7 Common Myths Of Speech Analytics

Scott Bakken, Founder and President of MainTrax
Scott Bakken, Founder and President of MainTrax

Speech analytics vendors often reel in customers by touting their products as the holy grail of contact center efficiency and profitability. They’re right, but only if you enter the world of speech analytics with clear expectations, precise business goals and healthy doses of patience and perseverance. In other words, it’s OK to be dazzled by a great demo as long as you put off taking the plunge until you do your homework.

Defining exactly how a speech tool will add value to your organization will prepare you to be operational from day one. Then, even if you or your boss have been “victimized” by a good demo, you can roll out a successful deployment that helps you improve agent performance, reduce operational costs, diminish compliance risk, capitalize on selling opportunities, isolate complaint trends, minimize customer churn and generate high-value, actionable business intelligence.

Heightening your awareness and understanding of seven common myths of speech analytics will shorten your learning curve and improve your contact center’s operational effectiveness.


Skillfully operating your speech analytics tool so that it delivers actionable business intelligence can be a daunting, time-consuming process. Misconceptions abound about the logistics and requirements of speech analytics, even in the minds of those who are currently using the technology. While identifying keywords and phrases may be fairly basic, that activity alone will not yield valuable insights without a working knowledge of how content relates to context, how to deftly interpret results, and how to account for various accents and dialects. Optimizing your speech analytics tool requires employees and/or consultants who are adept at both the art and the science of applying sophisticated methodologies to achieve your company’s business objectives.


You’re setting yourself up for success with speech analytics when you control expectations by viewing your speech tool not as a magical device that spins the straw of raw data into gold, but as a filter that pinpoints which portions of which calls pertain to a particular issue. In this way, speech analytics helps you uncover content that may otherwise have gone undiscovered. Used correctly, your speech tool can help you find the proverbial needles in your agent-customer conversational haystacks.


Moving forward with a speech analytics project without first reviewing 200 to 500 randomly selected calls from the appropriate business queue limits your chances of generating actionable business intelligence. You cannot truly know “the voice of the customer” until you listen to actual agent-customer conversations. Yes, listening to hundreds of calls from start to finish is a tedious process, but it’s the best way to identify the key words and phrases that reveal customer intent and establish real-world benchmarks of how often those phrases occur.

Contact center executives new to speech analytics often ask, “Aren’t there existing libraries of phrases ready to plug in? I mean, how many possible ways can there be to say ‘cancel my service’?” Actually, there are more ways than you might expect, and many of these phrases are unique to your customers. Yes, jump-starting the process by brainstorming a list of keywords and phrases can be helpful, but it can only take you so far.

Case in point: One healthcare organization turned to speech analytics to determine which medical conditions policy holders were mentioning. The organization polled their clinicians and compiled a long list of phrases; to identify calls pertaining to pregnancy, they listed medical terms like “doula” and “placenta praevia.” If your first reaction was, “That’s not the way normal people talk,” you’re absolutely right. By listening to recorded calls and learning what phrases were actually being uttered, we were able to double the list of phrases to include expressions such as, “How many months are you” and “Are you showing yet.”

Dissecting historical call recordings also allows you to:

  • Map different call types against predefined business objectives.
    Diagnose meaningful issues by listening not only to what is said, but how it’s said.
  • Create a hierarchy of keyword categories to enable the efficient extracting of business intelligence.
  • Identify the presence of issues that may trip up the speech engine, such as internal transfers, on-hold music or prerecorded messages.
  • Accurately detect and interpret a variety of accents, dialects and colloquial pronunciations.
  • Develop and implement a plan for acting on all the various triggers revealed by the call review process.


Granted, focusing on what agents are telling customers can produce immediate and measureable benefits. Establishing performance benchmarks, then coaching or training agents and tracking their behavior can reveal:

  • The quality of the training effort.
  • The degree of script adherence.
  • The willingness and ability of agents to improve performance.
  • What each agent is doing particularly well or poorly.
  • Which underperforming agents need to be weeded out.
  • Best practices that can be incorporated into agent coaching and training, both systemically and for specific situations.
  • Opportunities for capitalizing on selling opportunities.
  • Where operational enhancements are needed.
  • The financial impact of agent performance.

