This article is sponsored by NiCE and was written, edited, and published in alignment with our Emerj sponsored content guidelines. Learn more about our thought leadership and content creation services on our Emerj Media Services page.
Agentic AI wins in enterprise customer experience stall at the point of scale, often held back by fragmented systems, unprepared knowledge, and inconsistent governance.
A National Bureau of Economic Research study of 5,179 customer support agents found that access to a generative AI assistant increased the number of issues resolved per hour by 14% on average and by 34% for novice and low-skilled workers.
However, the risks of scaling poorly are visible as surfaced in a 2023 Consumer Financial Protection Bureau report, which notes that each of the 10 largest U.S. commercial banks has deployed chatbots in customer service, and that some customers experience significant negative outcomes because of the technology’s limitations. The CFPB warns that institutions deploying chatbots risk violating legal obligations, eroding customer trust, and causing consumer harm.
Emerj recently hosted Arun Chandra, Chief Operating Officer at NiCE and Paul Cuglietta, Customer Services Director at Simplyhealth on the AI in Business podcast in a series unpacking what it takes to turn enterprise knowledge into production-ready AI agents for enterprise-wide CX transformation at scale.
This article examines the key insights CX leaders can apply as agentic AI moves from isolated pilots to enterprise-wide production:
- P&L-anchored use case selection to secure CFO investment: Decide whether each deployment drives revenue or hard-cost savings before choosing use cases, and build token costs into the business case so gains hold up under financial scrutiny.
- Connected knowledge and systems for autonomous agent workflows: Organize enterprise knowledge, decision rights, and system-of-record access so agents resolve customer requests end to end instead of always handing them back to humans.
- Real-conversation testing before channel launch: Validate accuracy, brand tone, and human-handoff triggers under formal governance, since customers rarely give a new channel a second chance.
- Hybrid optimization teams as an adoption accelerator: Pair technologists with frontline advisors to own readiness, deployment, and tuning, and redirect human talent toward complex, high-empathy interactions.
Listen to the full episodes below:
Episode 1: Scaling Agentic Automation With Open Architecture – with Arun Chandra of NICE
Guest: Arun Chandra, Chief Operating Officer at NiCE
Expertise: AI-Driven Enterprise Transformation, Global Operations, Customer Experience, Corporate Strategy
Brief Recognition: Arun Chandra is Chief Operating Officer at NiCE, following senior executive roles at Disney, Meta, and Hewlett Packard Enterprise. At Disney, he modernized customer experience for streaming services reaching more than 200 million subscribers, delivering tens of millions of dollars in P&L impact through automation and operational restructuring. At Meta, he built global shared services across 45 locations, driving double-digit efficiencies across a multibillion-dollar operating budget. At HPE, he served on the corporate Operating Committee and chaired H3C, its multibillion-dollar joint venture in China. Chandra holds an M.S. in Computer Science from The Ohio State University and an MBA from Santa Clara University.
Episode 2: Inside the Shift Toward Smarter Automated CX – with Paul Cuglietta of Simplyhealth
Guest: Paul Cuglietta, Customer Services Director at Simplyhealth
Expertise: Customer Experience, Contact Center Operations, Claims Management, Customer Service Strategy
Brief Recognition: Paul Cuglietta is Customer Services Director at Simplyhealth, where he has spent a decade leading customer service, contact center, claims, and operational teams. Previously, he led customer service and business development at OPP, overseeing sales and service teams, European partner relationships, and customer experience strategy. His career spans more than 20 years across insurance, healthcare, finance, and learning and development, with a focus on building high-performing teams and improving service delivery.
P&L-Anchored Use Case Selection to Secure CFO Investment
Arun Chandra has watched organizations rush into AI pilots over the past 12 to 18 months because vendors were calling with solutions, or the C-suite wanted to know what they had done with AI. His alternative is what he calls classic operations management applied to AI:
- Set the objective: Decide whether the program drives higher revenue, lower costs, or both.
- Pick the use cases: Choose the most critical, impactful workflows to automate toward that objective.
- Map the dependencies: Identify the data, knowledge, context, and back-end systems those workflows need.
- Measure continuously: Track ROI on an ongoing basis.
