It has been roughly one month since Google introduced a new set of services called Contact Center AI at its  Cloud Next Conference 2018. As depicted in this slide show by Sheila McGee-Smith, executives from Genesys, Cisco, Mitel and Twilio lent legitimacy to the new suite and its ability to employ DialogFlow (the rebranding of API.ai’s core capabilities) to understand the intent of a caller. Google and its alpha partners have found the approach to be quite successful, as evidenced in this blog post in which Cloud AI’s head of product management, Apoorv Saxena, along with group product manager Geordy Kitchen note that “more than 800 customers have signed-up for alpha access, supporting collectively more than 300,000 agents.”

Google’s Contact Center AI amounts to training wheels for Conversational Commerce. With a focus on Intelligent Assistance, the use cases and product descriptions illustrate how the technology is being employed to inform virtual agents, suggest “next actions” to agents, or perform analytics in the background to support precision routing. Instead of automating customer care in contact centers, Google’s product managers see their new offering as part of an effort to “make support more human.”

The more cynical cohort of technology analysts see a thinly veiled effort by Google to gain access to new troves of unstructured data. Recall that the company used the hundreds of millions of calls to its directory assistance service to train its automated speech recognition API to do a better job of recognizing place names and proper nouns, as I noted in the 2010 post “Goog411 Has Served Its Purpose”. Yet, as Jesse Scardina, a reporter for SearchCRM was assured by a Google spokesperson in this article, “Google is only acting as a data processor for contact centers and will not retain the data or use it to train its models.” That’s a pity, and also not credible, because the only way that Intelligent Assistance algorithms learn and improve is by ingesting more training data and getting feedback from humans.

Looking Back to Look Forward

Thomas McCarthy-Howe in a post on Medium makes the observation that the new offering is really a look backwards. Core to his argument is that the percentage of households with “Plain Old Telephone Services” (POTS) or “landlines” has declined to 17% in the United States, while mobile phones, 75% of which are “smartphones” are in the hands of 73% of the population. The attention paid to bringing AI into the callpath of conversations that originate from telephones, he reasons, is misplaced. It ignores the fact that a growing percentage of the worlds population prefers text chat over voice.

Some readers will recall that, in late 2009, Thomas and I collaborated on a White Paper called “The Recombinant Telephony Ecosystem: Voice Mashups and the Telco API”. We described the opportunity areas as well as the challenges surrounding “voice mashups”, which were in the process of being defined. “Telco APIs” which are now mainstays in the world of agile service creation that accompanies cloud-based contact centers and data lakes we can claim some high ground in this conversation. Our vision was that entire new businesses and service offerings would result when developers “splice together the basic materials of today’s communication technologies with new software elements to introduce new services that support customer requirements.” The end-game was to “fulfill on the Internet’s promise to support a better user experience for commerce, communication and collaboration.”

Nine years ago, we got that concept exactly right; mostly because we defined it. Today, Thomas asserts that Google AI for Contact Centers is a retrograde phenomenon. While the universe spins toward applying AI to inform mobile apps on smartphones and text-based messaging platforms, Google appears to be stationary and falling behind by providing resources to support humans employing and employed by the so-called “voice channel”.  Yet the choice between “Speech and Text” is one of the “False Choices” I explicitly called out in this post at the end of 2015.

Keep Avoiding False Choices

Google, of all companies, well-knows that moving speech rec and natural language understanding to the so-called edge is powerful. Google Assistant, has established itself as a reliable and accurate conversational user interface on Google Home (and some other smart speakers) as well as billions of smartphones. “Hey Google” supports command and control of devices, natural language search and commercial transactions. This does not preclude Google from using its core natural language understanding, machine learning and cognitive resources to support consistent responses to queries or conversations through IVRs, enterprise intelligent assistants or contact center agents.

Ten years from now, it is safe to say that contact centers, as presently configured, will be quaint relics. Like “classic” IVRs, which were stand-alone boxes in a switch room, they are being supplanted by APIs or microservices delivered from cloud-based resources. Mashups are happening!

Bear in mind that the new architectures and resources support the full spectrum of text, voice and video. Conversations routinely combine all three modalities at once, driven by context and user preferences. The decline in the number of landlines that Thomas cites has not led to an equally dramatic decline in voice-based conversations between brands and their customers because the number of overall conversations – both text and voice (and a growing amount of video) – is exploding. Messaging apps on smartphones may be the predominant point of origin, making text or chat today’s darling. Yet a high percentage of long-lived conversations get “escalated” to voice-based interactions with live agents. That’s why it is important for Google to offer a flavor of its AI resources to inform IVRs or agents in the contact center.

In his post on Medium, Thomas McCarthy-Howe was asking why Google would bother to bring Contact Center AI to market. In my comment to his post, I note that Contact Center AI is a marginal investment to the search giant, meaning that my overarching question is, “Why wouldn’t they?” And now, with 800 applicants for the alpha program representing 300,000 agents (and untold IVR ports), it’s clear that the marginal cost to enterprise brands is minimal as well. Regarding the offering, as I noted above, as training wheels for more complex conversational offerings, this commoditization of elements of AI marks maturation, not regression, of the Conversational Commerce ecosystem.

Categories: Intelligent Assistants, Articles