The AI Reality Check - From Promises to P&L Outcomes

About the Webcast

Webcast Highlights.  You can watch the entire webcast and download a pdf of the slides.

Harry Rollason and Kelly Koelliker shared new research and practical guidance on AI in contact centers, drawing from three studies of nearly 6,600 respondents—including 5,000 consumers, 1,000 agents, and 600 business leaders. Their analysis explored how organizations are investing in AI, what outcomes they’re seeing, and how to close the gap between ambition and measurable results.

Key Findings

• AI investment is surging across customer‑facing tools, agent assist, workforce management, insights, and quality monitoring—but measurable impact remains limited.
• Nearly 90% of organizations have increased AI budgets, yet only half report meaningful reductions in routine agent work.
• Customer experience gaps persist: 51% of consumers say service falls short when they seek help.
• Agent strain and attrition risk are rising—31% are considering leaving within six months.
• Executives view contact centers as value centers, but poor data quality and availability remain major barriers to success.

Common Pitfalls

Three recurring failure modes emerged:

1. Paralysis by choice—too many options stall progress.
2. Stalled pilots—initiatives that never scale.
3. Disappointing production results—deployments that fail to deliver ROI.

Recommended Approach

Experts advocate a phased, workflow‑by‑workflow strategy that sets clear metrics and delivers measurable results at each step.

• Prioritize high‑impact, feasible workflows.
• Define and instrument metrics before deployment.
• Favor solutions that integrate with existing systems and include governance, model management, and domain expertise.
• Use hybrid human‑AI designs with domain‑specific or personalized models.
• Maintain continuous data pipelines and governance to sustain model performance.

Critical Insights

Data quality and governance are the top blockers to translating AI investment into value. Pilots often fail to scale due to stale training data and lack of ongoing maintenance. Consumers still prefer human agents for important interactions, so fully automated designs risk increasing churn. Matching the right model to the right problem, rather than relying on one‑size‑fits‑all LLMs, is essential.