As we reported, Google recently announced new speech and NLP enhancements to Cloud Speech-to-Text and Dialogflow that can improve recognition accuracy by up to 40%, while ensuring better virtual agent performance.
To provide further insight, we decided to reach out to Inference Solutions CEO, Callan Schebella. Speaking to Callan meant we could learn more about how Google’s innovations are being used by customer service departments to provide AI-powered self-service. We wanted to understand how these incremental advancements in accuracy and performance translate into practical business value for service departments.
Inference Solutions are resold by leading telecommunications carriers like AT&T, Telstra, and Vonage as well as Contact Center as a Service (CCaaS) providers like Cisco and 8×8. These offerings reach hundreds of enterprises with virtual agents around the world. Recently, Inference was even recognised as the global market share leader by DMG Consulting. As a well-respected voice in the UC space, Callan was in a great position to share some useful insights with us. He has a long-standing history in the industry, having spoken at recent events such as Enterprise Connect and VentureBeat Transform.
Why are Google’s updates important?
Google is innovating rapidly with its Contact Center AI (CCAI) framework, and with technologies like Dialogflow that underpin CCAI. This is impressive because the quality of the technology is advancing at a pace we’ve never seen before.
One reason that speech-to-text accuracy is improving is that Google is training their technology with millions of consumer interactions collected via smart speakers and devices. Google has also released text-to-speech that is extraordinarily life-like. There’s also access to natural language processing (NLP) that lets consumers speak more naturally to virtual agents. Google has recognised an almost universal consumer acceptance of speech as a primary user interface and of self-service as a critical requirement for customer care.
Today, Google is leveraging its substantial resources to invest in empowering virtual agents and ensuring they can satisfy rising user expectations for high-quality customer service.
Improvements to Dialogflow like Auto Speech Adaptation and the updates to Speech Context Parameters are two examples of recent updates that support today’s virtual agents. These tools enable agents to transcribe text more accurately while considering the context of a conversation for greater accuracy.
This allows virtual agents to do what humans do when trying to understand what someone is saying. For example, if you told me, “I’d like to book an appointment for an interview,” I’d expect that the rest of our conversation would be about booking that appointment. I’d then be less likely to confuse the word “meet” with “meat.” They sound the same, but given the context, I’d know that you weren’t trying to “meat” with me.
Perhaps the most critical update is that Google and other API providers have moved speech and NLP into the cloud, which is helping to reduce the cost of natural language self-service. This gives businesses of all sizes the ability to deploy applications where previously only large call centres like those used by banks and airlines could afford to leverage it.
Can you give us some real-world examples of businesses that are using these innovations?
Absolutely. Pizza Hut recently announced that they had deployed virtual agents using Google’s Cloud Speech-to-Text, Text-to-Speech and Dialogflow for conversational AI and Google’s mapping APIs to provide directions. This enables them to provide automated support to customers who are trying to find local restaurants and place orders.
The technology helps Pizza Hut to provide better user experience, especially for mobile consumers. With these innovations, the company can reduce both contact centre and store costs where routine inquires either incur the price of a live agent or require extra staff in a store to answer questions. You can find a case study about their project on our website.
Another example would be Red Lion Hotels, which recently announced that they had deployed virtual agents to support customer inquiries to 1,400 properties. Virtual agents are helping them answer questions from travellers and reduce the costs of non-revenue related calls.
What challenges will organisations face when attempting to deploy virtual agents?
I’d say the biggest challenge is still that developing virtual agents is too difficult for most organisations. Although Google and others are innovating at a breakneck pace, you still need a team of developers with expertise in JavaScript, JSON, and other programming languages to develop your agents.
There’s also the challenge of handling “fulfillment” – for example, once you understand that a caller wants to book a hotel reservation, the developer still has to complete the transaction by making API calls to a booking system like HotSOS. That’s the reason providers like Inference have developed solutions that make it easier for non-technical users to build and deploy self-service applications.
Another issue for businesses is understanding that building a natural language application is not the same as creating an IVR. With applications that use NLP, there is a 30-60-90 day learning phase where your virtual agents learn and become more capable over time. Once a service department understands this, they can establish KPIs to measure the increase in productivity over time.
Should service departments focus on using AI for agent assistance or virtual agents?
There’s no question that agent assistance is essential. Today’s businesses should focus on arming their agents with the best tools and the best answers to respond to customer questions.
We know that, as AI-powered self-service handles more and more of the routine interactions, more complicated inquiries are being routed to live agents. But that doesn’t mean that businesses should focus only on agent assistance. And in fact, very few of them do. Customer service managers for years have been trying to reduce costs with self-service using the web, IVRs, and other self-service channels.
With a growing number of consumers, expecting support across more and more channels, 24/7, they worry about being overwhelmed by the volume of service requests. They could try to hire their way out of the problem by onboarding more and more live agents, but the costs would be backbreaking. That’s why just about every service department is scrambling to put in place a robust self-service strategy and is embracing virtual agents to help.
The numbers speak for themselves. Analysts report that the number of live agent seats deployed plateaued in 2018 at just under 20 million worldwide and have now actually begun to decline. At the same time, Gartner forecasts that by next year, virtual agents will participate in the majority of all commercial interactions between people and businesses. They also forecast that virtual agents will drive $1.2 trillion in business value by 2030 through cost-savings, improved customer experience and new revenue generated.
Focusing solely on supporting your live agents and missing out on the opportunity to implement advanced self-service would be a mistake. It’s like investing in the pony express while others were building the telegraph, or like Steve Jobs investing in LPs instead of iTunes. Even worse, by the time you realised the mistake it could be too late to catch up.
That’s why virtual agents are a smart investment for businesses as they prepare for the next wave of customer experience.