That said, monitoring the customer side of conversations can also provide valuable insights. After all, it is caller intent that frames every call; identifying the specific desires of the customer will inevitably lead to more positive outcomes. Gathering and analyzing customer interaction data can:

  • Provide a deeper understanding of the “voice of the customer.”
  • Identify reasons behind customer complaints.
  • Reveal quality deficiencies in products, services, internal processes and agentcustomer conversations.
  • Help agents better recognize both opportunities and threats.
  • Enable managers to determine if agents are responding appropriately to specific triggers.
  • Develop strategies for satisfying unhappy customers.
  • Equip agents with more effective scripts.
  • Reduce customer churn.
  • Improve customer service.
  • Increase brand loyalty.
  • Prevent minor issues from developing into major issues.
  • Stay current with customer trends.


The quality of decisions made by people at all levels of your organization will likely be influenced by the quality of the data produced by your speech tool. Ensuring the quality of your data—and thus the quality of organizational decision making—requires you to optimize the accuracy detection of your speech tool. Doing so requires numerous iterations of testing and tuning to assure the highest levels of accuracy detection. This exacting work requires auditing search results and adjusting confidence levels repeatedly until you reach the sweet spot of accuracy that’s thoroughly aligned with your predetermined business objectives.

The importance of adjusting confidence levels cannot be overstated. Essentially, a confidence threshold is a setting (most effective at the phrase level) that allows the user to filter reported hits based on how certain the engine is that the hit is accurate.

Low-confidence thresholds direct the speech engine to be less discerning, increasing the odds that high-value phrases (such as, “I will sue you”) will be detected. Lower thresholds are typically set when an organization is seeking to capture high-value information and has the resources to review a potentially large number of hits. The trade-off: more false positives.

High-confidence thresholds direct the speech engine to be more discerning, increasing the odds of detection accuracy. Higher thresholds are typically set when the user wants to collect a sampling of calls with a high likelihood of accurate hits, rather than trying to catch every instance. The trade-off: more false negatives.


There are two ways to collect and measure your data: the easy way and the right way. A lack of rigor in your collection and analysis of data will almost certainly lead to unreliable results. Here are three common behaviors to avoid:

An unwillingness to dive deep. Building an accurate database requires that you verify both the content and context of detected phrases. For example, if your speech engine registers a hit for “awful service,” it is useful to know whether the customer was referring to the service delivered by your company or by your competitor.

A fixation on one area of focus. Myopic business analysts may focus on one area of importance while ignoring less obvious areas of interest. For example, focusing on underperforming agents while overlooking agents who provide high-quality “champagne” moments during tough conversations can distort perceptions of overall agent performance and lead to missed opportunities for enhancing the quality and depth of agent training.

A tendency to disregard all false positives. Analyzing hits that are not 100% accurate can still provide opportunities for insights and process improvements. For example, if you are searching for non-compliant phrases such as “guarantee you’ll get financial aid,” and the speech engine fires on “guarantee you’ll get aid,” you now have a new alternative phrase to import into your search criteria.


When its new low-cost fixed indemnity product began generating thousands of calls from dissatisfied customers, a prominent healthcare insurance provider determined that 12.3% of policy holders who called in a given month went on to terminate their coverage within two months.

The insurer’s attempts to use speech analytics to predict customer churn by identifying policy holders “at risk” generated mixed results because the vast majority of calls contained “cancel”-related language. Engaging MainTrax to expand the speech library beyond the most obvious “cancel”-related phrases nearly doubled the company’s predictive power of customer churn. Assisted by their speech analytics correlation tools, MainTrax analysts further determined that a specific combination of phrases was an even stronger predictor of behavior. Fully 38% of callers using a specific combination of phrases churned within two months of their call.

Armed with this insight, the insurer could now identify key segments of its customer base that were most likely to churn, and could employ remarketing and call-handling strategies designed to retain them. Did they put these metrics to work? They did not. They had not bothered developing an actionable program to save these customers. After investing in a sophisticated speech tool and three months’ worth of professional services, the insurer’s managers allowed themselves to be distracted by other priorities and brushed aside data that could have saved them countless customers and many thousands of dollars. The moral of the story? Have an action plan in place!

Scott Bakken, Founder and President of MainTrax, is a highly respected independent voice in the speech analytics industry. His company provides professional services that help end users use their existing speech analytics tools to deliver actionable business intelligence. Bakken was named an Ernst & Young Entrepreneur of the Year finalist.

– Reprinted with permission from Contact Center Pipeline,

Related Articles

Back to top button