Artificial intelligence is transforming customer experience (CX), giving businesses new opportunities to improve service quality, increase efficiency, and deliver faster, more personalized support. As organizations explore AI agents and automated customer service solutions, the challenge is no longer simply adopting the technology. The real challenge is scaling it successfully across the enterprise while maintaining customer trust, operational consistency, and measurable business value.
Research highlights the potential benefits of AI in customer support. A National Bureau of Economic Research study involving 5,179 customer support agents found that access to a generative AI assistant increased the number of issues resolved per hour by an average of 14%, with productivity gains reaching 34% for novice and lower-skilled workers. These findings demonstrate how AI can support employees, improve efficiency, and help organizations deliver better service.
However, successful implementation is not guaranteed. The Consumer Financial Protection Bureau has warned that limitations in customer service chatbots can lead to negative consumer outcomes and may expose financial institutions to compliance risks, customer harm, and declining trust. These concerns highlight the importance of responsible deployment, strong governance, and reliable systems.
In discussions featured on the AI in Business podcast hosted by NiCE Chief Operating Officer Arun Chandra and Simplyhealth Customer Services Director Paul Cuglietta shared insights into what businesses need to move agentic AI beyond isolated experiments and into large-scale customer experience operations. Their perspectives emphasize four essential priorities: selecting use cases based on measurable financial outcomes, connecting enterprise knowledge and systems, testing AI with real customer conversations, and building teams that combine technical expertise with frontline experience.
1. Align AI Investments With Measurable Business Outcomes
One of the most important steps in scaling agentic AI is establishing a clear business objective before choosing which processes to automate. Organizations may feel pressure to adopt AI because competitors are doing so or executives want to demonstrate progress. However, implementing technology without a defined purpose can lead to expensive experiments that fail to deliver meaningful results.
Arun Chandra recommends applying traditional operations management principles to AI initiatives. Businesses should determine whether a project is intended to increase revenue, reduce operating costs, or accomplish both. Once the objective is clear, leaders can identify the workflows that offer the greatest potential impact, determine the data and systems required, and continuously measure performance.
Financial accountability is particularly important when moving from small pilots to enterprise-wide deployments. While an organization may reasonably invest in an experimental project to understand what AI can accomplish, a larger rollout must demonstrate how the investment contributes to the company’s financial performance.
Leaders must also account for the operating costs of AI itself. Automation does not automatically translate into net savings simply because fewer human hours are required. AI agents consume computing resources and tokens, creating expenses that must be included in the business case.
Customer satisfaction improvements should also connect to broader business objectives. Higher satisfaction scores may indicate a better customer experience, but executives need to understand how those improvements affect retention, revenue, operating expenses, or other measurable outcomes.
At Disney, Chandra approached customer experience as a business function capable of generating value rather than simply reducing contact center costs. His approach included setting clear objectives for revenue opportunities, customer satisfaction, and operating costs. The goal was to create a model in which the business could serve a growing customer base without allowing expenses to rise at the same rate.
For organizations that lack their own AI performance data, Simplyhealth’s Paul Cuglietta recommends gathering evidence from comparable implementations. Businesses should ask potential technology partners about their experience with organizations of similar size and customer demographics, along with the results those implementations achieved.
AI investments should also support the company’s long-term strategy. By connecting the technology roadmap to broader business goals, organizations can build a stronger case for executive support and avoid pursuing projects that produce short-term excitement without sustainable value.
2. Connect Enterprise Knowledge, Data, and Systems
Even promising AI pilots can struggle when organizations attempt to expand them across multiple departments, customer channels, and business processes. This transition often exposes weaknesses in data quality, knowledge management, system integration, and decision-making structures.
For AI agents to deliver reliable results, they need more than access to large amounts of information. They must be able to retrieve accurate knowledge, understand the context of a customer interaction, and determine what actions they are authorized to take.
Enterprise knowledge must therefore be organized, current, and accessible. If policies, procedures, customer records, and operational guidance are fragmented across disconnected systems, an AI agent may provide inconsistent answers or fail to complete a request.
