Toshish Jawale, CTO and Co-Founder of Symbl.ai, tells how machine learning can aid in serving customers better
Every customer interaction starts with a conversation. Whether they are taking place on the phone, chat, or email, conversations are valuable bits of data that organizations can use to gather feedback, improve their products or services, and boost retention.
For conversations to be tracked and made use of, it’s essential that organizations have conversation intelligence tools in their AI solutions arsenal.
While the concept of conversational AI points to an AI system that is capable of responding or holding conversations with humans, conversation intelligence refers to an entire ecosystem that includes AI-powered tools, data tools, and more.
What organizations usually get wrong when analyzing conversations is that they treat transcripts as documents. Using document understanding ML models can only get you so far in analyzing text, but conversations require a different approach.
Building on this point, Toshish Jawale, CTO and Co-Founder of Symbl.ai, says:
“Conversations are not documents. They are free-flowing articulations of a thinking process because they are not well articulated, formally and accurately written pieces of information.
“Most NLP techniques are built to process documents and they typically don’t work on conversations, in a sense that they do not deliver meaningful results. They are not built for solving the problems calibrated towards naturally flowing human conversations.”
Human-to-human conversations, as opposed to human-to-machine conversations, are chaotic, don’t follow a fixed flow and are very contextual in nature. Examples include sales calls, interviews, lectures, health consulting, webinars, and more.
Although all of these conversations have an agenda and a goal, they don’t follow a fixed flow like a chatbot would, for example.
Getting value out of such conversations can be done through two modes: real-time and asynchronous.
Real-time mode refers to extracting value from conversations as they are happening, through in-app actions like sharing a document with a customer or offering a discount. Organizations can also extract value through live call monitoring.
Asynchronous mode refers to data available after the conversation ends, and this includes the length of the call, conversation analysis, key actions to follow-up, and more. Jawale explains this in detail:
“In the real-time mode, AI sits with you through the call and notes down what is happening on the call, like specific keywords. It can identify if the customer wants to know more about a product and automatically share a specific document which provides more information. This action is suggested based on the context of the call.
“In an asynchronous mode, AI gets access to conversations either through audio or video recordings and is trying to understand what happened. For example, it can then offer follow-up actions based on the context of the call. All of this can be done in real-time as well.”
From an AI perspective, conversations have temporal, conceptual, casual, natural, contextual, and relational aspects. All of these are part of a natural conversation which makes them challenging for AI software to understand.
However, advanced conversation intelligence has come close to understanding and aiding in serving customers better mid-conversation. Jawal says: “The key thing is humans are always going to relate something outside of the actual scope of the conversation. AI needs to have enough ability to contextually relate to entities, and temporally define them.
“It’s quite hard because there is no system that can take all of these aspects into account with just a few API calls. It ends up being a quite complex system.”
Symbl.ai helps organizations to easily process and analyze conversations across any channel. Learn how their offering can enable you to extract value from all customer conversations by visiting this page: https://symbl.ai/