9 Tips to Design Automated Conversations That Actually Work

By Harry Chang
November 11, 2022

Contact Center Automation is a sophisticated customer service solution powered by a combination of advanced AI, Natural Language Understanding, and Machine Learning.

But for all its technological firepower, Contact Center Automation is just as much of an art as it is a science. A thoughtful approach to automation takes into account factors like customer behaviors, industry standards, and brand guidelines.

Replicant’s Thinking Machine™️ places an equal emphasis on conversation design as it does Conversational AI. It’s built for not only the “happy path” conversations but the edge cases that can lead to frustrating experiences. The result: frictionless experiences that don’t simply deflect customers but fully resolve requests to drive operational efficiency.

Here are a few ways Replicant’s gets the most out of Contact Center Automation by creating conversations that customers love.

9 Game-Changing Conversation Design Tips:

1. Answer Immediately & Avoid Personification There are two things that can frustrate a customer in the first few seconds of a call. The first is long wait times. The second is the unnecessary personification of a machine. Replicant avoids both by answering every call immediately and being upfront with customers by stating “I’m a Thinking Machine™️ on a recorded line,” and not masking automation with human names or pleasantries. Both factors help gain customers’ trust without wasting their time.

2. Think Like a Machine & What’s In It For Me? Decades of clunky call trees have conditioned customers to be skeptical of automation. That’s why Replicant designs each Thinking Machine™️ to demonstrate value to customers as soon as possible. For example, if a customer attempts to zero-out to speak to an agent, the Thinking Machine™️ can make a data dip into a call center system to pull a live wait time. From there, it can tell the customer their request can be resolved by the Thinking Machine™️ before they’d get through to an agent.

3. Natural Language Intents & Colloquial Language The last ten years have seen massive technological advancements that allow for better automatic speech recognition than ever. The Thinking Machine™️ is designed to match unstructured speech to intents where older solutions get lost in translation. It can understand that a customer saying “I’ve been squinting at the computer screen and getting headaches” suggests they are seeking an eye exam, or that a customer who says they’re in the “Big Apple” is talking about New York.

4. Probabilistic Funnels & Knowledge Based Matching Most contact centers have highly probable call paths. When this is the case, it makes sense to build them directly into the Thinking Machine™. For example, if 70% of customers call to ask about contact lenses, it shave valuable seconds off of handle times to lead with “Are you looking for contact lenses?” Conversely, if the Thinking Machine™️ sees that a customer recently inquired about glasses, it can lead with “are you calling about eyeglasses?” to begin making progress faster.

5. Conversation Repair Sometimes customers change their minds. A caller scheduling an eye exam may initially request an appointment at the Flatiron clinic before realizing that it’s more convenient for them to go to the West Village location. Replicant builds in a “back button” into every Thinking Machine™️ to ensure that conversations don’t fall off the rails the second a customer makes a non-linear request.

6. Powers & Slot Filling & Contextual Carryover There are infinite ways for customers to say the same thing. This can be seen especially in the selection of specific appointment times. A customer may say “sometime around 10am works well for me,” but not offer a preferred day. The Thinking Machine™️ can react off the time to say, “Okay, I can make a 10am appointment for you, which day do you prefer?” without forcing the customer to rephrase their request.

7. Reduce Customer Effort When designed with the customer in mind, Powers can go beyond intent detection to reduce customer effort by detecting contextual information. For example, instead of reading back an inbound caller’s phone number, Thinking Machine™️ can log a new number and simply ask “is this the best phone number to reach you?” Maximizing simple answers further reduces handle times and creates a frictionless conversation.

8. Purpose-built Natural Language Understanding NLU models play a key role in the accuracy of the Thinking Machine™️. In a flow like email collection, things can get complicated quickly. Replicant has designed it’s NLU stack to go beyond simple transcription solutions that have long struggled to discern accurate information from alphanumeric strings. For example, it can accurately collect a unique email readout to know that “M as in Mary dot R I L E Y @gmail.com” is “[email protected].”

9. Omnichannel Experience & Conversational Continuation After designing these layers of complexity into a voice experience, the next step is filtering a great experience into every channel. The Thinking Machine™️ is built to serve every customer in the channel they reached out over, or switch between channels when convenient. For example, a customer can be given the option to receive an appointment confirmation over SMS, while remaining on the phone to refill a prescription.