Stop optimizing chatbots: Why customer care needs agentic orchestration

Today, automated solutions resolve only about 14% of customer queries. Customers expect to ask a question to a chatbot, get a clear answer, refine their question, change direction and still be understood. They expect systems to keep context, handle ambiguity, remember their needs over multiple sessions and respond in a way that feels natural, because consumer chatbots such as ChatGPT provide these experiences.

Then, they ask their bank, insurer or telecom provider and hit a different experience with their customer service team. They wonder, “Why doesn’t this chatbot work the same way as the chatbot I use?”

Instead, they get rigid flows, repeated questions and dead ends that push them toward requesting a human agent. When that happens often enough, customers lose confidence in digital channels altogether. They might turn to external AI tools as their primary interface for understanding products, policies and services, putting businesses at risk of losing control of the customer relationship.

For customer care leaders, the consequence of this experience is immediate: rising service costs, declining containment and a breakdown in the customer experience. Replacing human agents with chatbots that can’t resolve issues leaves customers frustrated and unsupported. When a customer can’t get help, they don’t just disengage, they leave.

Most customer service automation was built on a simple model: guide the customer through a predefined path. Press 1 for this, 2 for that. Even as chatbots replaced phone trees, the underlying architecture stayed largely the same. These systems used rule-based logic with limited AI under the hood. They could turn speech into text, detect intent, capture a few entities and move users through scripted flows, but only when the conversation followed the exact path the system expected.

However, customers change direction, ask follow-up questions and combine issues. They expect the system to understand and properly interpret what they mean. Here is where earlier generations of automation tended to fail, and why many organizations continue to see low containment despite years of investment in customer service automation.

To meet rising expectations, a different approach is needed. One that doesn’t force customers into rigid paths, but can understand intent, adapt in real time and complete the task.

Large, complex service issues become more manageable when they are broken into smaller, focused tasks. Instead of asking one system to do everything, organizations can assign parts of the interaction to specialized AI agents.

One agent can start the conversation and interpret what the customer needs. Simultaneously, another can retrieve relevant knowledge, while a third one can execute a transaction such as updating an account or checking claim status, and so forth. Each AI agent does one job well, and together, they help resolve the full issue.

This model is the core idea behind agentic orchestration in customer care. It shifts automation from monolithic if-then programs to a coordinated system of AI agents designed around resolution.

A customer experiences a medical emergency and is short on cash for the month. She needs to request a deferral on an upcoming USD 500 car payment. Instead of calling a contact center, she turns to her bank’s AI assistant.

Here is how an agentic system handles the interaction:

• The primary agent understands the request and breaks it into tasks.
• A credit agent retrieves her credit score to assess eligibility.
• A payment history agent checks for missed payments over the past several years.
• A deferral eligibility agent verifies whether she has requested a recent deferral.
• A loan calculation agent determines updated loan terms, including accrued interest and a revised payoff date.

Each agent completes a specific task and shares the results through a shared context layer. The orchestration layer then assembles the outcome and presents the updated loan terms to the customer for approval.

If she accepts:

1. One agent updates the loan agreement.
2. Another updates the CRM system.
3. The interaction is completed end-to-end without requiring a human agent.

In traditional systems, each step in this process would have required explicitly programmed logic. Developers would need to define every rule, condition and possible path in advance, making the system rigid and difficult to adapt.

With agentic orchestration, the model changes. Instead of scripting every outcome, teams can define goals, constraints and policies. The agents dynamically determine how to complete the task, within governed guardrails. This process significantly reduces development time, from weeks of programming to hours of configuration, and results in systems that are far less brittle when real-world complexity inevitably arises.

Identify the 10–15 intents that drive the most volume, the most cost or the most customer frustration. In many organizations, that includes use cases such as billing disputes, claims, account updates, delivery or service status questions and other interactions that are both common and difficult to resolve efficiently.

These journeys are the ones where the economics are often clearest. Automated interactions typically cost USD 0.03–0.25 per call, while live-agent interactions range from USD 3.00–6.50 per call. When too many of these interactions escalate, businesses remain locked into human-channel economics. The opportunity is to improve the percentage of interactions that are resolved through automation, especially in high-volume, high-friction journeys.

The goal is not to remove human agents from the process entirely. It is to involve them at the right moments, when judgment, empathy, negotiation or exception handling is required.

Done well, agentic orchestration improves human handoff quality by passing along full context, reducing repetition and helping agents pick up the interaction without forcing the customer to start over.

Another advantage of orchestration is that it creates a stronger feedback loop.

By analyzing interaction outcomes and agent performance, organizations can refine workflows, improve coordination between agents, identify gaps in knowledge or systems integration, and steadily increase resolution rates over time.

IBM® watsonx Orchestrate® gives customer care teams a way to move from fragmented automation to a more coordinated, action-oriented model. It enables organizations to build and govern AI agents that work across existing systems and workflows, helping teams improve resolution without requiring a rip-and-replace of their contact center environment.

It also supports an open ecosystem approach, including voice and conversational capabilities from partners such as Deepgram and ElevenLabs. This approach allows businesses to design experiences with the latest technologies that better reflect how customers want to interact.

Most important, it brings orchestration, action and governance together. That means customer care organizations can do more than generate better responses—they can connect conversations to business systems, preserve context across interactions and operate within the controls enterprises require. The result is a stronger foundation for improving containment, reducing live-agent dependency and delivering more consistent customer care at scale.

See IBM watsonx Orchestrate in action with Farmers State Bank

Explore IBM watsonx Orchestrate