Speech Technology with Impact - Speech Analytics Evolves a Step Further

Speech analytics (SA) is the application of speech technologies to the analysis of discourse, whether the speech consists of recorded calls from a contact center, wiretap, or some other form of media. A dozen or more vendors are now marketing such analytics, solo or in conjunction with complimentary products, primarily to companies for use in contact centers. Companies marketing SA include, but are not limited to, Utopy, CallMiner, Witness Systems, Mercom, SER Solutions, NICE Systems, Autonomy, Nexidia, and VoicePrint International.

The development of speech analytics came out of the thinking that existing analytic methodologies which typically only measure call center metrics - such as agent talk time, hold time, etc. - or post-call business process analytics - such as number of returns, amount of sales, etc. - leave a tremendous amount of valuable customer interaction data undiscovered and underutilized. Moreover, due to the logical restraints of not having the manpower to manually monitor all recorded calls, only a fraction of agent/customer interaction data is truly utilized. Speech analytics were developed as a way of mining this hidden wealth of data in an automated fashion in order to further help companies reduce churn, improve agent training, increase competitive awareness, improve first call resolution, and perform root cause analysis, among other call center metrics that managers worry about.

As with any technology, there are numerous approaches to the analysis of speech, from word spotting to emotion and intent recognition. This month we will look at the latest releases of two vendors supplying SA - Utopy and NICE Systems.

Utopy - SpeechMiner
Utopy, one of the premiere players in the speech analytics space, (http://www.utopy.com/) announced significant enhancements to its SpeechMiner platform in May. Simplistically, SpeechMiner identifies, evaluates and catalogs events within recorded conversations based on words and entire phrases in context of conversations. One of the enhancements to SpeechMiner in the 3.5 release is a significant increase in the applications ability to accurately capture the meaning of conversations. SpeechMiner's "Deliberate Listening Technology™" is way beyond word spotting.  SpeechMiner doesn't just count the number of times a person says "I want to speak to your supervisor," it classifies those by reason and intent, further defining the issues behind the request. With SpeechMiner SMART Administration tool, an administrator can very rapidly create dozens of reason codes and create call type categories, so when SpeechMiner searches for "supervisor" requests, those calls are located and separated into any number of categories from positive and negative reasons, and sub-reasons within each. This intent modeling provides valuable feedback to supervisors as to why customers are making requests, and how they should be handled. The system enables any form of key performance indicators to be monitored, including comparing agents by call type, reporting on training effectiveness, upselling and cross selling skills, and adherence.

As part of the 3.5 release, Utopy also added emotion detection capability, which checks for certain types of callers based on their manner of speaking, further enhancing the applications ability to capture and classify types of calls.

For using speech analytics in corporate environments, Utopy has upgraded SpeechMiner's ability to support multi-site locations, added centralized user management and built-in site partitioning so that managers can assign user-specific partitions for viewing data, preserving the privacy of other sites.

Perhaps most important of all, Utopy claims that they have greatly increased accuracy far beyond what the rest of the industry reports with classification rates at customer sites typically in the 80-85 percent range and as high as 97 percent for some call type categories.  Even if the percentages were far less, this type of accuracy combined with the ability to capture and categorize all recorded calls into finite buckets, has the ability to bring visibility to all sorts of issues that need to be addressed within an organization, whether it is for  improving the day to day running of a contact center, addressing back office problems, supporting sales activities, or analyzing business processes that need to be fixed in order to reduce call volume into a call center.

NICE Perform™
Having just read a book on identity theft, I was particularly intrigued by the addition of fraud detection to NICE Systems' NICE Perform™ application with their Multimedia Interaction suite. NICE Perform™, introduced in July of 2004, is a suite of tools that allows business analytics to be performed on multiple data sources across an enterprise - gathered and stored in a centralized data warehouse - to give a complete picture of what is going on with customer interactions. The addition of access to multiple data sources allows customers to go beyond just mining call center statistics, and lets them farm information from other channels of customer interaction including customer surveys, agent screen activity, call flow events, and other business data sources. Perform can capture all interactions, including agents' screens and recorded calls.

On top of business analytics, NICE employs speech analytics in the form of word spotting to search for pre-defined words and phrases, emotion detection to understand context and spot stress-related calls, and talk pattern analysis to better understand turn taking and conversational pauses between agent and customer.  Users of NICE Perform™ can manipulate aggregated data in any number of ways to gain understanding of customer intent, run current and future trend analysis, analyze statistics on call flow events, do audio analysis on agent/customer conversations, and evaluate customer feedback, etc.

In addition to multi-channel analytics, NICE has added speaker authentication to their suite, allowing companies to add a new level of fraud detection as well. The NICE fraud prevention system is a software-only solution that complements NICE's other offerings either as a standalone application or in conjunction with other data sources. The system enables corporations to prepare a high risk profile database of known "fraudsters'" voice prints to be used in comparison with authenticated users, allowing alerts to be placed on events where no extraordinary transaction has taken place. Being the recording vendor as well as the analytics provider, the system extracts the necessary "voiceprints" from already recorded interactions of customers, and hence no enrollment is needed. As with other call analytics, calls that utilize this database can be run based on rules. For example, verification can be run only on calls in which new accounts are being opened, addresses changed, or credit is extended. The use of the system can potentially alert a center when fraudulent activity is being perpetrated; stopping a known identity thief or helping identify one before any irregular transaction has been reported.

Financial losses from fraudulent use of credit cards and from complete identity theft are a huge and growing problem. Today, most organizations get post-notification when they fall victims of fraud, (usually it's the consumer that calls in the alert when they find unauthorized transactions on their credit card statement) and they lack the ability to prevent the same thief from doing it again. The creation of a black list to prevent fraud and of a "white list" of known users, in the first place, to be used in conjunction with customer data, such as CTI transactions etc., won't cure the problem, but can go a long way in stemming fraud to begin with. It will be extremely interesting to see what applications NICE develops with this new addition.

Have a cool, or noteworthy announcement? Please email me at nsj@jamisons.com

SpeechTek Covers
for qualified subscribers
Subscribe Now Current Issue Past Issues