Agentic AI in Customer Experience The Gap Between Vision and Operational Reality

Beyond the Hype: What Agentic AI Actually Means

The conversation between McKinsey partners Oana Cheta and Brian Blackader, alongside Parloa CEO Malte Kosub, makes an important distinction that many leaders miss. Agentic AI represents a fundamental shift from generative AI's content creation capabilities to goal-driven systems that make decisions and take actions autonomously.

Kosub describes four evolutionary phases: internal FAQ assistance, customer-facing chatbots, internal agentic workflows, and finally, external agentic customer interactions. Most organizations are stuck between phases two and three, struggling to move beyond basic chatbot functionality. This progression reveals a critical insight: the technology is advancing faster than organizational readiness.

Research from MIT's Computer Science and Artificial Intelligence Laboratory supports this observation. Their 2024 study found that organizations attempting to skip evolutionary phases in AI implementation experience failure rates exceeding 80%. The successful 5% typically follow a methodical progression, building organizational capabilities at each stage before advancing.

The ROI Imperative: From Curiosity to Accountability

Cheta makes a crucial point that resonates with current market pressures. After two years of AI experimentation, CFOs are demanding measurable returns. This shift from innovation theater to business impact accountability creates both pressure and opportunity.

The emphasis on ROI is well-founded but potentially problematic. Gartner's 2024 AI Investment Report indicates that organizations focusing exclusively on short-term ROI metrics miss 60% of AI's strategic value. The metrics that matter are changing. Traditional efficiency measures like average handle time reduction, while important, fail to capture AI's potential for revenue generation and customer lifetime value enhancement.

Consider the AI coaching example Blackader describes. Telecommunications companies in North America and Europe deployed algorithms identifying learning opportunities for call center agents, creating customized learning journeys over 12-15 weeks. Results showed improvements in average handle time and first contact resolution. However, the real value emerged in reduced employee turnover and enhanced upselling capabilities, metrics that manifested over quarters rather than weeks.

This temporal dimension of AI value creation challenges conventional business case development. Leaders need dual-horizon thinking: quick wins that justify continued investment alongside longer-term transformation that delivers sustainable competitive advantage.

The CIO-COO Collaboration Imperative

Perhaps the most important insight from the McKinsey discussion concerns organizational structure. The traditional boundaries between technology and operations are dissolving, yet most organizations maintain rigid functional silos.

Blackader observes that successful AI implementation requires "both teams working together to make sure it's secure, to make sure that it integrates with all the other systems." This sounds obvious until you examine how most organizations actually operate. A 2024 Harvard Business Review study of 300 large enterprises found that only 23% had formal collaborative frameworks between CIO and COO organizations for AI initiatives.

The consequences of this collaboration gap are severe. When AI implementation becomes purely a technology project, it fails to deliver operational value. When operations drive AI without technology partnership, security vulnerabilities and integration failures emerge. The sweet spot requires joint ownership of outcomes, shared accountability for KPIs, and integrated decision-making processes.

Several forward-thinking organizations are pioneering new structural models. A Fortune 100 retailer created a "Chief AI Operations Officer" role reporting jointly to the CIO and COO. A European bank established cross-functional AI delivery teams with permanent members from both technology and operations. These structural innovations signal recognition that AI transformation demands organizational transformation.

The Talent Transformation Challenge

Cheta's comments on talent reveal another critical gap between vision and reality. While acknowledging the need for technical specialists, she emphasizes that "the real challenge turns out to be a question of how we adapt the operational talent to complement AI."

This perspective deserves deeper examination. The McKinsey discussion focuses heavily on reskilling and upskilling, with relatively little attention to the displacement and disruption AI will inevitably cause. This optimistic framing, while politically safe, may leave leaders unprepared for the human dimensions of AI transformation.

Research from the Brookings Institution suggests that customer service roles face 70-80% automation potential over the next decade. While new roles will emerge, the transition period creates significant challenges. Organizations need honest conversations about workforce sizing, not just skill transformation.

The successful models Cheta describes merit attention: moving workers from routine tasks to relationship management, strategic thinking, and complex problem-solving. However, this transition assumes all displaced workers can successfully upskill to higher-value activities. Reality is messier. Some workers will transition successfully, others will require extended training periods, and some may need to pursue opportunities outside the organization.

