Moving to Next-Generation Speech Analytics:
Words are No Longer Enough
A Frost & Sullivan White Paper
Sponsored by:
frost.com
Executive Summary....................................................................................................................... 3
Introduction.................................................................................................................................... 3
The Complexity Problem.............................................................................................................. 4
The Speech Analytics Answer...................................................................................................... 4
Limitations of Traditional Speech Analytics............................................................................... 5
HP Autonomy’s Meaning-Based Approach................................................................................. 6
Conclusion...................................................................................................................................... 8
contents
Moving to Next-Generation Speech Analytics: Words are No Longer Enough
EXECUTIVE SUMMARY
Speech analytics is one of the most powerful tools on the contact center workbench. Its business value lies in
understanding the precise intent of the customer and driving actionable intelligence.
Traditional speech analytics solutions rely on spotting keywords, phonetics, and phrases, and miss out on the
context, relevancy, and hence the precise intent or meaning of the interaction.
Next-generation, conceptual speech analytics tools can unlock meaningful information and provide actionable
insights into customer and agent behaviors and expectations.
INTRODUCTION
In the past, call centers were charged with a very limited task: respond to voice calls from customers as efficiently
as possible, at the lowest cost possible. That simple mission has expanded to include new technology platforms,
accommodating greater demands by customers and having a much greater degree of revenue responsibility.
Ensuring an excellent customer experience has shifted, in the eyes of enterprises, from an expense that must
be kept low into a vital competitive differentiator. It is not enough to simply react– contact centers have to
anticipate changing business conditions in the customer base. Empowered customers readily turn to social
media to find buying information and express discontent. These customers are also highly mobile, creating
uncertainty about their location, status and even their identity when they call.
Enterprises therefore need to gather, understand and quickly act on customer intelligence while maximizing the
profitability, longevity and satisfaction of each customer. One of the ways that modern centers have engaged in
this more complex mission has been to add a layer of analysis to the data gathering. There is a special emphasis
on the unstructured voice recordings.
Speech analytics software has been helpful in turning voice conversations into vital customer and contact center
performance intelligence. Yet, for many centers, the need for actionable insight is outstripping the capabilities of
early-generation speech analytics.
Traditional speech analytics tools rely on keywords and phonetics. This poses a problem: these solutions miss
out on context and relevancy, both of which are essential to understand what individuals are saying. In short,
they do not provide meaning. Words by themselves are no longer enough.
The agent’s desktop also contains valuable information about the customer being served, which points to the
need for analysis tools that capture and digest an agent’s screen data. The same holds true for conversations
captured on other channels: IVR/speech recognition, Web and mobile app self-service, chat, e-mail, SMS/text
and social media, and freeform notes by retail and field staff.
Speech analytics captures the valuable insights in customers’ remarks to agents. For example: “I was on your
website but I couldn’t get into my account” or “I was in your store earlier today but the salesperson couldn’t
find what I wanted.” Contact center professionals need to get at the most relevant information through speech
and other analytics solutions in real time. The volume of data generated from speech and other channels
is enormous and growing. Moreover, contact centers are not the only organizations that can benefit from
information collected from customer-agent interactions. Accounting, billing, HR, IT, legal, marketing, product
development and sales departments realize great insights by applying analytics. But speech tools involve complex
deployments that often rely on expensive implementation, maintenance, and support. That has forced contact
centers to look to these other departments when making deployment decisions on speech analytics, as they
search for ways to extend the value of the tool enterprise-wide.
3
frost.com
THE COMPLEXITY PROBLEM
There are three ways in which this increase in complexity causes operational problems. First, customers
are coming to the interactions with more varied and demanding expectations. They have interactions that
cross contact channels; they seek out information from different sources before making their calls. They are
increasingly using mobile devices that add to their sense of urgency before and during interactions. All of
these shifts lead to more data that can be collected by the center and then mined for insight. But often that
data is scattered among silos that are not built to communicate.
This leads to the second problem with the increased complexity: most contact centers have not yet built
systems that give them an integrated view of the customer across all touchpoints or through time. Too often,
each interaction is seen in isolation, apart from previous interactions with the same customer, let alone
aggregated segments of similar customers.
According to a Frost & Sullivan survey, only 33% of contact center organizations have fully integrated their
contact channels. This is projected to grow to 54% by 2016.
