Contact centers are moving faster down the road to increased automation, where machines are becoming the norm and people the exception in customer interactions. Where technology also supports humans: whose roles are now to handle only what they cannot and as backups, oversight, and overrides.
This journey is being accelerated by AI advancements like agentic AI. It is being driven by the C-suite to cut costs and improve revenue retention-and-growing excellent customer experiences (CX): and not necessarily in that order.
But as organizations are learning (yes, once again), AI, like every other complex tool, also has its downsides that must be investigated. And, most critically, it must be carefully implemented, including thorough integrations with the people, processes, and other technologies.
In the July 2025 issue Feature, our expert panel touched on agentic AI and automation in routing and handling customer contacts. But we dug a little deeper and posed these related questions as organizations gain experience with these technologies.
Our panelists are:
- Kevin McNulty, Senior Director, Product Marketing, Talkdesk
- Sarika Prasad, Director, Product Marketing, Five9
- Neeraj Verma, Vice President, Advanced Technology, NiCE
Q. Does agentic AI have a role in routing customer contacts and if so, in which channels? In enabling more interactions to be handled by automated agents rather than by live agents?
Kevin McNulty:
Agentic AI has a big role to play in routing. But not in the way we’ve traditionally thought about it.
Instead of routing to a static queue or decision tree, agentic AI enables real-time reasoning based on context, intent, urgency, customer history, and business rules: and it does this dynamically, across channels.
With agentic AI it’s not just about “who takes the call” anymore. Instead, it’s about “what combination of AI and human agents can resolve this most efficiently and effectively.”
- In some cases, that may mean handing the interaction off entirely to a team of AI agents. One to authenticate the customer, another to process a refund, a third to update backend systems, and a fourth to confirm completion with the customer.
- In other cases, it might mean an AI copilot supporting a live agent while AI agents work in parallel on backend tasks.
This multi-agent model also means AI isn’t confined to chat or voice. It can orchestrate across both—and beyond—handling asynchronous messaging, email threads, app interactions, and even outbound engagement.
“…when [AI] understands context, remembers past interactions, and resolves issues without friction, [customers] stay.”
—Kevin McNulty
The bottom line is this. Agentic AI doesn’t just make automation possible; it makes it collaborative, continuous, and capable of handling complexity in a way traditional bots never could.
Sarika Prasad:
Agentic AI is no longer a future concept. It’s here and reshaping the contact center.
Moving beyond back-office automation, AI agents—empowered by varying levels of autonomy—are now taking on diverse customer-facing roles. They’re streamlining engagement, accelerating resolution times, and intelligently routing customers to the right destination at the right moment.
Traditional virtual agents are well-suited for handling routine inquiries and repetitive tasks, making them an effective first point of contact.
However, with the advancement of generative AI, today’s AI agents go much further. They can quickly understand the context and intent behind a customer’s issue and determine the most efficient path to resolution across channels.
These AI agents efficiently gather key information, resolve issues autonomously when possible, or seamlessly route the interaction to the most appropriate resource – whether that’s a chatbot or human agent – over voice or digital channels.
This dynamic routing ensures faster resolutions and reduces friction, all while supporting the omnichannel flexibility that modern customers expect.
In fact, we find nearly 60% of consumers’ preferred communication channel depends on the context of the situation, making it critical for companies to offer fluid transitions between touchpoints.
Agentic AI enables just that by learning customer preferences and responding accordingly. It can also be a key tool for human agents in the routing process, ensuring that if a situation escalates further, the agent has all of the necessary information to help the customer, without being repetitive.
In today’s high-volume, high-expectation environment, then AI agents aren’t a “nice to have,” they are a strategic imperative. They empower contact centers to scale efficiently, deliver faster service, and meet customers where they are, all while freeing up live agents to focus on high-value, complex interactions.
Neeraj Verma:
Traditional IVR is being enhanced by agentic AI to better understand and respond to customer intent without human intervention.