Leaders who acknowledge this complexity can plan more effectively. Best practices include extended transition timelines, robust training investments, and transparent communication about changing role requirements. Companies like IBM and AT&T have demonstrated that proactive, honest workforce transformation programs, while difficult, can maintain employee trust and organizational capability during major technology transitions.

The Implementation Reality: Three Critical Hurdles

Blackader identifies three implementation challenges that deserve careful consideration: security and regulation, data and systems integration, and undocumented knowledge.

The security and compliance dimension has grown more complex since ChatGPT's launch. The EU AI Act, implemented in 2024, creates stringent requirements for AI systems in customer-facing roles. Similar regulatory frameworks are emerging across jurisdictions. Organizations that treated compliance as an afterthought in early AI pilots now face costly retrofitting.

The data and systems challenge reveals a dirty secret of enterprise technology: most large organizations operate with fragmented, inconsistent data spread across multiple systems. Blackader notes that call center agents often access seven to ten different systems. AI agents require integrated data and seamless system connectivity. This means AI implementation often triggers broader digital transformation initiatives, multiplying cost and complexity.

The undocumented knowledge problem is particularly fascinating. Blackader cites research showing 70% of contact center workers report that at least 25% of their daily work isn't documented anywhere. This "folklore knowledge" represents a massive challenge for AI systems that require structured information.

However, this challenge also presents opportunity. The process of documenting and structuring organizational knowledge for AI consumption often reveals inefficiencies, redundancies, and improvement opportunities. Organizations that approach this systematically can achieve operational improvements independent of AI implementation.

The Vision: Personalized AI Agents at Scale

Kosub's vision of personalized AI agents deserves scrutiny. He imagines airlines with 100 million customers deploying 100 million personal AI agents, each building relationships and guiding customers through complete journeys across sales, marketing, and service.

This vision is technologically plausible. The economic and operational realities are more complex. Consider the infrastructure requirements: massive computing capacity, sophisticated orchestration systems, robust security frameworks, and continuous monitoring. The cost per AI agent may be lower than human agents, but the fixed costs of platform development and operation are substantial.

Moreover, the customer acceptance question remains open. While some customers will embrace AI-driven service, others strongly prefer human interaction for complex or emotionally significant transactions. Research from the Pew Research Center indicates that customer AI acceptance varies significantly by age, transaction type, and cultural context. A one-size-fits-all approach risks alienating valuable customer segments.

The most sophisticated organizations are developing hybrid models: AI agents handling routine transactions and information requests, with seamless handoffs to human agents for complex situations. This approach maximizes efficiency while maintaining service quality where it matters most.

What Leaders Should Do Differently

The McKinsey discussion concludes with recommendations to start small, think big, and focus on building blocks. This advice is sound but insufficient. Based on the patterns distinguishing successful implementations from failed pilots, several additional principles emerge.

• Establish clear governance frameworks before scaling AI initiatives. This includes defining decision rights, establishing quality standards, creating feedback mechanisms, and setting risk tolerances. The 5% who successfully scale AI don't move faster; they move more systematically.
• Investment in data infrastructure deserves equal priority with AI technology itself. Organizations that treat data integration and quality as prerequisites rather than parallel workstreams achieve better outcomes. This often means delaying AI deployment until foundational data capabilities exist.
• Change management cannot be an afterthought. Blackader wisely notes that "believing that technology is the solution is probably the wrong way to start." The most successful AI transformations invest 50% or more of their program resources in change management, training, and organizational alignment. Technology is the enabler; organizational change is the actual transformation.
• Resist the temptation to pursue comprehensive solutions immediately. The pattern of successful scale follows a different path: identify high-value, well-defined use cases; achieve demonstrable success; learn from implementation; then expand systematically. This approach builds organizational confidence and capability while managing risk.
• Establish metrics that capture both efficiency gains and strategic value creation. First contact resolution and average handle time are important, but incomplete. Customer lifetime value, revenue per interaction, and employee satisfaction provide a more complete picture of AI impact.

The Workforce Transition Reality

Returning to the talent challenge, leaders need frameworks for managing workforce transitions that acknowledge both opportunity and disruption. Several models are emerging from early adopters.

