What Every Call Center Needs To Know About Speech Analytics

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 customer and agent behavior. In fact, these systems can now actually detect emotions, giving organizations a powerful new way to gauge calls while they are in progress.

Today’s speech analytics solutions have uses beyond just improving agent performance. It can be argued that they now have greater value outside of the contact center than inside. That’s mainly because today’s solutions are capable of mining all of the call data, as opposed to just a small percentage of it, which enables organizations to see a more accurate view of the “big picture.” As a result, speech analytics are increasingly being used by executives in sales, marketing and operations to increase revenue, reduce costs, increase customer retention and improve the overall performance of the call center. As organizations come to adopt these solutions, the role of the call center is being redefined into that of a “business intelligence center” — the central hub of intelligence for the organization.

________________________Once the province of large enterprises only, speech analytics are increasingly starting to travel downstream in terms of the size of the organizations that can benefit from it.
_______________________

Once the province of large enterprises only, speech analytics are increasingly starting to travel downstream in terms of the size of the organizations that can benefit from it. “Interaction analytics can help any size of a contact center,” said Barak Eilam, vice president and general manager, interaction analytics, at NICE. “Obviously, large centers face greater management challenges and as such can benefit from the technology significantly. However, we have also been working with small call centers that are experiencing fast-paced growth. Implanting interaction analytics early in the development cycle of such centers helps them to improve significantly the management of such rapid growth and to meet the associated challenges. Any contact center with more than 50 seats can benefit from the technology and see a very good ROI.”

Don’t Confuse It With IVR Technology
There is a misconception that speech analytics solutions have to be as accurate as IVR solutions, in terms of speech recognition, to be of real value. The erroneous conclusion is that because today’s speech-enabled IVR systems are not entirely accurate that, similarly, speech analytics aren’t entirely accurate either. It’s important to consider that IVR is an application that comes into play at the front end of an interaction, whereas speech analytics are applied during and after the call.

“When people hear the word ’speech,’ they tend to think ’speech recognition’ — they think of IVR and desktop dictation systems, which speech analytics is not,” said Jeff Gallino, president and CEO of CallMiner, a speech analytics company. “Even with speech recognition systems, they have ridiculously high accuracy, and people still aren’t satisfied with them. That’s because it is a transactional based accuracy problem - they’re interested in the accuracy at the moment, not ’is it accurate enough that I can run my business on the results?’ When we talk about speech analytics, we talk about the ’business accuracy’ — how accurate is it to run your business — and that’s a major distinction.”

Dan Bodner, CEO and president of Verint, said, “It’s not about trying to measure people’s responses accurately, it’s about being able to analyze overall trends and getting the big picture — but getting the big picture in an accurate way. Customers using our solutions are finding that the accuracy in terms of categorization is very good now — we’re about 90 percent accurate — depending on the type of issue and the type of company. That’s enough to really drive insight. You have to realize that even incomplete information is better than no information. If you have no idea why people are leaving and you can find five reasons why that’s happening, you can do something about it.”

Not Just A Call Center Application
In addition to improving traditional efficiency metrics such as first-call resolution, script compliance, agent performance and cross-sell/upsell rates, companies can also gain valuable insights into what their customers want by mining the data retrieved from their recorded calls.

“The great value of speech analytics is that it opens up near limitless opportunities to gain a better understanding of the customer intent or attitude,” said Roger Woolley, Senior Vice President of Marketing at Autonomy/etalk. “At the most basic level, speech analytics simplifies access to recorded calls so that the call center can quickly, and without manual resources, find specific conversations, types of conversations or conversation trends for use in identifying at-risk customers, agent issues, training opportunities or process improvement initiatives. By providing an easy approach to finding relevant calls, speech analytics helps to bridge the gap between the call center and the rest of the organization so that the entire organization can analyze customer interactions for valuable insight into buying behaviors, product and competitive issues, as well as trends, opportunities and risks.”

Anna Convery, senior vice president of product management and marketing for Nexidia (News - Alert) Inc., said what’s really exciting about today’s solutions is that many opportunities for improving the customer experience are uncovered. “This ’voice of the customer’ insight can be shared outside the call center with other areas of the business, such as product management and marketing, enabling the customer-centricity that many companies aspire to but often fail to realize,” said Convery.

