Why the Smartest Voice AI Knows When to Send a Text

By Yaniv Masjedi, Chief Marketing Officer at Nextiva

It may seem counterintuitive, but it’s true. As businesses take on AI receptionists to answer phone calls, one of their most valuable capabilities involves written communication.

A new study shows why. The reason involves human psychology, emotions, and the complex challenges of resolving customer service failures.

AI voice agents are packed with unprecedented potential. They can capture missed calls, qualify leads, provide answers, relieve overburdened staff, and deliver a phenomenal customer experience (CX). Speech is our richest, most natural medium, so bringing human-like conversation to automated phone support can go a long way in boosting customer satisfaction.

But that doesn’t mean it’s always best. The moment something goes wrong in customer service, a written interaction can lead to better results. “In a real-world service failure context, customers who interacted with a chatbot (vs a voicebot) gave significantly higher satisfaction ratings,” researchers explain in a new study published by the European Journal of Marketing.

There's a crucial lesson in this. To find the best AI call handler for enterprises and small businesses, look for agentic voice tools that also send live texts -- and that know when to do so.

‘Affective Fading’

A typical caller may want to gather information, make an appointment, or schedule a service. Increasingly, customers appreciate doing these things through an AI receptionist that understands them and guides them to a quick resolution.

But when they’re upset about a mistake or something not working, the emotional balance changes. "In service failures, the increased emotional engagement associated with voicebot interactions may amplify distress, whereas chatbot interactions facilitate perceived relief,” the authors of this new study explain.

Spoken dialogue is rapid and synchronous. Humans typically speak at around 160 words per minute, compared to typing at roughly 50 to 55 words per minute, the study says. When a customer is frustrated, a spoken exchange usually causes them to remain that way. And the more steps are needed, the more upset the person gets.

Text, by contrast, introduces two important mechanisms that help to calm things. The first is “affective fading,” also known as “fading affect bias.” Negative feelings dissipate more quickly than positive feelings do. When an interaction is slowed, people have more of an opportunity to let those emotions subside.

“This extra time matters because negative emotions such as anger tend to decrease almost immediately after onset and can even fade naturally within about 90 seconds if not retriggered,” the study says.

The second mechanism is “attentional deployment.” It’s a term researchers use for the process of “shifting attentional focus within an emotional scene in order to modulate emotional experience.”

"Redirecting cognitive resources to composing a message can temporarily shift focus away from the emotional trigger, reducing arousal,” the study says. Written language generally “contains fewer emotionally charged words” than speech. “Speaking about choices fosters an emotion focus, whereas writing fosters a reason focus.”

So seamlessly offering a texting option mid-call is an essential part of designing an AI voice agent. In creating XBert, our AI receptionist, my team at Nextiva built this bridge directly into its architecture.

Urgency Erases the Advantage

Interestingly, this relief advantage is strongest when issues are routine or low-to-moderate in severity -- a broken setting, an account glitch, or an intermittent product concern. In emergencies, where immediate intervention is paramount, voice and text perform equally well.

"The urgency of problem resolution takes precedence over emotion regulation, as customers shift their focus toward obtaining immediate solutions rather than managing their emotions,” the study says. “In high-severity failures, both chatbots and voicebots present themselves as equally efficient problem-solving devices, and then no differences are expected."

While these findings serve as a guide, they don’t apply equally to everyone. People have unique preferences and emotional experiences. In any given scenario, some people will do better by speech, while others will benefit from a switch to texting. And the point at which that switch helps will vary as well.

The best AI call handler responds to each person in the way that's right for them. That's another reason unified customer experience management is so important. When a customer reaches out through any channel, a brand should know who that person is, and adjust the interaction based on everything it knows about them.

The future of frontline AI isn't a choice between voice-first or chat-first. It’s dynamically multimodal. In seeking out a solution, go for one that has the intelligence to bridge modalities in real time.