You’ve heard the buzz about AI, and how “smart” artificial intelligence is getting. Certainly, today’s AI is mind-bogglingly better than the AI of 20 years ago.
But if you’ve ever had a conversation with a voice AI, over the phone or through chat, you might not be so impressed. If you’ve had a bad conversation with a “robot,” that means it hasn’t been designed with humans in mind.
Any AI system still needs to be coached and tuned to be a great conversation partner. In other words, AI isn’t smart unless you make it so.
The Role of Conversation Design in Voice AI Training an AI to talk like a human requires serious effort. That’s where the concept of conversation design comes into play.
Conversation design is the craft of coaxing a voice AI to play nicely with people, on human terms. Conversation designers create scripts and flows to make interactions natural, intuitive, and efficient.
Think of it as writing a script for a play where the AI is one of the actors, but instead of memorizing lines, it needs to understand and respond to a wide range of human inputs in real-time.
Good conversation design requires understanding human linguistics, psychology, context, culture, and even a bit of humor to make interactions feel genuine. Conversation designers know how to train AI, how to craft tone and word choice, and how to create natural pacing and a sense of “flow.” These considerations ensure the AI responds like a good conversation partner.
When done well, it can make a voice AI feel more like a helpful guide and less like an answering machine. But when it’s off, users can quickly tell. They’ll often feel frustrated, repeating themselves, or like they're talking to a wall.
How To Tell if Your Voice AI Wasn't Designed With Humans in Mind Ultimately, conversation design aims to ensure users feel like they’re talking to a smart assistant, not a robot. Let’s examine the following signs to tell if a voice AI isn’t the sharpest tool in the shed along with some helpful examples.
1. It answers a question you didn’t ask There’s nothing more bewildering than asking a simple question and getting an irrelevant answer. It’s like asking a friend what time the movie starts, and they reply with the color of their shoes.
Let’s say your favorite restaurant has voice AI phone answering, and you decide to call to ask about the kitchen staff's exciting specials this week.
You : “What are your specials?” Bad robot : “Yes, we are currently open!” The AI clearly isn’t listening. Any AI’s first duty is to understand basic questions and provide the right answers. When a voice AI lacks proper training or context awareness, that’s a recipe to break down trust quickly.
At Slang.ai, our team has carefully trained and rigorously optimized our AI voice concierge for years to understand what callers say, the first time they say it. Slang concierges also give answers that are correct and specific to your restaurant. The Slang.ai dashboard allows you to configure and update your specials, and all your other restaurant information, as often as you need.
2. It misses important details A smart voice AI should be able to pick up on key details and respond accordingly. Say you have a big party of 20 people and want to make a reservation following a social event. Here’s how the call might go with a not-so-smart voice AI:
You : “Hi, can I make a reservation for 20 people?” Bad robot : “Sure, let me text you a link to our website!” Not bad, but there’s just one hiccup: In this scenario, the restaurant only accepts reservations for parties of up to eight people. The bot gave a generic “reservations” response but missed the detail that this caller was trying to book for a large party. It sent the wrong info and wasted the caller’s time.
Missing details like these can make interactions feel robotic and impersonal. A good voice AI is designed to understand context and nuances, ensuring the information it provides is relevant and useful.
Let’s try that again, only with a Slang.ai voice concierge instead:
You : Hi, can I make a reservation for 20 people? Slang.ai : Actually, we only take reservations for up to 8 people. For a party of that size, we can provide options for renting one of our private dining spaces. We’d love to talk with you about this, so let me transfer you to a member of our team. The Slang.ai voice concierge’s response takes into account the caller’s specific conditions and responds in context, using what it knows about a given concept’s policies.
3. It can only recognize topics, not nuanced questions If designed well, voice AIs can do more than recognize more than broad topics. A great voice AI can grasp nuance. Understanding conversational subtleties is crucial to avoid confusing a caller.
Let’s give another call to the inferior voice AI. This time, you want to bring a bottle of wine you’re excited to try after a trip to Sonoma wine country:
You : “Can I bring a bottle of wine?” Bad robot : “Yes, we have a full bar!” You : “No, I mean, do you have corkage fees? Can I bring in a bottle of my own?” Bad robot : “Yes, we have a full bar and an extensive wine menu!” It’s clear the AI only recognized the topic of alcohol and had a generic answer on repeat, completely missing the caller's actual query about outside wine and corkage policies.
When an AI fails to give a precise answer, it quickly frustrates users who have to repeat or rephrase their questions in hopes of getting a relevant answer. It demonstrates that this voice AI lacks sophistication in its design.
Let’s see how a similar situation would play out with Slang.ai:
You : Can I bring a bottle of wine? Slang.ai : You're welcome to bring your own wine to enjoy. Our corkage fee is $15 per bottle. Slang.ai’s concierges responds with relevant, detailed information about a restaurant’s exact policies, so that callers’ questions are answered right on the first try.
4. It’s socially awkward Interacting with a voice AI should feel natural, not uncomfortable. But if a voice AI will misses the mark, its conversations can be just plain awkward.
You’re calling again to hear about specials, but this time around, you get a strange response:
You : Do you have any specials today? Bad robot : I like you, too. Yikes. This AI only had one context for the word “special” and took it as a compliment. This kind of response is an attempt to be cute or friendly. But instead, it’s cringe-inducing.
Sometimes, developers try to infuse personality into voice AIs to make up for their lack of sophistication or accuracy. While a bit of personality can make interactions more engaging, it can also backfire (spectacularly in some cases) when the AI doesn't understand the context.
Once again, here’s how the conversation might play out with a Slang.ai voice concierge:
You : Do you have any specials today? Slang.ai : Yes, we have happy hour daily from 5 to 6, with $1 oysters and $5 glasses of bubbly. We serve entree and soup specials every day, too. Would you like me to text you a link with our specials? You : Sure. Slang.ai : Anything else I can help with? You : No, thanks. Slang.ai : Thanks for taking the time to call us! We look forward to seeing you soon. Slang.ai’s concierge knows that callers want good answers to their questions, not irrelevant banter. And even though it's AI, it has manners, too.
Building Trust Through Smarter AI Bottom line: For restaurant operators exploring the benefits of voice AI answering, the quality of design matters. When an AI can handle caller queries smoothly and accurately, it also builds trust and loyalty.
To recap, these are a few things to keep in mind to determine if your voice AI is “smart”:
It understands the context of a conversation to provide relevant and specific responses. It can handle subtle differences in questions, ensuring accurate and useful answers. It engages in a conversational and professional way, avoiding awkward or robotic responses. Slang.ai helps countless restaurants automate phone calls to boost staff productivity and capture more revenue. Book a quick 10-minute demo today to see how it works.