Not All Solutions Are Created Equal
Although most of today’s speech analytics solutions use a blend of phonetics and linguistics, some are more “bare bones” and do only word spotting (these are phonetic, or “speech-to-text” solutions), while others use a “whole vocabulary” approach (these are the linguistic, or large vocabulary continuous speech recognition (LVCSR) solutions). This latter type can determine the context of what is being said, providing deeper insights into the interactions.

“When you go through a speech recognition process, everyone starts out the same, whether they are large vocabulary or phonetic,” CallMiner’s Gallino explained. “They start comparing the voice signal to a model of what the voice might be saying … and everybody starts with phonemes. They slice the audio into very tiny slices, and then they guess, and say, ’this sounds most like ah or duh or eh or one of the other phonemes.’”

What comes after the “speech-to-text” stage is where things get really interesting: This is where vendors take those tiny slices of sound and create words out of them to build a vocabulary. They then build sets or groups of words that make up common phrases. This enables the speech engine to interpret not only the words, but whole phrases, so that it can understand the context of what is being said.

Some vendors have seen great success with deploying speech analytics as a managed service. With a managed service, the vendor hosts the call data and mines the information using its own preset methodologies. There are differences in architecture, as well, with some systems taking a completely centralized approach and others utilizing a hybrid, distributed/centralized approach. Which one is right depends on what type of call center operation you are running and what type of architecture you are using.

Do You Want To Mine In Real Time?
Thanks to faster processors and the advanced new algorithms being employed, today’s speech analytics are capable of mining calls in real time — using analytics to trigger screen prompts that “intervene” and tell the agent what to do during or call, or alert a manager that he or she may need to take action — though not all companies require this real-time response. Some call centers see negative unintended consequences in such real-time action triggering.

“Even though the engines can do analysis very quickly, and could be used to send up screen prompts to agents to help them react immediately, this is not recommended as you are shaping agent behavior in a negative way,” said Verint’s Bodner. “Furthermore, there’s no point in have it do analysis of what just one caller is saying when the agent is already trained to do that for you. The reality is that at this point during the call, they probably have better judgment than the system does.”

Still, many companies are developing solutions that can assist agents in making the “right” decision should a call happen to go awry. As Autonomy etalk’s Woolley said, “This can allow both for ’save-the-customer’ alerts and, in the case of using speech tools for information delivery, processing the conversation to offer information to the agent for more effective responses.”

Today’s Speech Analytics Solutions Can Be Easily Integrated With Existing Systems
It’s true that solutions that come pre-integrated with recording and quality monitoring solutions tend to be easier to implement. The challenge is not so much the actual integration, but in putting together a system that works smoothly and accurately.

“Technically integrating speech analytics applications into contact center workflow and processes is not difficult,” said Patrick Botz, global director of marketing for Voice Print International (News - Alert) . “Most speech analytics vendors can easily analyze the majority of recorded call audio and offer canned reports so that organizations can quickly identify valuable business intelligence. Most systems include notification or escalation mechanisms to ensure the insights gleaned and related call recordings reach managers who have the power to make decisions.”

_______________________It’s important to consider that IVR is an application that comes into play at the front end of an interaction, whereas speech analytics are applied during and after the call.
_______________________Fast ROI
Although they can be pricey, today’s speech analytics solutions can achieve a very fast return on investment (usually in three to six months). This is, however, highly dependent on the organization knowing its business inside and out, and therefore knowing exactly what to look for when mining calls.

With speech analytics, the ROI comes in many forms — for example, in the call center itself, you can use it to reduce agent call handle time, identify additional opportunities to cross-sell and upsell to customers and improve call resolution. Furthermore, call centers reduce the need for supervisors or managers to do manual call monitoring to uncover problems.

“With speech analytics, you can reduce manual listening by classifying calls into buckets by topics and subtopics, so that only relevant calls are monitored,” said VPI’s Botz. “Since most contact centers listen to over twice as many calls as necessary, implementing speech analytics can immediately reduce the time required to review calls by an average of 50 percent.”

In the early days of enterprise speech solutions, we saw a technology standing alone in the corner, looking for an invitation to join the call center party. It was widely assumed, at that time, that call routing (speech-enabled IVRs) would be speech’s call center “killer app.” Though speech has vastly improved the humble IVR, recent studies have shown speech analytics to be the number one growth technology for call centers. It seems the conventional wisdom was wrong, and speech has found a very comfortable — and profitable — niche in the call center while most of the rest of us were busy watching the IVR.

The author may be contacted at [email protected].