This site is operated by a business or businesses owned by Informa PLC and all copyright resides with them. Informa PLC's registered office is 5 Howick Place, London SW1P 1WG. Registered in England and Wales. Number 8860726.
Enlightening Generative AI
Across tech in general, and customer experience (CX) in particular, no topic has been more dominant this year than artificial intelligence (AI). It has become the primary topic in inquiries, product roadmaps, product announcements, and vendor events. AI in CX is not particularly new, but developments in generative AI have ignited a firestorm of new interest, investment, and imagination. Generative AI (gen AI) isn’t even new, but the public release of ChatGPT in late 2022 made it top-of-mind.
The public release of Open AI’s ChatGPT made obscure large language model (LLM) technology accessible and real. Some of the most advanced developments in computing hardware and LLMs suddenly were just a click away. ChatGPT reached 100 million monthly active users last January, just two months after its launch, making it the fastest-growing consumer application in history.
Yer a Wizard, ChatGPT?
ChatGPT was magical. It took the fiction out of the science fiction that we’ve all seen, where humans and machines interact seamlessly. Natural language is far more accessible than coding. Even better, generative AI could do the coding for us. Conversational bots aren’t new. They are frustrating gimmicks of deflection, but ChatGPT was different. It turns out that bots aren’t inherently frustrating, just the ones that don’t understand us are. It’s not hard to get from here to bots replacing agents — or is it?
Here we are in August, and generative AI doesn’t have much to show for itself in customer service. It has been touted in nearly every press release, but mostly doing background stuff – e.g., conversation summaries – which it is proving itself to be very good at. Most of the developments this year rely heavily on tried-and-proven predictive AI. This was the technology that was transforming customer service before ChatGPT.
I don’t mean to belittle summarization. It is a valuable feature. It reduces or even eliminates post-interaction processing time and often does it better than humans. But I will say call summaries feel like a disappointment. Generative AI seems like it should do more for customer service, and it will. But first, let me explain what’s taking so long.
We don’t trust generative AI and for good reason. We all know that these new bots tend to hallucinate — a fancy term for bald-faced lies. We are also terrified that gen AI can’t keep a secret, so we don’t provide gen AI the information it needs to be helpful. In addition to trust and security, other factors, such as compliance and cost concerns, prevent us from integrating generative AI with our business systems.
Gen AI’s Killer Feature in 2023
These pressures make summarization the killer generative AI feature in 2023. Summarization is a confined and specific task. Hallucinations are unlikely as summarization is so narrow. Summarization doesn’t require access to business databases, and it’s not too expensive to implement. Summarization can be accomplished without any training, though training can improve the outcome. For example, lots of industries use jargon that even a large language model (LLM) can’t understand. Supplementing the LLM with known terms, behaviors, sentiment information, and other conversational data can improve the quality of AI-generated summaries.
Generative AI will do more for the contact center, and I got my first glimpse of this at NICE Interactions, the provider’s recent user conference. There, CEO Barak Eilam announced and demonstrated three new products: Enlighten Actions for managers and executives, Enlighten Copilot to assist agents, and Enlighten Autopilot for self-service.
Generative AI was front and center when the Enlighten Actions demo started up with the prompt: “What do I need to know about my service operations today?” The response included three trends: CSAT, transfer rates, and conversion rates. The same prompt does not always return the same information. NICE has been building dashboards for years and has learned which metrics get double-clicked. Essentially, generative AI eliminated the dashboard.
Enlighten Actions then generated appropriate recommendations to address both good and bad operational metrics. In one of the examples, a programming change was needed to fix a routing issue. Enlighten identified, scripted, and implemented the change. Several other demos showed generative AI in all three new products.
Eilam finished the demos and proclaimed that “it is the future of CX available today.” It was a compelling keynote and demo because, clearly, NICE was doing things that I thought we weren’t supposed to do, like expose generative AI to customer data. I had some questions about what he demonstrated; here are his responses.
An interview with Barak Eilam, CEO, NICE
Barak Eilam is the CEO of NICE Ltd, and has been at NICE for nearly 25 years. During that time, he’s seen a few technology waves, though he indicated none as significant as what’s occurring now. NICE was recently placed in an upper left Leaders position in Gartner’s CCaaS Magic Quadrant.
Dave Michels (DM): What do you consider the main role of AI in today’s CX world?
Barak Eilam: Today’s CX organizations are tasked with fostering customer loyalty as they face increasing levels of complexity and rising labor costs. Our vision is leveraging AI as the ultimate technology super-wave to create a whole new approach to CX. We believe that AI is playing a pivotal role because it addresses these core business challenges in a way no other technology solution can. It increases the velocity of decision-making to overcome complexity by making smarter, more proactive, and faster decisions. It turbo-charges employee capabilities to amplify skilled labor capabilities. And AI is truly capable of creating personalized, humanized customer interactions that drive better loyalty.
DM: There’s tremendous hype and concern about generative AI in the contact center. Is Generative AI ready for mainstream use in contact centers?
Eilam: Generative AI is a game changer when it comes to making AI more accessible on a mass scale, opening up AI engines to a friendly conversational chat-like interface, and generating new levels of creativity never seen from AI before. We believe generative AI is more than ready for mainstream use, but it must be done correctly and with the adaptations necessary to address fundamental business requirements. Organizations that bought into the generative AI hype and tried to implement these generic types of solutions within their customer service environment quickly understood that these public domain solutions cannot be trusted to provide correct responses that are secure and aligned with their business goals. This is why we came up with Enlighten, which was created from the ground up as a trusted generative AI solution custom-built for CX and includes out-of-the-box AI models that are ready for mainstream contact center consumption.