These advancements are streamlining routing and resolution for routine queries, significantly lightening the load on live agents and enabling organizations to deliver faster, more efficient service.
I’d argue that “proactive” is where the real transformation is happening when it comes to agentic AI. If you think about the hierarchy of interactions, no business truly wants inbound.
Inbound is almost a symptom; something you aim to reduce, not optimize. When people say, “I want to improve my inbound experiences,” what they’re often really saying is, “I want to create a world where customers don’t need to reach out in the first place.”
Historically, proactive AI has been limited to basic use cases: simple reminders, alerts, or status updates.
But recently we’ve seen a major shift. Both agentic and generative AI are now orchestrating outbound interactions in far more intelligent and meaningful ways. They’re detecting patterns, identifying anomalies, and triggering personalized, conversational outreach that feels more human than ever.
As innovation continues, our focus must remain on using AI to power seamless, proactive, and deeply personalized interactions, raising the bar for what modern customer service can deliver.
Customer Acceptance
Q. AI, like agentic AI, has been heralded as a massively transformative tool. But is AI turning out that way?
Is it becoming like the grocery self-checkouts that customers appear to be avoiding and going back to staff?
Or more like bank ATMs and speech-enabled IVR, which customers have generally accepted, handling the simpler, high-volume interactions while leaving the complex engagements to people?
Kevin McNulty: The problem is that most AI still behaves like a vending machine: limited menus, rigid options, and zero awareness of what came before. It can only respond to what’s pressed, not to what’s meant. Customers aren’t rejecting automation; they’re rejecting bad automation that makes them do the system’s work.
When AI can’t recognize nuance, traps users in loops, or forces them to start over, they seek a human. But when it understands context, remembers past interactions, and resolves issues without friction, they stay.
That’s the turning point agentic AI creates; it replaces the vending machine with a system that can reason, adapt, and deliver outcomes proactively. Once customers experience that kind of competence, they don’t want to go back.
Sarika Prasad: While AI may have initially been met with skepticism in the CX space, with hesitancies that persist among consumers, it continues to be a transformative tool for customer interaction.
In the contact center, agentic AI is taking on the role that ATMs and IVRs once did: handling the high-volume, low-complexity interactions so human agents can focus on the more complex issues that require the judgment and emotion of a person.
“Agentic AI isn’t a ‘set it and forget it’ solution. Success requires ongoing evaluation and improvement. This includes sourcing and applying feedback…” —Sarika Prasad
We continue to take a human-in-the-loop approach to CX to ensure that AI doesn’t become the self-checkout experience that people abandon, but rather as an effective tool that consumers can trust for accurate and efficient service.
Neeraj Verma: AI in contact centers isn’t just about handling repetitive interactions. It is becoming a full-fledged extension of the workforce.
Modern AI can move across the entire service ecosystem, automating complete journeys from routine requests to mid-office approvals and back-office fulfilment, while working seamlessly alongside human employees.
It’s important to remember that AI is not new and AI in the contact center is not new. Rather than focusing on the newest large language model (LLM) or classes of AI, we have been building CX-specific AI solutions for decades.
That means bringing the most appropriate type of AI to deliver the best CX outcomes for consumers. This approach brings together AI memory, knowledge assets, deep industry expertise, the best LLM for each use case, security, scalability, and a comprehensive testing framework.
“Agentic systems need seamless access to fresh, unified knowledge across CRM, workflows, support systems, and historical interactions.” —Neeraj Verma
It allows building solutions that effectively copilot human agents while guiding customers effortlessly through their journeys, meeting the needs of all customers.
Companies that deploy AI solely for cost-cutting often fall short. But when AI is thoughtfully integrated, handling routine tasks while augmenting human capability, it strengthens both customer and employee experiences, delivering real outcomes rather than just responses.
In short, AI is no longer a support tool; it’s a collaborative partner, amplifying human capability, driving operational efficiency, and enabling smarter, end-to-end CXs.
