Why Chatbots Still Fall Short, and How Agentic AI, MCP, and Human-AI Orchestration Can Improve Customer Service

Customers are not rejecting automation. They are rejecting automation that fails, stalls, or traps them. New consumer research reveals exactly where the line falls.

Key Takeaways

Traditional chatbots fail when customer issues require reasoning, context, or action across enterprise systems. In a recent survey of U.S. consumers, 83% said speaking with a human agent is very or extremely important, yet 56% are satisfied with automation if it resolves the issue quickly. Agentic AI, MCP, and Avaya Infinity help enterprises orchestrate the right balance of AI, human judgment, and governed workflows across the full customer journey.

• 74% of consumers have quietly stopped doing business with a company after a single frustrating service experience.
• 70% have abandoned a service interaction because of difficulty switching channels.
• 69% say it is very or extremely important that AI and human agents work together.
• 87% say trusting a company to protect their data while delivering fast, human service is essential or very important.

For years, chatbots were presented as the answer to one of customer service’s biggest challenges: how to deliver faster support at lower cost without adding more pressure to human agents.

The promise was simple. Customers would get instant answers. Companies would reduce call volume. Agents would focus only on the most complex work.

But for many organizations, the reality has been more complicated.

Chatbots often work well for simple, repeatable tasks: checking an order status, resetting a password, answering a frequently asked question, or routing a customer to the right department. The problem begins when the customer’s need does not fit neatly into a predefined path.

A customer may have a multi-step issue. They may be frustrated. They may need an exception. They may be asking a question that depends on their history, their account, their eligibility, their current situation, or the policies of the business.

That is where many traditional chatbots break down.

They answer, but they do not always understand.
They respond, but they do not always resolve.
They deflect, but they do not always serve.

The data shows why this matters. In a recent US consumer customer experience survey, 83% said speaking with a human agent is very or extremely important when contacting a business with an issue. At the same time, 56% said they are satisfied with an automated assistant if it solves the issue quickly. Customers are not rejecting automation. They are rejecting automation that fails, stalls, or traps them.

For enterprises, the cost is bigger than one bad interaction. Every service moment can also be a loyalty moment, a retention moment, or a sales moment. The same survey found that 76% of consumers have chosen one brand over another because of a better customer service experience, and 74% have quietly stopped doing business with a company after a single frustrating service experience without complaining.

That is the real risk of poor automation. It does not always generate a complaint. Sometimes the customer just leaves.

The next phase of customer service requires more than a better chatbot. It requires a better architecture.

What is the Difference Between a Traditional Chatbot and Agentic AI?

A traditional chatbot is usually designed to respond within a defined set of rules, intents, scripts, or knowledge sources. It can answer questions and complete simple tasks, but it often struggles when the customer journey requires reasoning, planning, personalization, or action across multiple systems.

Agentic AI is different. It is designed to work toward a goal. In a customer service environment, an agentic AI system can help interpret intent, evaluate context, determine next steps, use approved tools, and adapt when the path changes.

That does not mean AI should operate without oversight. For enterprise customer experience, the goal should be governed autonomy, not uncontrolled automation.

A better way to think about the difference is this:

A traditional chatbot helps answer the question in front of it.
 Agentic AI can help determine what needs to happen next.

That distinction matters because customer service is rarely just about information. It is about action. It is about helping the customer move from problem to resolution.

And resolution is what customers care about most. When asked what matters most when contacting a business, consumers ranked accuracy of information first, ahead of speed, politeness, access to a human, and multiple contact options. In other words, speed matters, but being fast and wrong is not a customer experience strategy.

To serve customers well, AI needs access to more than language. It needs governed access to context, systems, workflows, policies, and human support.

What Role does MCP Play in AI-powered Customer Service?

Model Context Protocol, or MCP, is an emerging open standard that helps AI systems connect more consistently to external tools, data sources, APIs, and user context.

In practical terms, MCP can help solve one of the biggest problems in customer experience: disconnected context.

A customer may start in self-service, move to chat, escalate to voice, speak with an agent, receive a follow-up message, and later return through another channel. Too often, each step feels like a separate interaction.

That is exactly the experience consumers say they do not want. Nearly 71% said it is very or extremely important to switch between channels without repeating information, and 70% said they have abandoned a customer service interaction because of difficulty switching channels.

MCP helps create a more standardized way for AI systems to access the context needed to support those journeys. That context might include CRM data, knowledge articles, customer history, workflow status, eligibility rules, previous interaction details, or approved enterprise tools.

MCP does not magically connect everything by itself. It is not a complete customer experience strategy on its own. It is one important layer in a broader architecture.

That broader architecture needs orchestration.

Why Does Orchestration Matter More than Automation Alone?

Automation is powerful when the task is simple and the path is clear.

But enterprise customer service is rarely that simple. Large organizations often operate across many channels, systems, vendors, data sources, and regulatory requirements. A single customer journey may involve identity verification, account history, billing systems, CRM data, knowledge management, workforce tools, compliance rules, and agent-assist technology.

