Great Expectations: Self Service Automation and IVR

Susan Hura

Every call center manager knows that there are dozens of ways to measure various aspects of call center performance, including call center metrics for analyzing IVR self-service automation. Call resolution, call containment and average handle time are but a few of the call center measures of IVR performance, and for speech-enabled IVRs we can add recognition statistics, in-grammar and out-of-grammar rates, and more.

All of these call center measures leave out one vital factor for the success of speech self-service: The role of expectation. I’m referring to the expectations that callers have when they interact with a speech-enabled IVR application, based on previous experience with IVR systems, speech recognition and call center technology in general. Call center expectations are by definition soft and are expressed indirectly as attitudes, preferences and assumptions—all things that are notoriously hard to measure in the call center. Yet expectations about IVR and call center speech technology have a huge impact on how successful a call center project will be.

Too High and Too Low in the Call Center

Callers’ expectations are one of the most powerful factors in determining the way the customer approaches an interaction with a speech system. The complication is that there are widely varying levels of customer experience with call center speech technology among the general public today. Some callers have almost no exposure to speech technology so they don’t have any strong expectations of how it should work. These callers don’t know exactly what to say or how they should say it to a speech system. They’re trying to figure out the "rules of call center engagement" as they go along. Other callers have flawed, but overly positive, expectations for speech technology built on the Star Trek call center model, in which speech recognition is always available and works flawlessly.

Both of these groups of callers, tend to over estimate what speech technology can do and come into interactions with speech-enabled IVRs with unrealistically high expectations. The danger is that such callers tend to speak long, unconstrained utterances, just as you might talk to a person, which is likely to result in caller recognition failures. Most speech recognition IVRs cannot handle unconstrained conversational speech, even when they are built to handle "natural language." Thus speech systems fail to meet caller expectations, leaving them disappointed and often confused about what they did wrong.

Equally challenging for the call center is the caller who has low expectations of speech technology. These people have tried to use speech technology in the past and failed, often repeatedly, and are frustrated by the call center technology’s inability to understand them. Callers start out with good intentions—they tried in the past to give good responses to speech systems, and do not have a context that helps them understand why the misrecognitions occurred. It’s especially troubling when speech systems fail to recognize simple responses, like "yes" or "no," because callers don’t realize that this kind of response is more challenging for speech systems than for humans. Not surprisingly, callers who have had this sort of experience have very low expectations, which can unfortunately make the situation worse.

These callers treat the IVR like a person who is having trouble understanding, and speak in an over-enunciated and painfully slow way, which actually makes speech recognition more likely to fail. Speech systems are built to recognize utterances spoken at an average speed, with normal enunciation; any speech that deviates from normal (including speech callers believe is easier to understand) is actually more likely to be misrecognized. Thus very low caller expectations can also result in a poor caller experience.

Correcting the Expectations Mismatch in the Call Center

Some expectations are too high, some are too low, and all of them end with disappointed callers who are more likely to reject IVR self-service and opt out to live call center representatives. But there is a way to a way to avoid the expectations mismatch and deliver satisfying experiences to callers.

  • Do your best to understand the expectations of your customers before designing the speech IVR interaction. Conduct focus groups to collect direct input from customers on the call center self-service and call center speech technology. Think about the self-service options available to them today and their comfort-level with the call center technology. This will help define how much guidance callers will need to successfully interact with a speech IVR.
  • Remember that callers rely on the unwritten rules of spoken conversation when interacting with your IVR. Whether they expect too much or too little, callers fall back on communication strategies that work with people when interacting with IVRs. We can capitalize on this by crafting prompts that constrain callers’ responses by capturing the style and structure of human-to-human spoken language. The goal is not to write prompts phrased just like a person would say them, but rather to use wording that is intuitive to callers, so that they can simply respond without over-analyzing the interaction. This frees callers to focus on accomplishing their tasks rather putting them in the very vulnerable position of trying to figure out the limits of the call center technology.
  • It is also important for organizations to use speech technology wisely to automate tasks that can be completed in most situations. Avoid using call center speech technology wisely to automate tasks that can be completed successfully in most situations. Avoid using call center speech technology for its own sake, and instead look for opportunities in which speech provides distinct value to customers and the organization.

We are by no means doomed by expectations. By understanding the expectations of the caller and the call center, we can create speech IVRs that meet business goals and deliver a positive caller experience.

First published on Call Center IQ.