AI Observability

why it matters ai costs do not grow randomly they grow because usage changes more requests, different models, new teams, inefficient integrations, or silent failures all show up in usage data before they show up on a bill ai usage and observability gives you direct visibility into the activity behind your ai spend instead of guessing why costs increased, you can see exactly what changed, when it changed, and who or what is responsible this is the layer that connects financial reporting to real operational behavior without it, cost control is reactive with it, cost control becomes predictable how to use it use this page when you want to understand what is actually happening with ai usage across your organization start with the overview at the top of the page, you see total spend on ai activity for the selected time range the top resources driving spend the resources with the fastest changing usage this gives you immediate context you should be able to tell quickly whether usage looks normal or if something needs investigation review active usage metrics below the overview, you see time series data for ai usage metrics such as tokens, api calls, and errors by default only metrics with usage are shown the time range is set to the last seven days this provides a fast operational snapshot without noise you can optionally show metrics with no usage if you want to see everything that is available drill into a specific meter when something stands out, open the detailed view for that meter click on the button in the upper right of the graph that has two arrows this can be used to expand the graph into the full page and provide many ways to dig into the data from there, you can change the time range to focus on when usage changed adjust how data is grouped over time (hourly, daily, weekly, monthly) choose whether values are shown as totals, peaks, minimums, or averages these controls help you distinguish between steady growth, short lived spikes, and bursty behavior break usage down by business unit usage can be viewed combined across the organization broken out by business unit business units are flexible and can represent teams, applications, projects, or cost centers you can also select only specific business units to compare, keeping the view focused this makes it easy to identify ownership and isolate where changes are coming from filter to isolate the cause you can filter usage by dimensions such as model or provider for example view usage for a single model only compare behavior across providers isolate production traffic from everything else filters apply consistently to both charts and tables, so the data always matches what you see identify top consumers alongside the charts, you can see which business units are responsible for the most usage or spend over the selected period this helps you quickly answer who is driving usage, not just how much is being used inspect and export the data every chart is backed by a table showing the exact data points used in the visualization when needed, you can export the data as csv save the current configuration as a reusable view saved views are useful for recurring reviews or investigations and they can be found under the "saved views" navigation link when to use this page use ai usage and observability when spend increased and you need to explain why you want to catch abnormal usage early you need to understand which teams, models, or providers are driving activity you want operational answers before financial escalation this is how you move from reacting to ai costs to understanding them