How to deliver a delightful call center customer experience with conversational AI in 2021

Undelightful realities

But the realities of telephone customer communication today are significantly less than delightful. The call center industry has long been plagued by excessively high attrition rates - multiples higher than any other industry in the country. An attrition rate of 100% per year is the norm for on-shore call centers. Customers have had to put up with long wait times for just as long. These and other issues have only been exacerbated by the pandemic. Here are some ways in which the call center industry has been hit the hardest:

• Bewitched, abandoned and overwhelmed. In the initial stages of the pandemic, call centers had to shut down. Some managed to transfer to full-time work from home in a couple of months. For others, it took till the end of the year. At the same time, inbound call volume went up multiple times. We documented some of the horror stories that ensued in this post.
• Lack of resources. A call center agent is far from a high-paying profession. With the government unemployment subsidies at an all time high, many call center workers realized that they can get the same, or nearly the same amount of money, as they get paid, for free from the government and not have to do any work.
• Work from home. While a short term savior, is a long-term liability for an industry as, oftentimes, sensitive, as call center. Forget about taking credit card numbers over the phone. Also, the worker morale suffers significantly.

Here are the ways in which a call center user experience falls short:

• Long wait times.
• Multiple holds and transfers while the agent looks up information (or puts the customer on hold to make the customer feel like they are doing something - this is standard practice).
• Agents who can be either too morose or too talkative.
• Long call times.
• Not managing to resolve the issue.

Learning to delight the customer

The year is 2021. There are literally zero good reasons to keep using the old IVR + call center agent customer service stack. Zero.

Build a human-like conversational AI-powered application that will service your customers better, fairer, more efficiently, faster than ever before. Oh, you can also design your application to actually be a delightful experience for your customer.

Here are the benefits of human-like automation versus human agents:

• Always on, always running.
• Doesn't ask for overtime when it picks up a call at 2:30 AM.
• Zero seconds average wait times.
• Has direct API access to your databases. Doesn't need to put the customer on hold to type and look for data.
• Has the mood you programmed it to have. Does not have bad days.
• Is always pleasant and respectful (if you programmed it to be pleasant and respectful).
• Can handle as many calls at the same time, as you need it to.
• Actually follows the damn script.
• Costs significantly less than a call center agent does.

There is only one catch. It has to delight your customer. If it does not delight your customer, your customer will be asking to get transferred to an operator, rendering all of your automation efforts useless.

So how do you delight your customer with conversational AI? Easy. Just build a delightful conversational AI app.

Generally speaking, a delightful experience will:

• Be respectful of your customer's time.
• Solve their problem.
• Leave them with a positive feeling about your product or service.

There is another aspect that pertains specifically to conversational AI. Your customers are used to speaking with human. You can let them know that they are speaking to an AI but your AI has to be good enough to where the customer does not sense much of a difference from speaking to an actual human.

In terms of Dasha AI - here is how we solve for human-likeness and here is how you can use our tech to do the same:

• Voice. Most speech synthesis either doesn't sound human at all or sounds too perfect. We are constantly improving our speech synthesis engines, working on keeping the voice as human-like as possible through introducing minor mistakes, disfluencies, inconsistencies. These are the qualities that, to the mind of a human, define the speaker as being, indeed, human.
• Digressions. People throw curve balls in the course of conversations. For example, we can be talking about the economy and suddenly you may interject asking how the weather is where I am. As a human, I know how to respond to your question and then come back to the conversation we were having previously. The vast majority conversational/chat bots will not be able to do the same. The ability to leave the main flow of the conversation, go off on a tangent, come back and still retain the context of what was previously discussed is essential to passing the Turing test (the test of human likeness).
• Adaptability. The conversations your AI app has in the summer will have different digressions than the same conversations during Christmas season. Your app has to adapt.

Getting your app to human-like

Generally speaking, it's a three step process to get your conversational app up to a state where it can satisfy your customers' demands for human-like conversation.

Look at the existing conversations

You've been keeping recordings of your customer conversations, haven't you? Great. Now you need to study them for the following information:

1. Identify and outline the "perfect world conversation flow". Likely it's the one you handed over to your call center agents as the script to follow. Unfortunately (or fortunately) they don't follow it, because they are human. Don't blame them. Look through their calls, maybe you will identify a script which resolves the issue quicker and more efficiently. Who knows, stranger things have happened.
2. Identify the most common digressions. As mentioned, digressions are ways in which your customers sidetrack your call center agents with questions, requests, etc. "Who is this again," "what's your name," "by the way, when did you guys change your logo" are all examples of digressions. Your job is to identify as many of them at the outset as possible.

Now that you've got the perfect world flow and the digressions, you can create the baseline conversational AI app. You can follow this guide for a way to build a sample app. Apply the same principles to building your app.

Test the phrasing

Now that you've got your app prepared, you will want to make sure that you are helping the AI voice to pronounce the phrases in the most human-like manner. To do so, you may have to chang some punctuation and/or words or phrases in your apps.

For example, "Hi, John, this is Dasha, a conversational AI agent with ACME corporation, is it a good time for you?" Will sound very rushed. A better way to say the same thing would be "Hi, John. This is Dasha. I'm a conversational AI agent with ACME Corporation. Is it a good time for you?"

You can test which phrasing works best with this terminal command. Just run the command and open the MP3 in the folder. dasha tts synthesize "Hi! This is an example." -o test.mp3

Train on live customer conversations

This is where the true magic happens. The best way to teach your conversational app to have truly human-like conversations is to let it have conversations. As the conversations stack up, the dasha studio analytics view will highlight areas where your conversational app is failing, so that you can amend the conversational flow to meet customer expectations.

In my experience, it takes between 1-4 weeks of training to get your app to perform over 92% of all conversations correctly. If you want, here is a bit more info in our documentation on the training process.

TL;DR

• It's 2021 - there is no reason to rely on human call center agents and decades-old IVRs. Implement conversational apps and make them as human-like as possible, tailored specifically to your customers.

How to make your apps human-like:

• Define the perfect world workflow/script.
• Listen to conversations and identify key digressions.
• Let your app talk to real life customer in real life conversations. Review results in analytics view and make adjustments to your app to meet customer expectations.

If you want to get started building conversational AI apps with Dasha, join our developer community and build a sample app using a tutorial such as this one.