Generative AI is already being used in contact centers today, with more use cases emerging rapidly. For now, at least, these use cases seem weighted toward assisting agents, rather than customer-facing scenarios. Still, the lure of agent replacement may be too strong to resist, according to one contact center consultant I spoke with earlier this month.
To start with the current use cases, the big winner at the moment is abstract summarization, according to analyst Dave Michels of TalkingPointz. In a No Jitter webinar last week sponsored by Five9, Michels called summarization “the killer app” right now. Showing how such a feature works, Richard Dumas of Five9 described a use case where the contact center system generates a transcript of a customer call, which it then presents to GPT-3—the large language model (LLM) that preceded the latest-generation GPT-4, which powers the now-famous ChatGPT chatbot.
In the use case, the system tells GPT-3 to, “Summarize the call, making note of key information collected from the agent, such as customer name, address, and products mentioned.” The agent can opt to edit or approve the summary.
The LLM is “extremely good at summarization. It gets the salient points, it does it very quickly,” Michels said, adding, “Just having generative AI do the agent wrap-up in a contact center can offer a pretty significant time savings.” Dumas noted that saving even 1 minute from a 5 minute call means a 20% cost savings to the contact center.
The summarization feature is already in general availability in Five9’s offering. Other features in beta or coming soon include:
- Intent Classification: Contact centers already use natural language processing (NLP) engines in conversational AI to provide intent classification—understanding what the caller is asking for. This function could shift to generative AI. “It turns out that LLMs are great classifiers,” Dumas said. “We are experimenting with the idea that you could use GPT-3 as an alternative to one of the NLP models to do intent detection and classification.”
- Entity Extraction: Using generative AI to, for example, break down an address that’s spoken into the system by a caller. Instead of having to prompt the caller, in succession, for street address, city, state, zip code, the LLM-powered system can simply ask for the whole address, and then the generative AI can produce code that separates these elements out.
- Insights: Generative AI can be used to more quickly and easily glean business insights from a set of conversational data, for example analyzing reasons for calls (e.g., requesting an exchange or refund) against data like average handle time, average hold time, and average queue time.
Michels and Dumas emphasized these use cases as ways of freeing up the agent to spend time and be more empathetic and responsive to the caller. “The agent is now focused on the conversation,” Dumas said. “All of this transcription, the note-taking, eventually even the dispositioning will be happening in the background, with the AI assisting the agent.”
I was glad to see the focus on agent-assist functions for generative AI, especially after a conversation I had earlier this month with contact center expert Amas Tenumah, author of the book, Waiting for Service: An Insider's account of Why Customer Service is Broken + Tips to Avoid Bad Service.
In our conversation, Tenumah expressed the concern that contact centers will look to generative AI to provide faster, cheaper customer service, which he believes will lead too many contact centers to see generative AI as just a way to build a better chatbot—in spite of the fact that, as he put it, “Your customers have not asked you for another chatbot.”
He sees huge potential for generative AI in improving the agent experience. “We have largely ignored the most important human, which is the contact center agent,” he said. The agent’s biggest problem is cognitive overload, which generative AI is well positioned to alleviate. A generative AI system that listens to what the caller is saying and instantly provides the agent with supporting information gives the agent a “superpower,” he said.
“If we point this tool at the right humans, and not all this consumer-facing stuff, I think our consumers will be happy, and I think our shareholders will be happier because it will deliver on the ROI,” Tenumah said.
So if technology providers like Five9 are leading with back-end capabilities, and experts like Tenumah are there to remind the industry of the critical role of the agent experience, might some contact center decision-makers resist the temptation to see generative AI first and foremost as a customer self-service, cost-cutting tool? There’s always hope.