Dashboard
The Dashboard is where you check a number. It reports what your organization spent, who is using AI, which models answered, and which tools succeeded, and it stops there. When you want a conclusion with a fix attached instead, read Insights.
Three tabs, and a window picker offering the last 7 or 30 days. Trends compare the first half of the window against the second, so a 7-day view reacts fast and a 30-day view is the one to quote.
Overview
Adoption and spend together, which is the pair worth reading first.
Routing
How your routing policies are behaving, and what the models cost.
Cost per LLM call and credits used per month carry the trend. Spend by model, by vendor, and by provider answer the same question at three levels of detail, so start at model and widen only when a number surprises you. Tokens per day splits input, output, and cache reads, which is how you tell a workload that grew from one that stopped using the cache.
Gateway coverage by harness is the rollout number, not a cost number. It says which AI clients are actually routing through Gateway, and a harness missing from it is a harness producing no telemetry at all. Absence of traffic looks the same as absence of work, so read it against your employee list rather than on its own. Choose a rollout path covers closing those gaps.
Tool calls
Whether the tools you granted are working.
Success rate by Connector and by tool are the two reliability views: the Connector view finds a broken integration, the tool view finds one bad action inside a working Connector. Tool calls and credits by group show adoption per team, and credits and calls per tool show which tools are worth their cost.
A failure rate here is a real problem to chase, and Tool call logs is where you chase it, one call at a time.
Cards fall back to their own empty state rather than blanking the tab, so a Dashboard with model data but no tool calls renders the half it has.
Next
For findings that name what to change rather than numbers to interpret, read Insights.