The Contact Centre Expo conversation around AI agents reveals a widening gap between investment and outcomes. Organisations are deploying autonomous agents at scale—from Home Depot's multilingual voice systems to enterprise acquisitions like Sierra's purchase of Fragment—yet the operational reality appears misaligned with the hype. The core tension centres on cost-benefit: teams are spending substantially on agent infrastructure whilst reporting marginal improvements in resolution rates or customer satisfaction. This disconnect matters particularly for teams already committed to these platforms. What does this mean for organisations mid-implementation of Agentforce or similar enterprise solutions, where sunk costs make reversal difficult and stakeholder expectations have already been set?
The underlying issue appears structural rather than technical. Agent adoption data shows daily usage without perceived essentiality, suggesting tools are being used because they exist rather than because they solve genuine problems. Simultaneously, emerging risks—including security vulnerabilities around AI-generated fake service numbers and regulatory exposure from opaque AI decision-making—are creating compliance overhead that wasn't factored into initial ROI models. For CX leaders, this signals a need to recalibrate success metrics away from deployment velocity and towards demonstrable efficiency gains, whilst building governance frameworks that account for reputational and legal risk.
The conversation at Contact Centre Expo likely reflects a broader market correction: early-stage AI agent vendors are facing pressure to prove value, whilst established platforms must justify premium pricing against underwhelming performance data. For support teams and consultants, the implication is clear—vendor selection should now prioritise transparent performance benchmarking and integration depth over feature breadth, and budget allocation should reserve capacity for governance and monitoring rather than assuming agents will simply reduce headcount.