DM: Which generative AI model(s) is NICE Enlighten using?
Eilam: We are currently using OpenAI GPT, but we architected Enlighten in a generative AI-engine agnostic way to leverage other LLMs as well.
DM: Are you exposing customer data directly to OpenAI?
Eilam: No. We are using a private instance of OpenAI GPT, and the data we provide it is not used for bot training. Also, we are not uploading training data as we do with predictive AI. It would be cost-prohibitive to do so. Instead, we leverage our decades of experience in intent modeling and upload a structured database optimized for the LLM.
DM: How is a structured database helpful for training data?
Eilam: With generative AI, it’s less important to train the AI with answers and more important to provide it with the data necessary to get the answers. We are essentially providing the LLM a word-vector, so it knows where to retrieve necessary information. Generative AI can generate SQL queries in databases.
DM: So NICE is using generative AI to create SQL or other scripts to get information instead of natural language responses?
Eilam: Exactly, for AI to be useful, it has to do 3 things: converse, understand, and act. For the past decade, NICE has been focused on using AI to understand and act. LLMs and GPT represented a breakthrough in human-machine conversational technology. Now, with a natural language UI, we can immediately generate queries for action.
DM: You are positioning Generative AI as a sustaining technology. That it will leverage existing investments in AI. Many people have suggested that generative AI is disruptive, and prior expertise may prove to be a liability.
Eilam: Generative AI is disruptive in the unprecedented way in which it creates human-accessible interactions with AI. But as the world is learning the hard way, adapting a generative AI environment to an actual business requires harnessing this disruptive interaction power to the right information, data, and models – and this is where prior experience comes in. For NICE, the combination of generative AI with over 20 years of customer interactions data and our domain expertise, embedded across our entire platform and applications suite, is what transforms this from a niche disruptive technology to a viable business offering that could be implemented from day one.
DM: I understand summarization as an AI-powered feature. Does NICE view generative AI as a feature, tool, or something else?
Eilam: You have to consider a functionality like AI-powered auto-summarization as part of a larger whole. We are not looking at generative AI to perform one specific task. We think of it more as a daily AI companion to customers, executives, and employees that can provide real-time answers based on the right information (i.e., company-sanctioned data), can proactively alert the user on something unexpected they might want to look at, take an action or make a specific decision. For example, Enlighten Copilot is much more than an auto-summarizing solution. It is right there, on the contact center agent’s desktop, experiencing customer interactions (voice or chat), learning the conversation context, measuring customer sentiment and agent soft skills, providing real-time insights and callouts throughout, guiding the agent on the next best action or proactively suggesting the next activity, composing a post-interaction email, reviewing documents sent by a customer, and yes, also auto-summarizing the conversation at the end of the interaction.
DM: Do you see AI replacing agents? Will generative AI enable agentless CX interactions?
Eilam: We view generative AI as a force multiplier, allowing organizations to amplify both assisted and self-service interactions with user-friendly prompts that are informed, trusted, and secure. For customer service professionals, it opens up a completely new world of possibilities where technology and service can be brought together in exciting new ways. CX has always been about creating the best possible experience for consumers, the kind of experience delivered by the best service employee. When you combine Generative AI together with access to customer interactions data and analytics, CX models, and prediction – you can deliver a new version of the “best” service employee that melds together both human-assisted and automated service practices.
DM: How are the early adopters doing with Enlighten Actions, Copilot, and Autopilot?
Eilam: We’re seeing tremendous results from early adopters. For many of our customers, this is their first time using generative AI and their first experience with NICE Enlighten and its custom-built AI for the CX world. As part of our Value Realization Service approach, we work with our customers to define some clearly defined early business goals that are attainable within the first few weeks of implementation. We always enjoy seeing the delight of Enlighten customers who, in 30 days, manage to reduce interaction times, improve their resolution rates on a particular issue by 50%, or see their agents’ soft skills improve by prompting them in real-time on AI-derived measured metrics such as projecting empathy, taking ownership of conversation and many others.
DM: Can you give us a hint at what comes next? Where is the technology going, and when might we see it?
Eilam: We invested a lot in AI in the last few years, and this momentum is going to continue in full force. Just like we did in the past leading the WEM, Analytics CCaaS transformations, we are committed to leading the AI revolution that is sweeping the world of CX. We have the largest engineering force in the CX space, annually investing over $300M in our 3,000 plus researchers and developers, with the industry’s largest number of AI-dedicated R&D professionals, enabling us to execute our vision and lead the CX AI revolution with rapid time-to-market and lightning-fast adoption of the latest AI technology trends. Our plans are to continue innovating and enhancing our unified platform, taking AI to the next level based on our shared data layer, which allows us to harvest and leverage data to provide better insights. We have an extremely comprehensive roadmap that we will be sharing soon.
“Soon” will probably be in October, as that’s when NICE will host its next analyst event. The venue this year is Peru, not far from Machu Picchu. That’s quite the juxtaposition: The latest advancements in AI and CX in the midst of 15th-century ruins.
Dave Michels is a contributing editor and analyst at TalkingPointz.