How self-learning agents are improving customer service experience

Many organizations are turning to AI agents to automate customer service interactions.

Breakthroughs in generative and agentic AI have enabled a new generation of customer service AI agents that can handle complex conversations and connect with external tools and data sources (like CRMs, CDPs, and data warehouses) to complete transactions in much the same way a human agent would.

Creating AI agents that perform reliably for enterprises with large customer bases and contact volumes requires significant technical functionality that goes beyond the simple drag-and-drop interfaces offered by many conversational AI platforms. Users must be able to access fine-grain controls that enable them to tweak LLM, speech recognition, and text-to-speech performance.

This work isn’t over when the AI agent goes live. When your agent is engaging with real customers, you have the opportunity to tweak certain elements to enhance performance.

Until now, this has been a manual task handled by teams of product managers and IT professionals.

But now, a new ecosystem of self-learning agents is here to provide an autonomous improvement loop that drives optimal performance with minimal human intervention.

An ecosystem of self-learning agents

In the very near future, enterprises will leverage an entire ecosystem of self-learning AI agents to continually optimize customer service experience.

At PolyAI, we offer three types of AI agents that work together to enhance interaction, containment, and experience.

Customer service agent

This AI agent handles customer service interactions just like a customer service representative would.

It handles thousands of conversations simultaneously, across channels, and connects with other systems and data sources to complete transactions.

Judge agent

This AI agent evaluates conversations based on changes being made by a platform user.

For example, the user might update the knowledge base to include specific language, correct erroneous speech-to-text transcriptions, or LLM behavior. The feedback can also be more nuanced. For example, the platform user might provide feedback on examples that use emojis, so that the AI can generate more emojis in the future. Or perhaps the platform user prefers the AI agent to hand calls to agents in specific circumstances, or they prefer sentences that contain URLs. All of these preferences are trainable signals that can be incorporated into the judge agent.

As these changes are made manually by the platform user, the judge agent looks for opportunities to update rules and processes to enable consistency across interactions as platform users focus on making rapid enhancements.

Testing agent

This AI agent acts as a customer composite, generating realistic ways a customer might engage with the customer service agent, based on past conversations.

The testing agent then engages in simulated conversations with the customer service agent. It evolves by learning from real user behavior to reflect how callers are changing over time.

How the self-learning loop works

The customer service agent, the judge agent and the testing agent work together in a continuous self-learning loop, following these four steps:

1. Simulation: The customer service agent engages in simulated conversations with the testing agent.
2. Exploration: The testing agent provides varied and evolving scenarios, enabling the customer service agent to explore a wider solution space.
3. Evaluation: The judge agent assesses conversations, offering preference-based feedback aligned with human judgement.
4. Adaptation: Each agent continues learning.
   a. The customer service agent learns from real users.
   b. The judge agent learns from human team members using the PolyAI Agent Studio platform.
   c. The testing agent learns from the simulated conversations.

Autonomous agents and the role of humans

As agentic AI continues to evolve, more of the work typically handled by customer service teams will be automated. What does this mean for human workers?

Firstly, customer service representatives will still exist. We expect up to 90% of all customer service interactions to be automated in the future, but there will be a role for humans to play for quite some time, if not forever.

We’re experiencing a cultural shift towards comfort with talking to AI, thanks in part to ChatGPT which many of us now use in our daily lives. The voice channel, too, is on the up, with 86% of Gen-Z saying that the phone is their preferred channel for customer service.

But we still have a long way to go to prove that AI deserves the right to handle customer calls. Too many of us have had bad experiences with automated phone systems and until these systems are obsolete, it’s going to be difficult to earn customers’ trust. Until AI gets customers’ trust, enterprises are going to need to offer human customer service representatives.

But even in a world where AI is trusted – by both organizations and their customers – to handle all customer service interactions, enterprises will likely still keep great customer service reps on the payroll. These representatives will be very different from today’s tier 1 support reps, acting more like brand advocates, on-hand to build personal relationships with customers. If your 86 year-old customer is lonely and wants to talk to your service team for half an hour, and you truly value that customer, why not give them what they want – especially when loyalty is at stake?

Of course, AI will replace the work of thousands of customer service advisors, but the truth is that today, those staff members mostly don’t exist, which is why customers spend hours of their lives waiting on hold for customer service.

With AI handling more and more of these interactions, enterprises are already starting to upskill their top customer service representatives into AI agent supervisors. These supervisors will oversee AI agents, adapting systems in real-time in response to customer insights surfaced by AI. They will be rapidly adapting customer journeys to reduce friction and enhance experience at every step, as well as reporting back to strategic teams across the organization to inform a new level of customer-centric decision making.

Conclusion

Over recent years, customer experience has declined in the name of automation. As organizations focused on cutting costs by deflecting customers away from reps, automated systems offered clunky experiences, many of which we still encounter today.

But as AI agents become more and more capable of autonomous improvement, customer experience will improve, and service teams will be freed up to drive more strategic decisions across entire organizations.

At PolyAI, we’re proud to be leading the way with self-learning AI agents focused on improving customer experience and driving business results. Request a demo today to learn more.