition and speech analytics, they will continue to adopt these solutions and use them to improve contact center (and enterprise) operations.