That is why orchestration matters.

Orchestration determines how the right AI model, human agent, partner technology, business rule, and enterprise system come together in the moment. It helps decide what should happen, when it should happen, and which system or person should take the next step.

The survey data reinforces this need. 83% of consumers either expect human agents to know their history with a company or say it would be helpful, and 70% say the same about AI support agents. Even more importantly, when customers are transferred from an AI assistant to a human agent, 70% say it is very or extremely important that the human agent already knows their context.

That is the standard customers are setting. They do not want disconnected channels. They do not want to repeat themselves. They do not want a bot that forgets what happened five seconds ago. They want the company to operate as one connected experience.

For enterprise buyers, that changes the question.

The question is not simply, “Which chatbot should we use?”

The better question is, “How do we orchestrate AI, people, data, workflows, and governance across the customer journey?”

What is Human-AI Tandem Care?

Human-AI tandem care is a customer service model where AI and human agents work together across the same journey. Sometimes AI leads. Sometimes the human leads. Often, the best experience requires both.

This approach is important because not every customer interaction should be fully automated.

A billing question may be easy to automate.
A password reset may be easy to automate.
A routine appointment change may be easy to automate.

But many interactions require judgment, empathy, negotiation, reassurance, or policy interpretation. In those moments, the customer may not want a faster bot. They may want a capable human being supported by better context and better tools.

Consumers appear to agree. 69% said it is very or extremely important that AI and human agents work together so technology helps people help them better instead of replacing them. When including those who said it is somewhat important, that number rises to 90%.

That is the heart of the human-AI service model.

The most effective customer experience strategy does not force every interaction into self-service. It uses AI to improve the entire journey.

AI can help gather information, summarize the issue, recommend actions, retrieve knowledge, prepare offers, execute approved workflows, and handle follow-up tasks.

Human agents can then focus on what humans do best: understanding nuance, building trust, handling exceptions, and helping customers feel heard.

How Can Agentic AI Improve Self-service Without Removing Human Judgment?

Agentic AI can make self-service more useful by helping it move beyond static answers. With the right governance, integrations, and business rules, an AI-powered service experience can help:

• Understand the customer’s goal, not just the customer’s words
• Use approved enterprise knowledge instead of relying on generic responses
• Access relevant context from CRM, order, billing, or case management systems
• Recommend next steps based on policy and customer history
• Escalate to a human agent when the situation becomes complex, emotional, high value, or high risk
• Preserve context so the customer does not need to start over
• Execute approved tasks after the customer or agent confirms the path forward

The key phrase is “with the right governance.”

Enterprise AI should not be treated as a free agent making independent business decisions. It should operate inside approved policies, workflows, permissions, and audit controls. Especially in regulated industries, AI needs clear boundaries around what it can recommend, what it can execute, what requires review, and what must be handled by a human.

Consumers care deeply about that balance. 87% said it is absolutely essential or very important that they can trust a company to protect their personal data while still giving them fast, helpful, and human customer service.

That is where an orchestration platform becomes essential. The future is not just AI that can act. It is AI that can act within the right controls.

Example: AI and Human Agents Working Together in a Telecom Service Journey

Imagine a customer contacts a telecom provider because their internet performance has declined.

In a traditional chatbot experience, the customer might be asked a series of scripted troubleshooting questions. If those steps fail, the customer may be transferred to a human agent and asked to explain the problem again.

In a more orchestrated, human-AI tandem model, the journey could work differently.

A virtual assistant greets the customer, verifies identity, confirms intent, and collects basic diagnostic information. Avaya Infinity helps orchestrate the workflow across the relevant systems and tools. An agentic AI capability, operating within approved permissions and integrations, reviews network diagnostics, account data, service history, and available plan information.

If the issue appears to be technical, the system can help route the case toward resolution. If the data suggests the customer’s current plan no longer fits their usage patterns, the AI can prepare a policy-approved explanation and surface eligible options.

At that point, the next step depends on the customer and the business rules.

If the customer is comfortable, the virtual assistant may present the options. If the customer hesitates, expresses frustration, or appears to need guidance, the interaction can move to a human agent with the full context preserved.

The agent does not start from zero. The agent sees the issue, the diagnostic path, the customer history, and the recommended options. The agent can explain trade-offs, answer questions, apply judgment, and, where authorized, offer approved incentives or discounts.

Once the customer chooses a path, AI can help execute the next steps: updating the CRM, initiating the order, confirming the change, and documenting the interaction.

The value is not that AI replaces the agent.

The value is that the customer, agent, AI, and workflow stay connected.

Example: Governed AI Support in Regulated Industries

The same human-AI tandem model can also apply in regulated environments, but the governance requirements become even more important.

Consider healthcare. Consumers may be open to AI for certain routine tasks, but they strongly prefer humans for sensitive or high-stakes moments. In the survey, consumers preferred humans over AI for discussing symptoms or diagnosis by a wide margin, with 80% preferring a human. They also preferred humans for mental health or emotional support, with 80% choosing a human, and for questions about test results, with 74% choosing a human.

