The biggest constraint was not visual but technical. Google Analytics has strict request limits. Making a new server request every time a user clicked a filter or switched a tab would break the tool. Engineering initially proposed a traditional server-side filtering model where every interface interaction fired a new query.

I pushed back against this approach. Constant loading spinners ruin the experience of rapidly comparing different AI platforms. We compromised by shifting the heavy computation entirely to the initial load. We engineered the system to fire exactly three optimized queries right when the dashboard opens. We fetch the overall channel grouping, the specific LLM performance, and the landing page data for both the current and previous date ranges simultaneously. Everything is cached locally so we can calculate complex deltas directly on the client side.