In Chandra’s view, only two numbers convince a CFO: revenue gains and hard cost savings. He cautions that automation savings aren’t net savings, because leaders assume they’ve reduced human costs without accounting for what the agents cost to run:
“One thing we can maybe segue into is the cost of tokens. AI is not free. We’re talking about, AI can automate something, and so we assume that we’ve reduced the cost of the humans, but wait a minute, the AI agents have their own operating cost, which is in token costs.”
— Arun Chandra, Chief Operating Officer at NiCE
Even a CSAT score that climbs from 90% to 99% has to show up on the P&L, because someone will ask what it translates into. Chandra separates experimentation from scale:
- Sandbox pilots: Spending half a million or a million dollars to test is reasonable.
- Company-wide rollouts: In a large organization, these must connect to the P&L. A CEO who tells the street on an earnings call that the company invested $500 million in AI tokens will be expected to say what it got in return.
At Disney, where he modernized the customer experience function for the streaming business, Chandra approached CX as a general manager asking what business value the function could create, rather than as a contact center manager focused on handle time. He set a vision for CX to be a value-creating center, beyond a cost center, with clear objectives for:
- Revenue: How much cross-sell and upsell the function would deliver.
- Customer Experience: The CSAT scores it would reach.
- Cost: The cost structure it would operate within.
His ultimate goal was to bend the cost curve, so that as subscribers grow, the cost structure flattens out. Chandra says getting there took a clear vision and execution plan, followed by disciplined execution over many months, with course corrections along the way, because “you will learn so much in this process.”
For leaders who don’t yet have results of their own, Paul Cuglietta offers a way to build the case before the first deployment:
- Comparable proof: Ask potential partners when they have worked with an organization of similar size and customer demographic, and what results they achieved.
- Strategic alignment: Link the AI vision to the company’s purpose and strategic aims, looking three to ten years ahead rather than only the next 12 months.
A vision tied to strategy and backed by a like-for-like case study, he argues, gives leaders a strong chance of winning internal support.
Connected Knowledge and Systems for Autonomous Agent Workflows
Between AI pilots and enterprise-scale deployment lies what Arun Chandra calls a “valley of death,” and he says most efforts fail there operationally. The problems surface once an organization moves past the pilot and pushes for enterprise-wide scale. At that point the data has to be clean, and two further inputs become just as critical:
“More importantly, or as importantly, my knowledge has to be organized and available, because that knowledge and data is the lifeblood of what AI can produce for you in terms of outcomes. And then the third related thing is context also. So context has to be pretty well provided so that the AI can understand who reports to whom, what decision rights do I have versus what decision rights you have.”
— Arun Chandra, Chief Operating Officer at NiCE
His second requirement is connectivity into the back-end systems of record, so agents can take actions autonomously without involving a human.
Paul Cuglietta sees the same requirement from the customer’s side. A customer may call, email, and use live chat. Each channel needs to follow the same thread so the customer gets a consistent response and never has to repeat themselves. Joining channels up, through an API link or otherwise, isn’t easy and depends on how the organization’s architecture is set up. Cuglietta says the difference it makes to the experience is material.
Some of his channels are connected, and others are still being built out. His advice is not to wait until everything is connected before going live, because that means waiting too long. Instead, he recommends using data to identify the most popular channels and how to connect them. He works hand in glove with the chief technology officer so both sets of priorities shape what gets built.
Architecture choices determine how far that connectivity can extend. Chandra is a firm believer in open systems and in open protocols such as Model Context Protocol (MCP), which his own team uses both internally and externally.
SaaS companies developed vendor lock-in, he notes, and AI is opening that up. He expects strong pressure against vendor lock-in, and he expects open-source LLMs to drive down token costs across the industry. Before committing to open or proprietary components, though, he says leaders need to:
- Fully understand the enterprise architecture before deciding where open and proprietary systems belong.
- Confirm that the biggest workflows will support that choice and that the interconnectivity is there to scale.
- Make sure privacy, compliance, and legal requirements are covered across the open architecture. Chandra calls this the number one consideration for large organizations, particularly in regulated industries such as financial services and healthcare. If those requirements can’t be managed, he says, the architecture doesn’t work for anybody in the company.
- Ensure the organization can interconnect across the organization using MCP for its own use cases.
