Customers want better service, not more AI

AI must improve the customer experience

Customers care about the outcome of a service interaction. They want a faster answer, less effort, and a complete resolution. AI creates value when it delivers those outcomes.

Chad Anderson, Head of Customer Care & Roadside Relations at Mercedes-Benz USA, put the issue clearly: “Customers aren’t begging for AI, they’re begging for better experiences.” This distinction should shape investment decisions. An AI system that lowers operating costs while adding friction to the customer journey can create a larger commercial problem through weaker trust and loyalty.

Poor automation often moves work back to the customer. People provide information to an AI system, fail to resolve the issue, reach a human agent, and then repeat the same information. The company may have automated part of the interaction while making the full journey longer. For executives, end-to-end customer effort is therefore a more useful design constraint than the percentage of contacts handled by AI.

Mercedes-Benz USA applies three criteria to AI initiatives: save the customer time, save the customer effort, and go above and beyond expectations. Anderson said deployments that fail these tests do not move forward. The framework connects AI investment to customer value and protects a brand whose relationships depend heavily on trust.

Michael Darwal, Chief AI & Digital Officer at ibex, raised a related problem. He cited an MIT study that found a majority of AI implementations fail or fall short of expected results. Darwal linked these failures to weak customer-journey design. Companies often pursue “deflection,” where automation prevents a customer interaction from reaching a human agent, without first understanding what happens to the customer as a result.

This changes the executive question. The relevant target is successful resolution with low customer effort. Cost savings should follow from better process design rather than become the sole design objective.

Treat AI and human service as one operating system

AI does not need to complete an entire customer interaction to create measurable value. Michael Darwal, Chief AI & Digital Officer at ibex, describes customer-service automation as a spectrum. Different parts of the same interaction can be assigned to AI and human agents according to complexity.

That distinction matters. Many service calls begin with predictable work: verifying identity, collecting account information, establishing the reason for contact, and routing the request. AI can complete these steps before a human becomes involved. The agent can then enter the conversation with the customer’s context already available.

Darwal said automating these early tasks can reduce human-agent handle time by 30% or more. This is meaningful operational leverage. It increases agent capacity without requiring AI to solve every customer problem independently. It can also remove repetitive work from the agent’s role.

ibex’s deployment for Philippines Airlines provides a concrete example. The company deployed AI agents in six weeks and supported three languages, including Tagalog and Taglish. The system handled frequent, relatively simple requests such as seat changes and flight modifications. AI-handled interactions achieved a customer satisfaction score of 4.7 out of 5, compared with 4.3 out of 5 for typical human interactions.

Those results support a more precise automation strategy. High-frequency and predictable work is a strong starting point because the required actions and expected outcomes are easier to define. More complex or sensitive cases can move to human agents with the context gathered by AI.

The key technical constraint is the handoff. Authentication, customer intent, information already supplied, and actions already attempted must transfer with the interaction. Without that continuity, customers repeat themselves and much of the efficiency disappears.

Executives should therefore evaluate AI at the workflow level. Full AI resolution is one possible outcome, but it should not become the default measure of success. Handle time, customer effort, resolution quality, satisfaction, and the quality of AI-to-human transfers provide a broader view of business performance. The goal is to place automation where it produces a better customer outcome while making scarce human expertise more productive.

AI makes the remaining human work more complex

As AI absorbs routine requests, human agents receive a higher concentration of difficult cases. This changes the economics and design of the contact center. Training, agent-assistance technology, staffing models, and performance metrics must evolve with the workload.

Chad Anderson, Head of Customer Care & Roadside Relations at Mercedes-Benz USA, expects frontline roles to change as agentic AI handles more routine tasks. “We need to make sure that we’re training them to be able to prepare for that and to enhance their skillsets, because a lot of roles are not necessarily being eliminated, but they are evolving as AI gets more implemented,” he said.

The underlying constraint is agent capability. Complex interactions can involve incomplete information, unusual requests, frustrated customers, and decisions that require judgment. When these cases become a larger share of an agent’s day, conventional training designed around standardized service processes becomes less effective. Companies need deeper product knowledge, stronger problem-solving skills, and clear escalation procedures.

ibex is using AI to support agents during these harder conversations. Michael Darwal, Chief AI & Digital Officer at ibex, said his team has developed desktop tools that listen to calls in real time. The software identifies possible answers and suggests responses before the customer finishes speaking. This reduces the time agents spend manually searching knowledge bases while a customer waits.

“The goal is to be able to utilize that technology to make the job of the agent easier, even on those more complex interactions,” Darwal said. This is an important second use of AI in customer service. Automation can remove routine interactions from the queue. AI assistance can also improve the human handling of cases that remain.

The change has implications for performance management. Genelle Chamberlain, IT Manager at PrimeSource, warned that established contact-center KPIs may become less representative as case complexity rises. Average handle time is an obvious example. An agent handling difficult exceptions may need longer to achieve the right outcome. First-call resolution can also change when the remaining cases require more investigation, coordination, or specialist input.

Agent morale deserves similar attention. Employees accustomed to resolving a mix of simple and difficult requests may eventually spend most of their time on demanding cases. That creates a different work environment. Executives should track workload complexity, quality outcomes, employee experience, and retention alongside conventional productivity measures.

The management priority is clear: redesign the human role at the same time as the automated workflow. Companies that automate routine work without adapting training, tools, and KPIs risk creating a more difficult operating environment for the people responsible for their most challenging customer problems.

