What Should You Do About GenAI in the Contact Center?
Customer service functions are on the cusp of a transformation because of generative artificial intelligence. Large language models (LLMs) can redefine how contact centers operate, and customer service leaders must understand both the near-term and long-term possibilities.
The AI-Led Reinvention of Contact Centers: Medium-Term View
First, because of the seeming ubiquity of AI and the near-constant hype and FOMO, it’s worth repeating that we are in the early stages of adoption. In fact, the AI technologies that may well reinvent the contact center are still in the labs.
- Multimodal LLMs: For example, multimodal LLMs that can process customer interactions in multiple formats, be it text, voice, or visual inputs such as images and videos, can be a game changer for customer service. Seamless handoffs and transitions between channels may be within reach with multimodal LLMs. But while we have excellent text-based (and to an extent image-based) LLMs, multimodal LLMs are still being cooked up.
- New speech tech paradigms: To take another example, OpenAI said its latest GPT-4o model includes advanced capabilities such as directly processing speech input without first converting speech to text. It is still undergoing performance and safety testing.
- Agentic AI: This involves AI agents that can divide up a task into multiple parts, accomplish each part using the best-fit tool, and then stitch them back together. Arguably, AI agents will be able to automate more customer service interactions, but AI agents (which sound a bit like RPA bots with brains) are also currently experimental technologies.
The impact of these newer AI technologies will likely be felt more in the medium term, say the next three to five years.
Meanwhile, there are genAI technologies to consider in the near term for customer service leaders looking to increase human agent efficiency and better serve customers.
The AI-Led Transformation of Contact Centers: Near-Term Game Plan
When equipped with the right guardrails, current genAI technologies can handle level 1 and level 2 support tasks. The customer service workflows will consist of human agents and digital agents (AI bots), and clearly defined transitions between the two. To be sure, there are stories of AI chatbots gone awry and that’s why guardrails are essential. This allows for automating a greater portion of routine inquiries and enables human agents to focus on more complex tickets and value-adding issues.
RAG and Fine-Tuning
Many companies might not be comfortable deploying external customer-facing LLM chatbots but are more open to leveraging agent-facing LLM applications. Human agents can be better equipped to handle higher-complexity queries using an approach called retrieval augmented generation (RAG). In RAG, when a user prompts an LLM, the relevant information and context from the enterprise repositories is passed to the LLM system, allowing for more accurate and relevant responses. In short, RAG can help assemble contextually meaningful answers rather quickly and boost agent productivity. Note that RAG requires a bit of AI application development.
Fine-tuning, where the LLM is tweaked by training it with additional domain-specific data, is another useful approach. Additional training can enable responses to follow brand guidelines and communication standards. Fine-tuning can help ensure a consistent and polished customer experience. Note that neither RAG nor fine-tuning eliminate hallucinations (plausible-sounding but factually incorrect material) completely. But there are techniques to provide factual grounding to LLMs; assess them for your service scenarios.
Cost of LLMs
As AI models improve, we are also seeing their costs decrease. But the current costs are distorted by AI companies’ aggressive pricing strategies as they battle for market share. Factor in the software and services cost of building RAG applications, any cost of fine-tuning (and updates), and ongoing LLM usage charges as you build the business case.
GenAI and Global Sourcing
There may be larger shifts in the global shared services and third-party global services sourcing models. As AI capabilities advance, the traditional approach of offshoring contact center roles to inexpensive locations may get a relook. If the mix of work performed by human agents changes significantly, the locations might change as well.
The changes discussed here are profound but they won’t happen overnight. They are likely to unfold over the next several years. As such, customer service leaders should stay informed about the current capabilities of AI and its progress. x
Kashyap Kompella, CFA, is an industry analyst, author, educator, and adviser. He is the founder of the AI advisory outfits RPA2AI Research and AI Profs and is a For Humanity Certified AI Auditor.
Related Articles
The tensions and undercurrents in the wake of genAI should be better understood.
30 Aug 2024
AI-based voice apps can help in the early diagnosis of disorders such as Parkinson's and Alzheimer's.
09 Jul 2024
It is the best of times, it is the worst of times for this powerful technology.
14 May 2024