Article

Revolutionizing CX with Generative AI and LLMs

Innovations in customer and agent experience with AI

Generative AI and LLM for Contact Centers

The key drivers of adoption of CCaaS solutions are the desires to better enable customer and agent experiences, increase the availability and efficacy of customer self-service capabilities, and gain advanced insights into customer experiences both within and beyond customer service activities, according to the Gartner Magic Quadrant for CCaaS 2023. We would be remiss not to mention the growing excitement around generative AI, large language models (LLMs) and their ability to impact all three of the mentioned drivers, among others.

Up until now, CCaaS integration of generative AI and LLM technology has primarily been limited to conversation summarization capabilities. However, many CCaaS providers and independent software vendors adjacent to this market are actively developing capabilities that promise to address all of the three market drivers and more. This has tremendous potential for the operational customer service improvements that these solutions could bring. Gartner expects that these enhancements will largely augment agent activities within the five-year planning horizon and in doing so, increase CCaaS vendor revenue.

Meanwhile, organizations are increasingly looking to contact center as a service (CCaaS) and Workforce Engagement Management (WEM) providers to offer platforms that orchestrate and analyze seamless customer journeys to help agents support customers throughout the process and across all digital and voice channels.

This decade, applying artificial intelligence to various aspects of the customer journey has become a best practice in the industry and SuccessKPI is a vanguard of AI innovation to improve the customer (CX) and agent experience (AX). We have been quietly building intelligent functions for more than three years and just announced our AI Capabilities Portfolio and AI Roadmap. From 2021 through 2023, we have developed:

  • Sentiment Analysis & Monitoring – uses natural language processing and text analysis to decipher the emotional tone behind the language customers use to determine their feelings about a company and its products and services.
  • Automated Quality Monitoring (QM) – machine scoring predicts every outcome of interactions agents have with customers to help them improve their performance.
  • Topic detection and Phrase Enrichment – adds training phrases to agent and chatbot vocabularies that are often associated with a particular customer request. This helps determine the customer’s full intent and satisfy their request more quickly.

Recently we also introduced our newest capabilities: Forecast Traffic, Agent Assist and Call summarization.

  • Forecast Traffic – This capability predicts customer interaction volume and staffing needed, factoring in the contact center operator’s desired service levels and staffing characteristics using AI/ML forecasting algorithms. Consequently, schedules can be generated from staffing forecasts that account for experience, availability and working hour preferences of individual agents.
  • Agent Assist – This allows customers to monitor calls in real–time and assist agents with targeted guidance scripts and articles. The dual AI/ML technology predicts the outcome of each question a customer asks, enabling agents to move faster and more accurately toward resolution. This clearly improves the customer experience by elevating the agent performance while also enhancing regulatory compliance. Our customers have enjoyed our offer which can be accessed from within CCaaS desktop to Chrome extension to within Salesforce object layout. Ability to take real-time actions like passing topics to salesforce knowledge management or creating a lead or case is just icing on the cake.
  • Call Summarization: All calls can now be summarized in few sentences and completely eliminate the need of having spent 1-2 mins after the call to summarize the conversation. Now machines can do that with much better accuracy. ROI is huge in this case and should not be ignored.

Next, the addition of generative AI featuring large language models will raise employee productivity by reducing average handle times; improve the quality and accuracy of interactions by creating reusable knowledge content; and drive higher self-service containment rates through better conversational virtual agents.

Going forward in 2024, we plan to deliver a range of new generative AI capabilities such as generative intelligence, generative AI Expansion, Diagnose Topics and Intelligent Summaries that will make agents more productive and efficient, increase customer satisfaction and loyalty, and drive down the costs CCaaS operations, all securely, safely and in compliance with industry regulations.

Of course, responsible operators will weigh any perceived risks associated with using generative AI. At SuccessKPI, we have accounted for the security aspects in our development of AI in our labs. It was our first priority to make sure that not a single byte of data leaves customers’ accounts. Their data stays in the boundaries of their deployment. Any model training that we do is exclusively for that specific customer and their data is secure within their boundaries.

This is security within a private cloud. As part of our AI Pipeline process we follow this sequence of steps:

  1. We collect data from the customer
  2. We create models
  3. We evaluate those models, and how good they are
  4. Once validated, we say “Start predicting the next set of costs or next set of events.”

All of this happens within the customer’s private instance. Not many companies are fully aware of the security of data. For SuccessKPI, it’s not just the product and effectiveness of the product, but also the approach towards building the product that keeps customer data secure. It’s foundational.

Now with generative AI, these large language models are pre-trained with a lot of public data that could be limited or influenced by the particular search engine being used with that version of generative AI . What sometimes results from this is an AI “hallucination.” Customers, however, do not want their machine to hallucinate or recommend the wrong product because the LLM AI is trained on billions of data points outside just customer data. At SuccessKPI, we filter the data and make sure that the machine is aware of customer specific problems and domain and business and we do it securely. We are also making sure that these pre-trained models are not hallucinating and recommending something outside certain boundaries. This is important not only for accuracy but also for security.