• The "skills passport" approach, pioneered by several European telecommunications companies, creates detailed skill profiles for every employee. As AI transforms role requirements, the system identifies transition pathways based on existing capabilities and learning potential. This data-driven approach to workforce planning enables proactive rather than reactive management.
• The "AI augmentation academy" model, implemented by a major North American bank, creates internal training programs specifically designed to prepare customer service workers for AI-augmented roles. Rather than generic upskilling, these programs teach workers how to collaborate effectively with AI systems, interpret AI recommendations, and handle exceptions AI cannot resolve.
• The "graduated automation" approach phases AI introduction over extended periods, allowing workers to adapt gradually. Rather than replacing entire functions overnight, organizations automate specific task components, monitor results, and adjust based on both performance data and employee feedback.

These models share common characteristics: transparency about changing requirements, investment in worker development, and extended transition timelines. They recognize that workforce transformation is measured in years, not quarters.

Addressing the Questions McKinsey Avoided

Several important dimensions receive insufficient attention in the McKinsey discussion. Leaders evaluating agentic AI investments should consider these carefully.

Customer privacy and data ethics represent significant concerns, particularly as AI agents accumulate detailed behavioral information over time. The personalized agents Kosub envisions require extensive data collection and analysis. Organizations must balance personalization benefits against privacy risks and regulatory requirements. The recent backlash against surveillance capitalism suggests customer tolerance for data collection is not unlimited.

Algorithmic bias in customer service AI has received growing attention from researchers and regulators. AI systems trained on historical interaction data can perpetuate or amplify existing service disparities across customer segments. Organizations deploying agentic AI need robust bias detection and mitigation frameworks. This is not merely an ethical imperative; it represents a significant legal and reputational risk.

The competitive dynamics of AI transformation deserve consideration. If the vision of universal AI agents materializes, what becomes the basis of competitive differentiation? When everyone deploys similar AI platforms, competitive advantage shifts to data quality, process design, and change management execution. Leaders should think carefully about where sustainable differentiation will come from in an AI-enabled world.

The cybersecurity implications of autonomous AI agents making decisions and taking actions on behalf of customers are substantial. Each AI agent represents a potential attack vector. As agent capabilities expand, so does the potential damage from compromised or manipulated systems. Security architecture must evolve in parallel with AI capabilities.

The Path Forward

The gap between the 5% who successfully scale AI and the 95% stuck in pilot phase will likely persist for several years. This creates both risk and opportunity for business leaders.

The risk is straightforward: competitors who execute effectively will establish advantages in operational efficiency, customer satisfaction, and employee productivity that compound over time. The opportunity is equally clear: most organizations are still figuring this out. Leaders who move systematically, learn quickly, and scale thoughtfully can establish sustainable advantages.

The most important lesson from McKinsey's discussion may be one that remains implicit: agentic AI transformation is fundamentally an operational and organizational challenge, not primarily a technology challenge. The technology exists and continues improving rapidly. The bottlenecks are organizational readiness, change management capability, and leadership commitment to sustained transformation.

Cheta's observation that successful companies are asking "what should AI own as we execute on our vision?" rather than "where does AI fit?" captures the mindset shift required. This reframing treats AI as a core capability, not a tool to be integrated into existing processes.

Leaders should approach agentic AI with both ambition and realism. The vision of personalized AI agents transforming customer experience is achievable. The path to that vision requires patient investment in foundations: data infrastructure, organizational alignment, workforce capability, and systematic learning from implementation experience.

The conversation McKinsey facilitates provides valuable insights from practitioners at the forefront of AI transformation. However, the optimistic framing should be balanced against implementation realities. Most organizations will find this transformation harder than anticipated, taking longer than planned, and requiring more fundamental organizational change than expected.

Those who succeed will combine strategic vision with operational discipline, technology investment with change management focus, and efficiency objectives with value creation opportunities. The 95% still struggling to scale AI implementations represent not just a statistic but a warning: technology alone is never the answer. Organizational transformation, leadership commitment, and sustained investment in both technology and people determine who captures the value agentic AI promises to deliver.

The future of customer experience Kosub describes is coming. The question is not whether AI agents will transform customer operations but which organizations will lead that transformation and which will struggle to keep pace. That outcome will be determined less by technology choices and more by how effectively leaders manage the complex organizational, workforce, and operational changes AI enables and requires.