The Low Hanging Fruit In CCAI
In this article I analyse the Gartner AI use-case Prism for AI in Customer Service. The question is often asked, what are the components of Contact Centre AI (CCAI), and how can CCAI be broken down into use-cases for manageable implementations? In this report Gartner breaks Customer Service down into four segments. And identifies 18 use-cases for AI within Customer Service.
The TL;DR
- The Customer Service process is segmented into four steps; these four steps in sequence are Getting Connected, Process Orchestration, Resource Management and lastly Knowledge & Insights.
- Spread across these four segments, Gartner identifies 18 AI use-cases within the Customer Service area.
- In this article I focus on six of the use-cases raised by Gartner.
- In a recent Gartner report emphasis was placed on areas of concern in the chatbot deployment cycle.
- Gartner advised that chatbot teams must focus more on customer intent, rather than technology features and metrics like containment.
This report from Gartner simplifies the CCAI idea by segmenting CCAI into four areas of concern. The four segments start with the customer making a connection, from where orchestration follows.
Orchestration is one of the neglected aspects of CCAI. In a recent conversation I had with one of the founders of OneReach.AI, it was raised that everyone is speaking of orchestration, but no-one is doing it.
A simple example to illustrate this point, consider the following… currently a customer is not able to drop a call midstream while speaking with a voicebot, and subsequently start a session with a chatbot, with the conversation continuing where it was interrupted.
The next element is resource management, ending in knowledge and insights.
CCAI Use-Cases
The list of 18 CCAI use-cases are indicated below and the report indicates in which areas of the customer interaction the use-case is relevant.
The six use-cases which are of particular interest to me are listed here, with their respective rank of priority from Gartner. This priority is based on business value and feasibility.
0️⃣1️⃣ Speech Analytics of Sentiment or Topics
0️⃣2️⃣ Intent Training
1️⃣2️⃣ Conversational Customer Assistants for Self-Service (Chatbots)
1️⃣4️⃣ Virtual Assistant for New Agent Onboarding
1️⃣5️⃣ Real-Time Agent Coaching
1️⃣8️⃣ Intelligent Contact Routing
1️⃣ Speech Analytics of Sentiment and Topics
An advantage of speech analytics and determining sentiment and topics is the luxury of performing this analysis off-line.
Topics is useful in determining the reason for a user making contact. I guess one could see topics as a first step in determining intents. This use-case is useful not only for agent conversations, but voicebot and chatbot interactions.
2️⃣ Intent Training
In a report on chatbot deployment, Gartner identified one of the main contributors to chatbot implementations failing, is a lack of focus on intents, which leads to poor user experience.
Whilst customer conversations and utterances do exist from current customer care interfaces like IVR, email, etc. These utterances/conversations can be used for creating intent clusters which are semantically similar.
Considering CCAI, Gartner also places high emphasis on intents and ranks “Human-in-the-Loop Intent Training” second highest.
Gartner also rates Intent Training highly feasible in CCAI, yet the necessary focus is not there.
1️⃣2️⃣ Conversational Customer Assistants for Self-Service (Chatbots)
Chatbots find themselves with high business value, but feasibility is low. The business value has been established over and over again. Yet the feasibility or probability of delivering on ROI is in doubt.
Looking at the Gartner Hype Cycle for AI 2021, it is apparent that chatbots are currently in the trough of disillusionment. The red arrows in the hype cycle (added by me)denotes other human language related technologies.
It is evident from the hype cycle why the feasibility rating for chatbots are low. There are efforts afoot to remedy this, in terms of bootstrapping with search, knowledge bases, LLMs and industry specific presets.
1️⃣4️⃣ Virtual Assistant for New Agent Onboarding
Chatbots (Virtual Assistants) have the capability to address an audience of one. A number of companies have started their customer care chatbot journey with an in-ward, agent facing conversational interface.
A good source for the conversation structure can be existing agent training material and also archived customer and agent chat transcripts. Using past conversations are becoming a focus area for planning and understanding how to serve customers better. While we are on this topic, item 10, Redacting of Personal Identifiable Information (PII) is receiving special attention due to the need to re-use customer conversations for internal AI training.
Microsoft recently launched and added to their cognitive services a feature to detect and redact Personally Identifying Information (PII) in conversations.
1️⃣5️⃣ Real-Time Agent Coaching
To me real-time agent coaching feels closely related to a virtual assistant for new agent onboarding. The one assistant can be an extension of the other. One agent assist conversational interface makes sense. One approach taken at a large telco was to ask users to state the reason for their call by the IVR, prior to transferring to an agent. The user utterances were recorded, and these 10 second voice recordings were subsequently submitted to the acoustic model for Advanced Speech Recognition (ASR) and transcription to text. The accuracy of these machine transcribed recordings only had a 4% deviation in intent recognition compared to human transcriptions of the same audio.
Having this intent prior to agent transfer, goes a long way in preparing the agent’s desktop for the incoming call and facilitating real-time, call intent specific, agent coaching.
1️⃣8️⃣ Intelligent Contact Routing
There can be debate about what intelligent contact routing entails exactly, a large part will definitely be customer identification and a customised interface; no AI needed for this.
But I do see orchestration as part of this process, creating a seamless customer experience with continuation across the mediums the user decides to use.
So this does not only include routing a customer on their current interaction, but also treating the customer within the context of their previous interactions, with the relevant continuation.
Finally
One thing which is evident from this list of technologies are the dire need for orchestration, and matching the user intent with business intent.
Considering the use-case prism below, the human-in-the-loop intent training is highly feasible and has only a medium business value. The latter is highly debatable.
The chatbot use-case has a close to high business value but medium feasibility.
The low feasibility is most probably related to the low level of success in terms of chatbot implementations and reaching the objectives set for success.
Other Gartner research has shown that whilst investment in chatbots is on the rise, 70% of customers who try self-service during their resolution, only 9% are wholly contained within self-service.
