Insights | Merge

Insights reads the past few weeks of AI usage and lists findings, such as an expensive model doing work a cheaper one handles. For raw numbers, see the Dashboard.

Insights is off until Merge turns it on for your organization. Ask your Merge contact.

View workforce can read findings. Manage workforce can dismiss one or run its fix. Nothing changes until a person acts.

Insights reads Gateway request logs and tool call metadata. It never reads prompts, tool arguments, or tool results. Figures on a finding are computed in code; a language model writes only the text.

Finding types

Finding	Detects	What to do
Overpowered direct calls	Direct calls to a model at least twice your average price, over 5% of Gateway spend, where a cheaper routed model serves similar prompts	Add a policy in Routing
Idle Connectors	Provisioned Connectors nobody uses	Remove access in Connectors
Suggested skills	The same tool path of three or more steps, run repeatedly by several people	Click Create skill to save a draft in Skills
Flex tier candidates	Regular, non-streaming overnight work on the standard tier	Move jobs to the flex tier, about half the price with more latency
Cache opportunities	Projects re-sending a prompt prefix of 4,096 tokens or more without cache markers	Add cache markers, or a policy in Routing

Create skill is the only fix that runs from a card. To dismiss a finding, click the X on its card; it stays dismissed.