Context is equally important. Agents need to understand relevant business relationships, responsibilities, and decision-making authority. Without clear boundaries, they may struggle to determine which actions they can perform independently and which require human approval.
Connectivity to back-end systems is another essential requirement. An AI agent that can answer questions but cannot securely access the systems needed to update an account, process a request, or retrieve a customer’s latest information may offer only limited value. Meaningful automation requires the ability to carry out appropriate actions within approved systems.
A connected customer experience also depends on consistency across communication channels. Customers may contact a company through telephone calls, email, or live chat, expecting each interaction to reflect the history of their previous conversations. When channels operate independently, customers may have to repeat information, receive conflicting answers, or experience unnecessary delays.
Cuglietta recommends using customer data to identify the most frequently used channels and prioritizing integration accordingly. Organizations do not necessarily need to connect every system before launching an AI initiative, but they should develop a practical roadmap that addresses the most important customer journeys first.
Technology leaders and customer service teams should collaborate closely during this process. Their combined perspectives help ensure that integration decisions support both technical requirements and real customer needs.
Architecture choices also influence how easily AI solutions can expand. Chandra supports open systems and open protocols, including the Model Context Protocol (MCP), as potential ways to improve interoperability. He also expects open-source large language models to contribute to competitive pressure on AI operating costs.
Nevertheless, organizations must evaluate their existing architecture before deciding where open-source, proprietary, and other technologies belong. Integration capabilities, workflow requirements, privacy, compliance, and legal obligations must all be considered. These issues are particularly important in regulated industries such as financial services and healthcare.
Businesses should understand their own requirements before selecting vendors and ask specific questions about how a proposed solution will operate within their environment. A technology partner should be able to explain how its platform supports the organization’s workflows, security requirements, and long-term scaling plans.
3. Test AI Agents With Realistic Customer Conversations
Launching an AI agent before it is ready can damage customer confidence and discourage people from using the channel again. Customers expect automated systems to understand their questions, provide accurate answers, and make it easy to reach a human when necessary.
One common mistake is designing agents primarily around frequently asked questions. In actual conversations, customers rarely use precisely the same wording as a prepared FAQ. They may combine several questions, use unfamiliar expressions, change topics, or communicate differently depending on whether they are typing or speaking.
As a result, AI testing must reflect the complexity of real customer interactions rather than relying exclusively on ideal scenarios.
Cuglietta recommends extensive user acceptance testing using sandbox scenarios and real data. Teams should evaluate whether agents answer questions correctly, communicate in a way that reflects the brand, maintain the organization’s preferred tone, and avoid generating unsupported information.
Testing should take place through small, methodical experiments that allow teams to identify problems and make improvements before expanding the deployment. Organizations need sufficiently varied examples and appropriate sample sizes to understand how an agent behaves across different situations.
Accuracy alone is not enough. Businesses must also determine whether the agent handles sensitive conversations appropriately and recognizes when human assistance is necessary.
For example, a routine request for account information may be suitable for automation, while a customer reporting the death of a family member may require a more compassionate, human-led interaction. The right approach depends on the circumstances, the customer’s needs, and the organization’s service standards.
Governance should guide every stage of deployment. Cuglietta describes using an internal AI governance framework that requires stakeholder approval and documented decisions. Maintaining records of approvals supports accountability and provides an audit trail for the solution being introduced.
Organizations should establish clear expectations for agent behavior, define escalation procedures, and identify who is responsible for monitoring performance. These safeguards help ensure that speed and efficiency do not come at the expense of customer trust.
Businesses also need to decide which use cases to tackle first. Some organizations begin with simpler workflows involving fewer systems and less complicated data requirements. Others choose a more significant business problem that exposes the challenges they will eventually face at scale.
Chandra favors taking on meaningful problems with appropriate executive backing because doing so can reveal important operational gaps early. However, either approach requires careful planning, suitable partners, and a willingness to learn and improve throughout implementation.
