This documents includes some sample KQL queries and report samples that were created based on the logs that were written as part of application execution. These are samples that can be extended for your own use.
This calculates AOAI consumption cost based on the number of tokens sent to AOAI and number of tokens created by AOAI services. Based on the tokens consumed, total cost is calculated by multiplying the numbers with the consumption cost based on the pricing listed here
Below query calculates cost. Do select the time range when executing this query
traces
| extend total_tokens = todouble(customDimensions.total_tokens)
| extend prompt_tokens = todouble(customDimensions.prompt_tokens)
| extend completion_tokens = todouble(customDimensions.completion_tokens)
| extend aoai_response_MS = tostring(customDimensions.["aoai_response[MS]"])
| extend day = format_datetime(timestamp, "yyyy-MM-dd")
| summarize prompt_cost=sum(prompt_tokens)/1000 * 0.03, completion_cost=sum(completion_tokens)/1000 * 0.06 by day
| order by day asc
| render columnchart
This calculates the length (number of unique dialogs) of each conversation. A dialog is what the user said and the corresponding response from the bot for that utterance. Since this a multi-turn conversation, multiple dialogs could be happening, within the conversation. One assumption that could be drawn is that, more the number of dialogs, the more engaging the experience is.
traces
| extend conversation_id = tostring(customDimensions.conversation_id)
| extend dialog_id = tostring(customDimensions.dialog_id)
| distinct conversation_id, dialog_id
| summarize dialog_count = count() by conversation_id
| render columnchart