That does not mean AI has no role in healthcare service. It means AI needs to be applied carefully.

AI may be appropriate for scheduling, reminders, follow-ups, intake, summarization, knowledge retrieval, or helping agents prepare for the conversation. But when interactions involve diagnosis, symptoms, emotional support, insurance complexity, or personal risk, human judgment and governance become essential.

The same principle applies in financial services. When asked about serious financial matters such as fraud or claims, 86% of consumers said they always or usually prefer human interaction.

In a governed model, the journey starts with consent, identity verification, and policy-approved data access. AI can help gather relevant information, retrieve approved knowledge, and summarize available options. A human agent, licensed representative, or specialist can remain involved where judgment, compliance, disclosure, or customer reassurance is required.

If consented data sources are used, they must be handled according to approved business rules, regulatory requirements, privacy policies, and access controls. AI can assist the process, but sensitive decisions should remain governed by the organization’s compliance framework.

In this model, AI helps reduce friction and improve personalization. Humans provide oversight, empathy, and accountability.

That is the balance enterprises need.

Why Personalization Must be Connected to Trust

Personalization is no longer a nice-to-have. It is becoming part of what customers expect from service experiences.

The survey found that 69% of consumers always or usually notice when experiences feel personalized, and 92% say real-time personalization based on history and preferences across systems is at least somewhat important. Similarly, 92% say it is at least somewhat important that customer support experiences feel tailored to them.

But personalization without trust can backfire.

Customers want companies to remember context, but they also want to know their data is protected. They want AI to help, but not to overstep. They want fast answers, but not careless ones. They want automation, but not abandonment.

That is why AI-powered customer experience must be designed around three connected principles:

Context: Does the system understand the customer’s situation?
Control: Does the organization govern what AI can access, recommend, and execute?
Continuity: Can the journey move across AI, humans, and channels without losing the thread?

When those three principles come together, personalization becomes useful instead of invasive.

Why Choice of AI Matters for Enterprise Customer Experience

Many organizations are experimenting with AI in customer service. The challenge is that experimentation can quickly create fragmentation.

One team may use one AI model.

Another may use a different bot.
Another may use a separate analytics tool.
Another may rely on a different CRM or knowledge platform.

Without orchestration, each tool may improve one piece of the journey while making the overall experience harder to manage.

“Choice of AI” AI orchestration gives enterprises more flexibility. It allows them to choose the right AI models, partner tools, and deployment approaches for their business while maintaining a more consistent experience across channels and workflows.

That flexibility matters because customer behavior is already changing. In the survey, 47% of consumers said they had used ChatGPT in the past three months to search for information, ask questions, or get help online, while 28% had used Google Gemini and 16% had used Microsoft Copilot.

Customers are becoming more comfortable with AI in their daily lives. But their expectations are also becoming sharper. They know when AI is helpful. They also know when it is blocking them from getting help.

That is why enterprises need an AI strategy that is not tied to a single model, bot, or channel. They need an orchestration strategy that can adapt as models, tools, customer expectations, and compliance requirements evolve.

What This Means for Contact Centers

For contact centers, the next phase of AI is not just about increasing containment. It is about improving resolution, trust, efficiency, and customer value.

Agentic AI, MCP, and human-AI orchestration can help contact centers move toward:

• Better first contact resolution
• More useful self-service
• Faster escalation when a human is needed
• Less repetition for customers
• More informed agents
• More consistent policy application
• Stronger governance over AI actions
• Better service-to-sales and retention opportunities
• More connected customer journeys across channels

The consumer data points to a clear operating model. Customers want automation when it works. They want humans when the issue is complex, emotional, regulated, or high-stakes. They want context to move with them. They want personalization, but not at the expense of trust. And they are willing to leave brands that fail them, often without saying a word.

The goal is not to automate every interaction. The goal is to design every interaction around the right balance of automation, human judgment, context, and control.

Some moments should be self-service first.
Some moments should be human first.
Many moments should be AI and human together.

That is where customer service is heading.

The Bottom Line

Chatbots fell short because many were designed to answer questions, not resolve journeys.

Agentic AI can help close that gap, but only when it is connected to the right context, governed by the right rules, integrated with the right systems, and supported by human judgment when the stakes are high.

The research makes the mandate clear: consumers want fast service, but they also want accurate information, human access, channel continuity, personalization, data protection, and AI that helps people help them better.

MCP helps standardize how AI systems access tools, data, APIs, and context. Avaya Infinity helps orchestrate those capabilities across customer journeys, human agents, enterprise workflows, and an open ecosystem of AI and technology partners.

The future of customer service will belong to organizations that stop treating AI as a standalone bot and start treating it as part of a governed, connected, human-aware customer experience architecture.

That is how self-service becomes real service.

Obtain more information about MCP for customer experience.

Obtain information about Avaya Infinity.