Figure 1 - Current and Future Integration of Contact Channels: North America, 2014-2016
5%
17%
Don’t Know 40%
Partially Integrated
Mostly Integrated 49% 50%
Transitioning
Fully Integrated to full 64% Growth Rate
integration
54%
33%
Current, 2014 Future, 2016
Source: Frost & Sullivan 2014 Contact Center Survey, N=305
The third element is a traditional contact center quandary, made worse by increasing complexity: cost control
requires that you constantly improve the efficiency and productivity of your labor base. It requires investment
to deliver high-quality service, and customers are unwilling to pay for better service in the form of higher
prices. As a result, companies are under pressure to find new ways to boost the service experience without
letting costs spiral out of control. One way has been to use speech analytics to leverage the enormous
amount of data already collected in the form of call recordings.
THE SPEECH ANALYTICS ANSWER
The contact center generates, through the course of its normal operations, an enormous amount of raw
data about its activities. Data is spun out of nearly every system, from the telephony platform to the quality
monitoring engine.
4
Moving to Next-Generation Speech Analytics: Words are No Longer Enough
Speech analytics is a set of tools that parses that particular set of unstructured data contained in recorded
audio calls. It aggregates voice calls from a contact center and finds patterns in them. These can be patterns
in behavior of a single customer or group of customers. Analytics can also detect patterns in how agents
respond to a particular type of problem or sales interaction.
To illustrate, an ACD report can tell you how long one of your agents was on the phone with a caller. Speech
analytics, on the other hand, can tell you what that call was about and issues that may have impacted the length
of the call. As a result,speech analytics applications have become powerful tools for customer experience
enhancement and cost control. They help centers obtain valuable insights into agent and customer behavior
that bolster contact center performance and productivity. Speech tools use a variety of word-detection
techniques to drill down into customer conversations, turning voice data into meaningful intelligence that can
be acted on. They rely on business rules that launch scripts to agents based on what customers are saying in
real time. It’s the automated rendering of the conversation married with the system’s understanding of the
available enterprise resources (Web forms, FAQ, manuals, scripts, product recommendations, etc.) that can
help resolve the issue or drive up-sell activity. This understanding will get the right resource to the agent.
Two elements give speech analytics its power: the power of voice and its methodology. The spoken words
contain emotion, context, relevance and insights into each individual’s attitude and state of mind. This is
surpassed only by face-to-face, in-person interactions and high-quality video. Speech analytics tools are
designed to perform the difficult task of uncovering meaning from conversations. They provide an ordered,
objective, efficient, systematic and readily retrievable means of obtaining this knowledge – especially when
compared to monitoring live calls or listening to recordings and taking notes.
Speech analytics uncovers ways that agents can deliver higher-quality service and capitalize on cross-sell/up-
sell opportunities. Applications listen for churn risks and provide information on “moments of truth,” leading
to better agent performance and higher revenue. They also allow managers to find the “friction points” that
led to the inbound call. Customers can also give clues about where the leads came from and which new
products or services, pricing or promotion drove them to make contact with the vendor.
LIMITATIONS OF TRADITIONAL SPEECH ANALYTICS
For all of the advantages of speech analytics, several years of use have uncovered limitations of the early
technologies. Conventional tools typically rely on keywords and phonetics. Keyword searches return files
that contain the terms queried by the user, relying on vocabularies in the system. It is simple and intuitive.
Phonetics goes deeper by matching search terms to words broken down into their phonemes, or smallest
compositions of speech, and matches them to words, even though they may not be in the system dictionaries.
While these methods are proven for general or common interactions, they are less effective with more
complex exchanges. As centers try to push as many routine interactions into self-service to control costs,
the calls that make it to agents are more complicated and unique.
Keyword-based technologies risk misunderstandings by missing out on concept, context and relevancy. For
example, they may miss that the sentence, “I have a question about last month’s payment,” is a billing matter
because the words bill, invoice or statement are nowhere in the phrase.
Several speech analytics solutions use Boolean searches. This technique, which is related to keyword searching,
permits deeper dives into conversations by using conjunctive words such as “and,” “or” and “not” to
expand, define or narrow the fields. This requires users to know exactly what they are looking for and to
manually conduct these searches.
5
frost.com
The keyword, phoneme and Boolean tools suffer from two limiting weaknesses. The first is that they do not
try to determine the meaning of words. The second is they also cannot differentiate multiple expressions of
the same concept having the same conceptual meaning within a certain context. For example, “supervisor”
and “manager” are sometimes synonymous but sometimes are not.
