Understanding Speech Analytics & Natural Language Processing

As technology is slowly improving to mimic human-like interaction, many businesses are looking to provide such experiences to their customers—the kind that comes with systems, like Amtrak Julie, that can understand the intent of the caller. Suddenly natural language processing (NLP) has become a buzzword much like “cloud computing.” How do you know if NLP is the right technology for you? That requires a longer discussion.

To understand NLP, it’s important to understand how automatic speech recognition (ASR) works. The way I generally explain it is that it’s very similar to your understanding of foreign languages. When we learn a foreign language, we build a list of words we can understand. These words or phrases in the ASR system are called grammars. If you speak anything outside that list, ASR system doesn’t understand what you are saying. Based on some research done in the IVR industry, the average rate at which callers say things that are not on the list of pre-defined IVR grammars varies from 15–30%. Assuming 95% accuracy in understanding grammar-driven words, our current IVRs can only understand around 70 percent of total caller interactions without using speech analytics.

One way to reduce this out-of-list grammar problem is to introduce more comprehensive lists of words and phrases but this process is expensive and time consuming and, in the end, you can do only so much. A true natural language–based solution solves these problems as it works by building a statistical model using machine learning with the goal of understanding the caller’s intent rather than matching a set of words. Let’s take an example: you are interacting with an IVR and you have to enter a four-digit PIN; you say something like, “One, two, five, four—not five, four—three, four.” It would be pretty hard for an IVR to understand but NLP would do a great job. A successful natural language deployment means fewer agent transfers, more containment, and higher customer satisfaction.

So what is stopping businesses from completely switching to NLP technology? Well, it takes a huge amount of time (often years) and resources to train the machine. Businesses find it very hard to justify the ROI, and often results vary widely from one system to another. Although I do not see IVR replaced by NLP technology in the near future, I do believe it has the potential of becoming a force to be reckoned with.

What do you think? Check out our webinar OnDemand, What are 99% of Your Customers Hiding from You? Find Out with Cloud-Based Speech Analytics to learn more about listening and analyzing your customer contact center interactions.