Three Things They Won’t Tell You About Speech Analytics

Me: “Hey Siri? I’m bleeding badly. Call me an ambulance!”
Siri: “Okay. From now on I’ll call you ‘Anambulance.'”

Most all of us have humorous, and often aggravating, anecdotes about trying to communicate with Siri, Alexa, or any of the other voice prompted technology apps available to us. I am quite regularly thankful that no one is around to hear the tirades I scream to the disembodied, robotic female voice of my car’s voice prompt technology. It amazes me, then, to know that businesses spend huge amounts of money on speech analytic technology as a way to replace their Quality Assessment (QA) programs.

Let me start with full disclosure. Our company, C Wenger Group, has spent a quarter century monitoring and analyzing our clients’ phone calls as a third-party QA provider. Sometimes our clients hire us to be their QA team, and other times they hire us to provide periodic audits and reality checks to their internal efforts. Over the past few years we have learned that speech analytic technology has become a competitor to our service. I can quickly name two clients who have dismissed our services in favor of speech analytic software.

The promise of speech analytics is in the ability to monitor vast quantities of phone calls. Most human QA efforts, by comparison, utilize relatively small random statistical samples. Our data over the years reveals that our team can quite capably provide an accurate reflection of service performance with relatively few calls. I remember calling one skeptical client after our initial month listening to a minimal sample of calls for sales compliance. I gave him the names of three sales people whom our call analysis identified as problems. He laughed and told me that all three had been fired the previous day agreeing that our sample and analysis was, indeed, sufficient.

Nevertheless, the idea of being able to catch the needle in the haystack has certain appeal. Random samples don’t capture every instance of irate customers, lying representatives, and forbidden behaviors. That’s where tech companies and their (big ticket) speech analytic software promise nervous executives a peaceful night sleep knowing that every phone company can be monitored by computers and flag problem calls when they occur.

Just like Siri flawlessly hears my every spoken request and never fails to provides me with exactly the answer I was looking for.

I have followed up and spoken to both clients who dismissed our company’s services in favor of speech analytics. In one case, my contact admitted that they abandoned the technology after a year of unsuccessfully investing significant resources (money and man hours) trying to get it to provide meaningful results or value. In the other case my client contact admitted that the technology never worked, but that his company continued to play the political game of pretending it was working because they didn’t want to admit that they’d wasted so much money on something that wasn’t working. I have also spoken to representatives of other companies with similar words of warning. As with most technologies, it’s important to know what you are, and aren’t, getting before you sign on the dotted line.

My conversations with those who’ve employed speech analytics reveal three key things that should be considered when considering it as a technology investment.

It’s going to require a lot more work to set it up, monitor, tweak, and successfully utilize it than you think. At one industry conference I attended a forum of companies were using speech analytics. I found it fascinating that all of the panelists admitted that the technology required far more time and energy than they anticipated when they purchased it. One company on the panel admitted that they hired five full time employees just to make the technology work and to keep it working. Many people don’t realize that you have teach the speech analytic software what to listen for, what to flag, and what to report. Then you have to continually refine it so that it’s catching the things you want it to catch and ignoring the things you don’t.

In many cases, this process is not intuitive. It’s more akin to computer programming. Operations associates who thought they were going to save themselves time having to personally analyze phone calls find themselves spending even more time mired in IT issues related to the technology.

The technology is going to give you a lot of false-positives. I love that I can say “Hey, Siri” and my iPhone will come to life and ask what I need. I have also been annoyed and embarrassed at the number of times in normal conversation or business meetings that I say something that my iPhone mistakenly hears as “Hey, Siri” only to wake-up, interrupt my conversation, and ask what I want. In similar fashion, you can expect that for every instance of speech analytic software catching the right thing, it is going make at least as many, if not more, mistakes.

One of my former clients told me that the speech analytic software they employed never worked as well as advertised. “Every time it flagged a call for us to listen to there was nothing wrong with the call,” he admitted. They quickly stopped listening to any of the calls flagged by speech analytics because they soon saw it as the proverbial child crying “Wolf!”

Speech analytics can monitor volume, pitch, and words that are said, but cannot intelligently analyze content across calls. Our team recently monitored a randomly sampled set of phone calls for a customer service team. The CSRs were articulate and professional in the words they used and the tone with which they communicated with callers. Across the calls, however, we quickly noted a pattern:

  • “Let me get you to the person who handles your account.”
  • “I don’t handle your area.”
  • “You’ll need to speak with….”

In various ways, using different words, many of the CSRs were refusing to help callers. They would immediately “pawn them off” (one customer’s words) to other CSRs or dumping callers into voice mail. In some cases we heard veteran employees claim that they didn’t know how to do the most basic of customer service functions in an effort to avoid helping callers.

Our team quickly recognized that our client was struggling with a culture on their call floor in which CSRs tried to avoid helping callers (in the most professional sounding way). Customers were being dumped into voice-mail and transferred unnecessarily as CSRs played an internal game of “that’s not my customer, that’s your customer.” We addressed it with our client, citing examples. They quickly moved to address the issue and are already making significant progress toward changing behavior on the call floor.

I tried to imagine how I would tell a speech analytics program to catch such an occurrence. The ways that CSRs communicated that they couldn’t help were as varied as the CSRs themselves and their own communication styles. Customers frustration never escalated to shouting or profanity. It was all very subtle, and required experienced analysts making connections across multiple calls to recognize the pattern of behavior. Speech analytics could never do that.

Like most technologies, speech analytics has its place and its purpose. For those companies who have the resources to successfully employ it, speech analytics can analyze vast quantities of interactions and flag, with relative degrees of accuracy, when certain words are spoken or certain levels of emotion are expressed. Those considering this technology as a replacement for a thorough and well structured QA program should understand, however, that the technology has requirements and drawbacks that the technology salesperson will be quick to ignore or minimize.

  2 comments for “Three Things They Won’t Tell You About Speech Analytics

  1. March 7, 2017 at 12:17 am

    Please, do not compare SA to Siri. It’s like saying: “Formula 1 cannot be real. I had a Ford Fiesta, and it was nowhere near that fast.”
    There are many kinds of Speech Analytics.

    The engine behind Siri is designed to understand short commands. The NLP-based engines behind call center conversations analytics differentiate between speakers, understand context, measure sentiment and emotion, and correlate findings with metadata.

    Sure, they need to be set up, just like a QA agent needs to be trained. However, it takes nowhere as long as it would 10, 5 or even 3 years ago. SA of today is much more customer-friendly, most solutions come with a dashboard-for-dummies type interface.

    QA has its advantages like higher accuracy or… nope, that’s the only one I can think of. Every aspect of QA can be recreated with SA models. While spotting three lazy agents is an accomplishment, it’s a drop in the ocean.

    QA has to embrace Speech Analytics. Your experience is still valid, and your analytical intelligence remains relevant, but it’s time to drop random manual call scoring and use tools that will allow you to do the same thing on a much bigger scale and at a fraction of the cost.

Leave a Reply

Fill in your details below or click an icon to log in:

WordPress.com Logo

You are commenting using your WordPress.com account. Log Out / Change )

Twitter picture

You are commenting using your Twitter account. Log Out / Change )

Facebook photo

You are commenting using your Facebook account. Log Out / Change )

Google+ photo

You are commenting using your Google+ account. Log Out / Change )

Connecting to %s

%d bloggers like this: