{
"@odata.type": "#Microsoft.Dynamics.CRM.expando",
"imagesCount@odata.type": "#Int64",
"imagesCount": 0,
"modelType": "default",
"text": "{\n \"Summary\": \"******.\",\n \"ContractingParty\": \"**********\",\n \"CounterParty\": \"*********\",\n \"StartDate\": \"2026-04-01\",\n \"EndDate\": \"2029-03-31\",\n \"ContractType\": \"Sale Agreement\"\n}",
"thoughtSteps": "",
"finishReason": "stop",
"totalTokens@odata.type": "#Int64",
"totalTokens": 19202,
"promptTokens@odata.type": "#Int64",
"promptTokens": 18989,
"completionTokens@odata.type": "#Int64",
"completionTokens": 213,
"modelName": "gpt-41-mini-2025-04-14",
"costAsAiBuilderCredits@odata.type": "#Int64",
"costAsAiBuilderCredits": 20,
"costAsCopilotCredits": 0
}
The flow also uses the AI builder action Recognize text in an image or document to extract the text from the pdf to pass to the Run AI Prompt action, which just returns the number of AI builder credits used. I can add that to the spreadsheet as well.
In the minimal testing I've done so far, it is taking about 18 AI builder credits per document. If the averages hold up, I'll need about 198,000 AI builder credits for this project. If 25,000 Copilot credits cost $200, then this will cost us about $1,584. To do this manually, it would probably take someone at least 20 minutes to read through each contract to generate a summary and extract the above metadata in the list and then update it in the list, which is probably a conservative estimate.
The goal once this is done is to then get Copilot Microsoft 365 licenses for the key users of this SharePoint site so they can get information faster. We had considered creating a Copilot Studio agent, but after some research and doing some testing, we felt it would work better with Microsoft 365 Copilot licenses than with an agent.
I hope I didn't distract too much away from the original post, I just wanted to share a way I am estimating the number of credits needed for a current project.