Media agencies build auditing tools to manage AI token costs

The shift toward agentic advertising requires a new layer of financial governance to prevent automated systems from depleting campaign budgets. As agencies like PMG and Quad's Rise build custom monitoring stacks, the industry is acknowledging that efficiency gains from automation are currently offset by the high cost of compute and the risk of 'agent drift.' This trend forces a re-evaluation of the 'human in the loop' model, moving it from a philosophical preference to a fiscal necessity for maintaining profitability. Watch for whether SSPs like PubMatic integrate more granular cost-capping features directly into their platforms to attract budget-conscious agencies.

The push to control AI spending is not limited to media agencies. Across the telecom and enterprise software sectors, vendors are racing to commercialize agentic AI platforms while grappling with the same cost-overrun dynamics. In June 2026, Ericsson launched its AI in RAN commercial software subscription, claiming up to 20% higher downlink throughput and up to 10% better spectral efficiency across more than 15 live deployments, signaling that even infrastructure vendors are packaging AI capabilities as subscription products with predictable pricing, a model agencies may soon demand from ad-tech platforms.

Nokia has moved aggressively to build a full agentic AI stack for network operations, a parallel to the multi-layer cost stacks agencies are now auditing. At DTW Ignite 2026 in Copenhagen, Nokia partnered with Google Cloud to build six specialized Gemini-powered agents capable of reducing network problem-solving times by 50% to 80%, with the platform launching on Google Cloud Marketplace in September. The company separately combined with AWS and Databricks to build a unified telco data platform under its Autonomous Network Fabric, claiming automation rates above 90% and service interruption periods of one minute per year or fewer. These deployments illustrate the scale at which AI token consumption can compound when agents operate continuously across domains, the same risk agencies face when agentic systems run programmatic campaigns without cost caps.

The technical architecture choices underlying these cost dynamics are becoming a differentiator. Light Reading reported that Ericsson and Nokia are diverging sharply on AI-RAN strategy, with Nokia running all Layer 1 functions on Nvidia GPUs via CUDA while Ericsson limits GPU usage to forward error correction only, a distinction that directly affects per-token compute economics. For ad-tech, the analogy is clear: agencies auditing AI token costs must understand whether their platform partners are running inference on expensive GPU clusters or offloading to cheaper CPU-based pipelines. Verizon's public call for industry-wide interoperability standards for agentic systems, disclosed alongside its 60,000-site vRAN deployment, underscores that no standardized protocol yet exists for agentic command, control, and cost assurance, a gap that standards bodies may eventually need to address for programmatic AI agents.