Governance determines whether AI can scale safely

Fast AI deployment requires disciplined preparation. Customer journeys must be understood before automation begins. Data access needs explicit controls. AI behavior needs continuous monitoring. Regulatory and brand requirements need to be built into deployment decisions from the start.

Mercedes-Benz USA illustrates the governance demands of a large global company. Chad Anderson, Head of Customer Care & Roadside Relations at Mercedes-Benz USA, described the company’s approximately 140-year brand legacy as a reason for deliberate implementation. Different regulatory environments in the U.S. and Europe add complexity, while customer expectations also vary by market.

“We have to really be cautious and not necessarily rush to do something because it’s the next big thing, but really being intentional to see how it can further enhance the legacy of our brand,” Anderson said.

For C-suite leaders, governance therefore extends beyond compliance. AI-generated errors, inappropriate data access, poor customer experiences, and inconsistent deployment decisions can create operational and reputational exposure. The controls need clear ownership and must persist after a system enters production.

Genelle Chamberlain, IT Manager at PrimeSource, described this challenge through the company’s experience with Microsoft Copilot. “One thing we found with Copilot is it kept pulling from [external sources], even though we set those guards, we set those boundaries, it was still trying to pull outside,” she said.

That experience highlights a practical limitation of static controls. Configuring boundaries is an initial governance action. Teams then need to test whether those boundaries work under real use, monitor unexpected behavior, and adjust access rules as systems and data environments change.

PrimeSource has established a cross-functional AI council with representatives from data, logistics, customer service, and marketing. This structure gives AI decisions input from the business functions that understand the relevant data, workflows, customer consequences, and operational risks. It can also reduce fragmented adoption in which departments establish incompatible practices independently.

Michael Darwal, Chief AI & Digital Officer at ibex, connects governance with deployment speed. His team invests heavily in mapping the customer journey before introducing automation. He said this preparation helped ibex deploy AI agents for Philippines Airlines in six weeks, across three languages, while competitors spent months struggling with hallucinations. In generative AI, hallucination refers to the system producing information that appears plausible but is incorrect or unsupported.

“The goal is to be able to do it both thoughtfully and quickly,” Darwal said. “Anybody can do anything bad really fast, but the goal is to be able to do it once well and getting the best outcome possible.”

The executive lesson is that preparation and speed can reinforce each other. Clear customer journeys, defined data boundaries, testing procedures, regulatory review, and accountable owners reduce the number of problems that surface late in deployment. Governance works best when it is part of system design and operating practice from the beginning.

AI literacy must start with practical, role-specific use cases

Successful AI adoption depends on whether employees can use the technology effectively in their daily work. Access alone does little to develop that capability. Employees need concrete examples tied to tasks they already understand, supported by training that builds confidence through repeated use.

Chad Anderson, Head of Customer Care & Roadside Relations at Mercedes-Benz USA, said the company holds weekly AI meetups. Employees share how they use AI in their roles and discuss developments across the industry. This gives workers a regular setting to learn from practical examples and see how colleagues apply the technology.

Anderson sees specificity as essential. “Unless you’re someone that is very techie or just really interested and curious, you’re not going to proactively dive into AI,” he said. “Leaders need to focus on specific use cases for how their employees can use AI, as opposed to saying, here’s AI, use it.”

The constraint is practical understanding. Employees need to know which tasks AI can improve, what information they can safely provide, how to write effective instructions, and when generated output requires verification. Training should therefore reflect the actual work of each function. Customer-service teams, marketers, analysts, and operations employees face different workflows, data risks, and quality requirements.

Genelle Chamberlain, IT Manager at PrimeSource, described a similar model. PrimeSource’s cross-functional AI council runs training based on real workplace scenarios and adapts those scenarios to different roles. Some applications are deliberately simple. An employee might use AI to change the tone of an email, for example. Small, well-defined tasks give employees a clear way to develop useful skills.

“A lot of them have a fear of even getting into AI, like they just don’t even know where to start,” Chamberlain said. Structured training addresses that barrier by providing a defined starting point and clear expectations for appropriate use.

For executives, AI literacy also has a governance function. Employees who understand approved use cases, data boundaries, verification requirements, and escalation procedures are better equipped to use AI within company policy. Training and governance should therefore develop together. As tools gain new capabilities, both the approved use cases and the required employee skills will change.

Leaders should measure adoption through business outcomes. Useful indicators can include time saved on specific workflows, quality improvements, employee usage within approved processes, and reductions in repetitive work. These measures connect training investment to operating performance.

The priority is to turn AI literacy into an organizational capability. Regular peer learning, role-specific training, clear usage rules, and practical examples give employees a structured path from basic experimentation to productive use. That creates a stronger foundation for scaling AI across the business.

Key highlights

• Prioritize customer outcomes: Deploy AI where it reduces customer effort, saves time, and improves resolution. Treat cost savings as part of the business case without allowing them to weaken trust or service quality.
• Design AI and human service together: Automate predictable steps such as authentication, information gathering, and routing, then give agents the full context. ibex says this model can cut human-agent handle time by 30% or more.
• Prepare agents for harder work: As AI absorbs routine requests, agents will handle a greater share of complex cases. Invest in advanced training, real-time AI assistance, and KPIs that account for rising case complexity.
• Build governance into deployment: Define customer journeys, data boundaries, regulatory requirements, testing, and ownership before scaling AI. Cross-functional governance and continuous monitoring can reduce operational and reputational risk while supporting faster execution.
• Make AI literacy role-specific: Give employees practical use cases tied to their daily work, supported by clear data and verification rules. Measure adoption through workflow improvements, quality, and time saved rather than tool usage alone.