Enterprise organisations have solved the governance problem for AI agents but remain unable to accurately track their operational costs. Across 107 enterprises surveyed, the typical organisation runs three orchestration platforms simultaneously, selecting them for model flexibility rather than vendor lock-in—Microsoft leads in primary usage whilst Anthropic commands forward momentum. This multi-platform approach reflects a mature understanding of agent deployment: teams know how to set permissions, enforce controls, and manage permissions across systems. Yet this sophistication in governance masks a critical blind spot. When a CX team deploys agents across Zendesk, Salesforce, and a custom orchestration layer, the cost visibility collapses. Token consumption, inference latency, and per-interaction pricing become opaque across platforms, making it impossible to calculate true cost-per-resolution or justify agent investment to finance teams.
The implications for CX professionals are immediate and uncomfortable. If your organisation runs multiple orchestration platforms—which the data suggests you likely do—you cannot confidently answer what your agents actually cost to operate. This creates a governance paradox: you can enforce that agents follow compliance rules, but you cannot enforce that they operate within budget constraints you cannot measure. For teams already running Agentforce or similar enterprise solutions, this means the business case for agent expansion rests on incomplete financial data. The risk is not that agents are uncontrolled; it is that they are ungoverned financially, leaving CX leaders unable to optimise spend or defend headcount decisions when cost pressures emerge.
The underlying issue is architectural. Multi-platform orchestration was the right choice for flexibility, but it fragmented cost attribution. Each platform meters differently, reports in different cadences, and lacks cross-platform visibility. Until enterprises demand unified cost metering as a non-negotiable requirement—equivalent to the security and governance standards now enforced—CX teams will continue operating agents in a financial fog. This becomes especially acute as agent complexity increases and organisations move beyond simple routing into autonomous resolution, where token consumption and model selection directly impact margin.
Across 107 enterprises, agentic orchestration is not a choice of a single platform.The typical enterprise runs three orchestration platforms at once, and selects them for flexibility across models rather than affinity to any single one. Microsoft leads primary usage while Anthropic leads forward con