4. Build Hybrid Teams to Support AI Adoption
Technology alone cannot guarantee successful AI adoption. Organizations must also prepare employees for changes in responsibilities, workflows, and customer interactions.
Simplyhealth addressed this challenge by creating an optimization team that brought together technical specialists and frontline customer service employees. The team combined technology knowledge with the practical experience of advisors who understood customer concerns, common service issues, and the realities of daily operations.
This structure supported business readiness, technical deployment, and the ongoing optimization of AI solutions. It also helped bridge the gap between the people building the technology and the employees who would use it.
Hybrid teams can be particularly valuable because frontline workers often recognize problems that may not be obvious during technical development. Their feedback can help organizations improve conversational flows, identify missing information, refine escalation rules, and ensure that automated processes reflect actual customer needs.
Change management is another essential part of the process. Leaders need to explain how AI will affect employees’ roles, which tasks will be automated, how human handoffs will change, and what new skills workers may need to develop.
Employee attitudes toward AI can vary. Some customer service advisors may welcome the technology because they can see how it will simplify their work. Others may need more time, support, and practical evidence before they understand its benefits.
Organizations should address these concerns through communication, training, and employee involvement rather than treating adoption as a purely technical exercise.
The objective should be to help employees focus on the work where human capabilities are most valuable. AI agents may handle suitable routine inquiries and repetitive tasks, while human advisors concentrate on complex problems, emotional conversations, and customers who need additional support.
Empathy, compassion, judgment, and relationship-building remain important elements of customer service. When automation is implemented thoughtfully, it can help employees dedicate more time to these responsibilities instead of replacing the human connection customers may need.
External technology partners also play an important role. Technical capability matters, but organizations should evaluate whether a vendor is willing to collaborate with internal teams, share knowledge, and help develop a practical implementation roadmap. Long-term success depends on cooperation between the business, its employees, and its technology providers.
5. Create a Sustainable Strategy for Enterprise-Wide AI
Moving from a successful pilot to enterprise-wide adoption requires an integrated strategy that connects financial objectives, technical readiness, governance, and workforce preparation.
Organizations should begin by identifying the business outcomes they want to achieve and establishing measurable indicators of success. These indicators should account for revenue, operating costs, customer experience, and the full cost of running AI agents.
Next, businesses need to prepare the foundations that make reliable automation possible. This includes organizing enterprise knowledge, improving data quality, connecting essential systems, clarifying decision rights, and establishing appropriate privacy and compliance safeguards.
Before an agent is released to customers, it should undergo structured testing that reflects realistic conversations and sensitive situations. Clear human-handoff procedures and governance requirements should be in place, with ongoing monitoring to identify areas for improvement.
Finally, companies must invest in the people responsible for implementing and using AI. Cross-functional teams, employee training, and close collaboration between technical specialists and frontline advisors can help turn the technology into a practical business capability.
These efforts should continue after launch. Customer expectations change, business processes evolve, and AI systems require regular evaluation. Continuous measurement and optimization allow organizations to address emerging problems and adapt their solutions as requirements develop.
Conclusion
Agentic AI offers significant opportunities to transform enterprise customer experience, but scaling it successfully requires more than selecting a capable AI platform. Businesses must connect technology investments to measurable financial results, prepare their data and knowledge, integrate essential systems, test agents against realistic customer interactions, and establish strong governance.
The experiences shared by Arun Chandra of NiCE and Paul Cuglietta of Simplyhealth demonstrate that effective AI adoption depends on the combination of technology, business strategy, and human expertise. Financial accountability helps organizations justify investment, connected systems enable more complete automation, rigorous testing protects customer trust, and hybrid teams make implementation more practical and sustainable.
Most importantly, AI should be introduced with a clear understanding of where automation creates value and where human support remains essential. By taking this balanced approach, organizations can move beyond isolated experiments and build customer experience operations that are more efficient, consistent, and responsive while preserving the empathy and accountability customers expect.
The path to enterprise-wide AI success is not simply about automating more tasks. It is about designing intelligent workflows that serve business goals, empower employees, and deliver meaningful improvements for customers.