CLOUDFLARE has updated its AI Gateway User Insights to add much-needed context to AI traffic analysis. Since the initial launch, User Insights now helps teams understand not just which models and request counts are being used, but also the kind of work behind those requests—whether it is coding, research, writing, summarisation, or data analysis.
The new overkill view highlights when a model’s capabilities exceed the task requirements, enabling organisations to identify which users, agents, or applications drive that usage and how task, model, cost, and conversation patterns relate. This context supports targeted investigations and changes, such as adjusting models or workflows, with the added benefit that the insights are available to AI Gateway users at no extra cost.
Cloudflare also introduces a broader workflow around turning insights into action. The platform captures analysis signals after a conversation to classify traffic into categories, returns a confidence score, and links task complexity, intent ambiguity, and context dependence to the traffic. Users can connect this data to individuals and teams via AI Gateway behind Cloudflare Access, so usage can be attributed to specific users and sessions.
The update also links to new features: a Potential Savings view to highlight faster or cheaper model options without sacrificing quality, and an Auto Router (now in closed beta) that automatically routes tasks to appropriate models based on task type and cost considerations. Together, these enhancements aim to make model selection and routing more intelligent and cost-aware without requiring bespoke routing rules.