# How Kong AI Gateway Controls and Optimizes Your AI Token Costs

This session focuses on controlling AI costs before they spiral, especially for teams running experiments that quickly become expensive in production. The speaker explains that many organizations approve AI projects based on capability first, then struggle with overruns, missed forecasts, and poor real-time visibility into token spend, citing research that 79% face cost overruns, 85% miss budget forecasts, and only 25% have real-time cost visibility. He frames the core problems as attribution sprawl, model and provider sprawl, and constantly changing pricing, then argues that every dollar of AI spend needs policy-based governance. Kong’s approach is to use its AI gateway and broader connectivity stack to track token usage, model calls, pricing, and identity context in one place, enabling dashboards, budget thresholds and actions, pricing catalogs with contract overrides, attribution mapping, anomaly detection, and projected savings from controls such as prompt compression, with chargeback, monetization, and enterprise billing as later stages.