To test vendors who claim to be open, Chandra advises coming in knowing your own architecture, needs, and use cases, asking pointed questions about what your environment requires, and refusing to accept high-level hand-waving in place of an answer.
Real-Conversation Testing Before Channel Launch
Paul Cuglietta’s case for a measured rollout rests on one observation:
“From my experience, you get one chance to get a new channel right. Otherwise customers may switch off to that channel and think, you know what, I think I’ll just stick to utilising the telephone or utilising another channel that I’m more familiar with. So what we don’t want to do is to get it wrong. And I think sometimes companies fall foul of wanting to evolve quicker than perhaps they’re ready.”
— Paul Cuglietta, Customer Services Director at Simplyhealth
One challenge he has heard from peers is building agents around frequently asked questions. Customers rarely phrase a question exactly like an FAQ. They use a variety of words and language, ask multiple questions in one go, and on voice channels speak in a different tone.
Cuglietta’s approach is a significant set of user acceptance testing on both sandbox scenarios and real data, worked through methodically in very small tests and learns. The tests check whether the agent:
- answers the customer’s question,
- represents the brand the way the organization wants,
- uses a tone consistent with how the organization speaks to customers,
- avoids hallucinations or anything other than the intended answers.
Cuglietta sets the bar at 100% certainty that the agent is acting the way the company wants, and building that certainty takes time, data, and adequate sample sizes.
Above cost and other concerns, Cuglietta says the organization’s risk appetite sets the pace. Everything new goes through an internal AI governance framework and needs sign-off from all stakeholders. Those meetings are minuted and recorded for audit, so everyone is on record signing up to what will be deployed.
Cuglietta’s second consideration is the customer experience itself. His team asks whether it is comfortable with a customer interacting with AI in a given scenario. In most situations, the answer is yes. A customer reporting that their next of kin has passed away is different: the team may wish to move from the agent to a human much more quickly.
Arun Chandra adds a perspective on what to put through that process first. One school of thought starts with simple, contained use cases that involve fewer interconnected systems and less data cleansing, and builds momentum by showing success. The other takes on the hardest, most meaningful problem for the company, backed by a coalition across the organization and executive support.
Chandra recommends the second, because it surfaces the problems a company will face when scaling, so the gaps and needs show up early on. If those can be overcome, the path to enterprise scale is faster and smoother. He leaves the choice to each organization and adds that the partners it selects matter, because they help guide the way.
Hybrid Optimization Teams as an Adoption Accelerator
To execute its AI deployments, Paul Cuglietta’s organization dedicated an internal team, which it called the optimization team, drawn from both sides of the business:
“This team was made up of people who were, from a technical background in our organization, but also from my front line as well. So those on the front line that had the knowledge, they had the passion, they had the career ambitions to move into a slightly more technical role. Bringing those two together, technology and service, into an optimization team, they helped execute the AI deployments in our organization.”
— Paul Cuglietta, Customer Services Director at Simplyhealth
The team covers business readiness, the technical aspects of deployment, and optimizing what the organization already has. Cuglietta says it made a big difference to the organization’s success.
Arun Chandra names change management of the people involved as a key factor in whether AI deployments succeed or fail. In his framing, that means answering three questions:
- Roles: What is the new role of humans in the process?
- Handoffs: What needs to be different in handoffs, and where do they disappear entirely through automation?
- Skills: How will the workforce be reskilled?
He stresses that this applies to AI adoption broadly, including engineering and product management, not only CX. He expects a future in which human and AI agents coexist, with every organization needing some mix of both.
On the service floor, Cuglietta finds that opinion on agentic AI is polarized; some advisors are bought in and see how it will help both customers and themselves. Others are less far along the change curve and don’t yet see the value, and some sit in between.
He aims to make his team’s jobs easier and simpler, so they can focus on what they’re skilled in: soft skills, support, empathy, and compassion. When customers who can self-serve do so, advisors are freed for complex inquiries and vulnerable customers who genuinely need a person.
Cuglietta also judges outside partners on collaboration; a vendor may have the best technical capability, but if its teams won’t collaborate with your own technical teams, he warns, you will run into blockers and barriers. He advises choosing a partner on both technical and people grounds, then building a roadmap together toward the organization’s AI vision.