Methods have been developed to obtain meaning from conversations and to make querying more intuitive.
They, too, have limitations. Grammar/lexicon parsing uses grammar and lexicon rules to understand what
was said. It cannot, however, give the right weighting to multiple ideas that co-exist in the same sentence:
“I installed some software on my laptop and it isn’t working.” Parsing would have difficulty determining
whether “it” refers to the software or the laptop.
Natural language analysis approaches human conversations by letting users pose questions rather than keying
in specific words to get answers. It only recognizes precise questions and stored matching answers. This is
a major shortcoming because customers’ conversations cover many different and complex topics, such as
billing and service issues, often even within the same sentence.
Speech analytics tools also have difficulty coping with accents, dialects and words with multiple and different
meanings. Language and usage is rapidly evolving with new words, expressions and meanings entering the
common lexicon.
Another speech analytics solution barrier is the lack of integration with multichannel and screen analytics
solutions. As a result, they can fail to provide an accurate and unbiased view of agent performance and
productivity or even customers’ experiences. Most analytics applications fall short because they are designed
for precise queries, when the critical need is to make sense of what is being said, in real time, whether by
voice or text.
Analytics for speech and other media help contact centers obtain real-time information and insights
from customers’ conversations. Contact centers are now seeking to put this into a broader context that
includes other forms of important data and types of customer experience, including customer surveys
and mobile applications.
Speech analytics solutions are, however, frequently deployed as standalone tools, without direct linkages into
enterprise systems. This can limit inter-departmental information-sharing.The search tools in these applications
may be too complicated for managers to learn and keep refreshed on them if they only have occasional need
for the tools. Frost & Sullivan identified this lack of flow-through in an earlier study, “Redefining Contact
Center Analytics.” In it, we noted that many contact center professionals describe what is missing in their
operations as a “control center” or a conduit of useful information to the executive level.
HP AUTONOMY’S MEANING-BASED APPROACH
A more effective and productive approach to speech analytics is known as meaning-based computing (MBC),
pioneered by HP Autonomy. MBC stresses relevance, not just accuracy, in its methodology. It uses advanced
statistical analysis to understand the intent of the unstructured conversations, and likely topics, not just a literal
readout of what was said. HP Autonomy’s meaning-based computing platform, which powers all of its solutions,
is called HP IDOL (Intelligent Data Operating Layer).
IDOL goes beyond traditional speech analytics solutions in that it examines the entire universe of customer
interactions to obtain the context required to form an understanding of this human-friendly information. It
6
Moving to Next-Generation Speech Analytics: Words are No Longer Enough
incorporates proven speech analytics tools: keyword, phoneme and Boolean searches; grammar/lexicon parsing;
and natural language, but more than that, it pulls from sources including customers’ previous conversations (as
well as live ones), CRM records, IVR, and text-based and Web interactions. It uses automatic hyperlinking that
connects to a range of relevant documents, services or products that are connected to the original text. IDOL
then incorporates the features of standard speech tools while overcoming their limitations by tapping other
sources to give relevance and meaning.
IDOL shows users, like contact center managers, patterns or themes in conversations and interactions, and in
data sets, by applying heuristic or computer learning and sorting methods, such as quantum clustering, to form
them. It understands that there is an X probability that the content in question deals with a specific subject by
studying the preponderance of one pattern over another.
IDOL is built on algorithms rooted in Bayesian Inference and Shannon’s Information Theory. With Bayesian
Inference, the more information given, the more accurate the outcome; also, prior experience should be used
to inform new data. MBC begins with a blank slate and allows incoming data to dictate the model; it mixes new
information with a growing body of older content to refine and retrain the search engine. It “learns” what
customers are saying, which improves relevancy, accuracy and user understanding when querying conversations.
Shannon’s Information Theory provides a framework for software to skim through information to grasp meaning
from it, just as one skims through a news story, as much of what is in there is redundant. This theory enables
IDOL to determine the most important, or informative, concepts within a document, like a recorded call.
IDOL then replicates how humans think and act, in software form. It provides a better analysis by seeking out the
gist of what the call or text is about. It can compare apples to oranges by understanding that they are both fruit.
IDOL does not have to hear the word “statement” to see that a customer is calling back about a billing issue.
It understands the context of the call, recognizes that the caller previously called about a similar issue, and
compares it with other calls and interaction types (e.g., e-mails, chats or social) to identify the root cause of the
call. IDOL then matches its conceptual understanding of the call with potential solutions and suggests the best
to the agent to drive successful resolution.