AI Challenges
Q. There have been many reports of AI applications, like agentic AI, not meeting their promises and resulting in fewer benefits than were hoped for. But what are you hearing in the contact center? What challenges in deploying these technologies have they encountered?
Kevin McNulty: The main challenges we see are companies relying on single bots, sometimes even agentic bots, that are simply too limited in scope.
They can handle narrow tasks but when conversations stray from scripts or require coordination across systems, they quickly fall short.
Sarika Prasad: In contact centers, one of the most challenging hurdles to agentic AI adoption that CX leaders must overcome is consumer trust and reluctance.
Despite the promise and advancements of AI, consumers remain cautious. We found in our paper, “2025 Customer Experience Report – Consumer Edition,” that nearly a quarter (23%) of consumers still report they are uninterested in using AI for customer service.
Neeraj Verma: What I’m hearing from conversations with contact center leaders and integrators is that the challenges with agentic AI aren’t about the technology being overhyped. [Instead], they’re about the realities of bringing it into complex environments.
The promise of AI that can reason, act, and learn on its own is exciting. But many organizations are quickly discovering that the foundation underneath (data, integrations, governance) just isn’t ready yet.
First, the data and context infrastructure is often underbuilt.
Agentic systems need seamless access to fresh, unified knowledge across CRM, workflows, support systems, and historical interactions. When those sources are stale, siloed, or disconnected, the AI hits dead end or hallucinates.
[Second], we also see challenges around error compounding and governance.
In a multi-step task, one misinterpreted step early can cascade into costly mistakes downstream. Leaders want to innovate but are cautious about compliance, auditability, and reputational risk. Human oversight isn’t optional; it is core to maintaining confidence as AI takes on more responsibility.
The challenges we’re seeing aren’t failures of AI; they’re lessons in execution. Agentic systems can absolutely deliver value, but only when organizations treat them as part of an evolving ecosystem that blends people, process, and technology.
Recommendations
Q. What are your recommendations to obtain the optimal results from agentic AI, including avoiding and responding to its issues?
Kevin McNulty: The success or failure of agentic AI comes down to data: how it’s organized, shared, and continually refreshed. Too many AI systems are built on brittle integrations and siloed data, which leads to hallucination and drift.
The answer is a living data lake architecture that unifies structured and unstructured data: voice transcripts, chats, CRM records, and knowledge articles into a single accessible fabric.
On top of that, a knowledge mesh ensures AI agents can find the most current and authoritative information, regardless of where it resides. This turns data into a dynamic asset rather than a static warehouse.
Organizations that design for discoverability, traceability, and continuous learning get AI that evolves with the business. Those that don’t end up with digital self-checkouts: technically functional, but easily abandoned.
Sarika Prasad: To address consumer hesitancies, companies have to increase [their] focus on deploying agentic AI tools that are trustworthy, intuitive, and efficient. Additionally, to win over reluctant users, transparent deployment, coupled with compliance and accuracy, is essential.
Agentic AI isn’t a “set it and forget it” solution. Success requires ongoing evaluation and improvement. This includes sourcing and applying feedback from customers and agents to better understand what is working, what isn’t, and what can be improved.
Companies that regularly update and optimize their AI agents can address small issues before they become major problems, [thus] ensuring the technology meets its promise rather than its hype.
Neeraj Verma: To achieve optimal outcomes with agentic AI, organizations should adopt a strategic approach that emphasizes purposeful design, robust training, and continuous oversight.
It starts with purpose-driven design, defining clear objectives and leveraging historical and operational data to train AI agents so they can deliver meaningful impact.
Equally important is human oversight: robust governance frameworks, continuous monitoring, and audit trails ensure AI operates safely and transparently within enterprise standards.
Agentic AI should augment, not replace, human expertise. Seamless collaboration between AI and frontline teams improves both efficiency and CX.
Finally, continuous learning is critical, as AI models must evolve based on real-world feedback and changing business contexts to stay relevant and effective.
