Why Natural Language IVR Is A Nightmare for Customers

The rise of automation in everyday life is often bemoaned for its displacement of the human touch. This is especially true when a technology is introduced before it can provide the same or better level of service than what it’s replacing—such as a low-level chatbot meant to fill the role of a real-life representative. 

Furthermore, even though many companies are able to engage with their customers via multichannel and omnichannel communication methods, 76% of customers still prefer to contact customer service centers via phone. 

Nevertheless, many companies are implementing interactive voice response (IVR) systems into their contact centers and leveraging natural language processing (NLP) to allow customers to solve their own problems and route their own calls by speaking naturally. Of course, that’s easier said than done—because if an IVR is implemented poorly, its predetermined prompts and menus can seem cold, impersonal, and unhelpful.

As a result, although the intention of adding NLP features to an IVR is to improve the overall customer experience, one false move during its implementation can do the opposite. 

Reasons Natural Language IVR Can Be A Nightmare For Customers

Like any technology that attempts to mimic humans, generative and conversational AI models are trained via millions of real-life examples. Voice assistants such as Siri and Alexa, for instance, are able to offer personalization because they are backed by a vast corpus of data.

In the case of a natural language IVR, its success depends on the accurate interpretation of caller requests and the application of database knowledge to make good routing decisions. Failure to do so would result in the system routing callers to a live agent in the wrong department or providing the agent with the wrong information about the call—ultimately causing the customer to waste time repeating themselves and thus defeating the purpose of the entire system. 

Here are three core concepts you have to keep in mind when incorporating NLP into your IVR system:

The system will not understand every customer

Although AI-powered speech recognition has come a long way in its ability to convert speech into text that it can comprehend, there is not a one-size-fits-all solution. Yes, most individuals have first-hand experience with embedded natural language processing in apps and other voice-activated parts of life, but spoken commands still tend to function most reliably on individual user devices that can adapt to certain individuals. 

Open-ended prompts can be too unpredictable for IVRs

Allowing customers to respond in their own words can lead to significant challenges, mostly because callers are not always prepared to react to an open-ended prompt with a clear and concise response. Instead, many callers will end up giving meandering, roundabout explanations for what they need or what’s going on—and this can send automated systems in all kinds of directions. 

In other words, an automated system that uses natural language processing will have difficulty handling long-winded explanations because they can cause the machine to consider multiple routing options and ultimately offer the wrong solution as a result. 

Meanwhile, despite their advancements, natural language processing systems can also struggle with the diverse range of dialects, regional accents, and mispronunciations that customers may use, potentially leading to further inaccuracies. Similarly, other potential flashpoints of allowing free-flowing conversations to occur include the challenges of word choice like industry jargon and slang.

Altogether, if any of these elements cause the IVR to become confused by conflicting information, the potential inefficiencies of its subsequent performance can be frustrating for both customers and the business itself. 

Open-ended prompts can be too confusing for callers

Conversely, open-ended prompts like, “How can I help you?” can also confuse callers—especially those determined to respond with exactly what the system needs to send them in the right direction. 

Many customers may also lack the relevant vocabulary or precise product knowledge to produce adequate, on-the-spot responses without any suggestions or nudges from someone else. As a result, communication problems can quickly escalate, with many users becoming frustrated after a few failed attempts. 

How to Increase the Chances Of Natural Language IVR Being Successful

While the challenges of adding NLP technology to your IVR menu system may seem daunting, they aren’t insurmountable. There are a few actions you can take to ensure that yours is useful enough to serve your customers well. 

Get the right person to train your IVR system

Fine-tuning your natural language IVR for optimum performance typically requires a strong understanding of the underlying AI models and architecture. Hence, you may need the help of a developer or prompt engineer to train and/or design everything to your benefit.

Hiring NLP developers

Most chatbots are powered by large language models (LLMs) like OpenAI’s GPT-4. LLMs are a crucial subset of NLPs and can perform a variety of natural language processing tasks using transformer models. 

However, most businesses simply use them “out-of-the-box,” which isn’t adequate for many of the industries that require customized LLM algorithms. While these chatbots have been revolutionary, they still have the tendency to “hallucinate” and produce erroneous information and feedback.

Therefore, you may need to hire an NLP developer or software engineering team to create tailored solutions for your unique needs—especially if you’re in fields such as finance, manufacturing, healthcare, automotive, and logistics. While transformer models translate text and speech in real time, developers can make them focus on the most relevant segments of language to produce better results.

Hiring prompt engineers

The LLMs in the public domain come preloaded with massive amounts of information and training. However, they tend to lack a targeted understanding of a given business’s needs and the intentions of its callers. 

Thus, for your system to perform at its best, the AI-powered language interpreter must be primed for your organization or industry’s knowledge and jargon, used in the proper context. Fortunately, you can access LLMs through an application program interface (API). This makes it possible for someone to make adjustments and create parameters that will dictate or guide how an LLM responds.  

This is called prompt engineering, which entails sending questions and requests to a language model so that it learns how to provide the output your customers want. By hiring a prompt engineer, then, you can enhance the quality of your natural language IVR’s neural network architecture. 

Reconsider if you really need a natural language IVR system

While people prefer to speak conversationally, you must ensure that this use case suits your business structure. In other words, NLP technology shouldn’t just be another shiny object in your toolbox—it should serve a purpose. 

For example, if your organization can get by with a traditional speech IVR that handles simple “yes or no” questions, then you can save a lot of time, money, and other resources by holding off on implementing a natural language IVR system. 

Here are several indications that you need or could use one:

• You have an adequately expansive organizational structure. If you have a large number of departments and destinations to which you need to direct calls, consider adopting a natural language IVR. 
• You have a high volume of varied calls. If your contact center or customer service department consistently experiences high call volumes, you are likely a prime candidate.
• You have the financial capacity to pull it off. Implementing and maintaining a natural language IVR is more expensive than traditional, simple dialogue-driven IVR systems, so you should only consider it if you have the budget for initial setup and ongoing maintenance costs.

Have some sort of education for customers when you roll it out

Most customers are familiar with (and may still expect) old-school IVR systems, so it’s not a great idea to thrust a new system upon them without warning.

While some businesses decide to create tutorials and provide answers in FAQ pages, the best approach may be to offer direction within your IVR system itself. This way, your customers will receive real-time guidance when they encounter issues. To provide this kind of help, you can set up your system to teach the caller how to proceed—perhaps by having it say something to the effect of, “You can now say things like, ‘I need to reschedule an appointment,’ or, ‘Pay my account balance.’”

Creating an email campaign can also offer your customers a streamlined and structured learning process for your natural language IVR. This can give them a way to learn at their own pace, discuss the system, ask questions, and offer feedback. 

Lastly, remember that there may be some growing pains as your customers adjust to the new system—even when you provide great educational resources.