IDOL takes general queries and automatically pulls up answers with very little drill-down on the user’s part, which
increases timeliness, effectiveness and productivity. It is easily extendable to most managers in an enterprise,
compared with traditional speech tools that require training and experience to use the right search words or
phrases to obtain insights.
The technology is language-independent as it derives its understanding from context; it is not restricted by
specific grammars or vocabularies. It is easily adaptable to multiple languages, accents and dialects, and to rapidly
changing word usages and lexicons.
IDOL powers HP Autonomy’s speech and multichannel analytics solutions. It provides access to more than
1,000 data types, allowing users to assemble structured and unstructured data and connect to more than 500
content repositories. Its service-oriented architecture allows businesses to easily add HP Autonomy solutions
as needs arise.
HP Explore is HP Autonomy’s multichannel analytics product that puts IDOL’s meaning-based capabilities into the
hands of contact center managers. It slices the time and effort they need to obtain a more thorough, actionable
understanding of customer interactions. It is designed to function across multiple channels. It can identify which
customers should be moved to self-service versus agent-assisted service based on certain situations, which
improves customer experience and operations performance.
7
frost.com
HP Explore automates not just speech recognition but also audio-to-text and video, taking a snapshot of what
was said and the emotions and sentiment behind them. It automatically spots trends, both over time and as
they occur, by grouping together related concepts. It can automatically alert staff based on thresholds and other
triggers, such as repeated interactions with a customer on the same issue. Moreover, HP Explore is extendable
to the rest of the enterprise. It is Web-based, intuitive and has a small footprint.
CONCLUSION
Enhancing the customer experience is critical to business success. Firms that clearly understand the intent
of customers and prospects and can leverage these insights across the enterprise will succeed. They can also
cut costs.
Traditional word/phoneme/phrase-based speech analytics solutions, however, are no longer up to the complicated
job. Next-generation, conceptual analytics tools are gaining importance across contact centers.
HP Autonomy’s meaning-based computing methodology that powers its speech analytics solution is attuned to
the complete range of customer interactions. It provides context and relevancy across multiple channels and
permits the gained insights to be shared across the enterprise.
ABOUT HP AUTONOMY
HP Autonomy is a global leader in software that processes unstructured human information, including social
media, email, video, audio, text and web pages. Using HP Autonomy’s information management and analytics
technologies, organizations can extract meaning in real time from data in virtually any format or language,
including structured data. A range of purpose-built market offerings helps organizations drive greater value
through information analytics, unified information access, archiving, eDiscovery, enterprise content management,
data protection and marketing optimization.
Additional information is available at autonomy.com.
8
Auckland Dubai Milan Shenzhen
Bahrain Frankfurt Mumbai Silicon Valley
Bangkok Houston Moscow Singapore
Beijing Iskander Malaysia/Johor Bahru Oxford Sophia Antipolis
Bengaluru Istanbul Paris Sydney
Buenos Aires Jakarta Pune Taipei
Cape Town Kolkata Rockville Centre Tel Aviv
Chennai Kuala Lumpur San Antonio Tokyo
Colombo London São Paulo Toronto
Delhi/NCR Manhattan Seoul Warsaw
Detroit Miami Shanghai
Silicon Valley
331 E. Evelyn Ave., Suite 100
Mountain View, CA 94041
Tel 650.475.4500
Fax 650.475.1570
San Antonio
7550 West Interstate 10,
Suite 400
San Antonio, TX 78229
Tel 210.348.1000
Fax 210.348.1003
London
4 Grosvenor Gardens 877.GoFrost
London SW1W 0DH [email protected]
Tel +44 (0)20 7343 8383 www.frost.com
Fax +44 (0)20 7730 3343
Frost & Sullivan, the Growth Partnership Company, works in collaboration with clients to leverage visionary innovation that
addresses the global challenges and related growth opportunities that will make or break today’s market participants. For more than
50 years, we have been developing growth strategies for the Global 1000, emerging businesses, the public sector and the investment
community. Is your organization prepared for the next profound wave of industry convergence, disruptive technologies, increasing
competitive intensity, Mega Trends, breakthrough best practices, changing customer dynamics and emerging economies?
For information regarding permission, write:
Frost & Sullivan
331 E. Evelyn Ave., Suite 100
Mountain View, CA 94041
