HOW TO - Track model credit consumption with Power BI

[jk_one]

Hi all,

As the AI Builder team has been receiving questions on how to track credit consumption per model, here is what you can do as of today to build a Power BI report that tracks consumption per model type and per model, leveraging the Dataverse AI Event table:

• Find and copy the environment URL (in Power Platform Admin Center -> Environments -> <Env name> -> Details)
• Open Power BI desktop
• Select "Common Data Service (Legacy)" as data source 
• Enter environment URL (do not forget the https:// and the / at the end)
• Select tables/entities msdyn_aievent, msdyn_aitemplate and msdyn_aimodel
• In the Model view, establish a many to one relationship between table msdyn_AIModel column msdyn_templateid and table msdyn_AITemplate column msdyn_aitemplateid
• In the Report view, add a Matrix visualization
  • Select “msdyn_uniquename” (from msdyn_AITemplate) and “msdyn_name” (from msdyn_AIModel) as rows, “month” from date hierarchy of msdyn_processingdate (from msdyn_AIEvent) as columns, and Sum of “msdyn_creditconsumed” (from msdyn_AIEvent) as values: this will show the monthly consumption per model type (prebuilt models) & per model (custom models)

You can of course change the processing date hierarchy to fit you needs, or change the type of visualization. 

Some additional info:

- admins see all data, makers see only their own;
- <blank> model id/type is when a model has been deleted;
- check out Monitor model activity (preview) | Microsoft Learn for the official documentation;
- the AI Builder consumption report - AI Builder | Microsoft Learn remains the source of truth for credit consumption tracking as some consumptions are not recorded in the AI Event table (ex: object detection training, prediction models consumption, errors when recording the predicts in the table etc). 

Hope that helps !

JK from the AI Builder team

[jjpreston291]

@jk_one 

An excellent and very timely article for me

I was trying this today using the new dataverse connector but i could not see the msdyn_aievent table

So have switched to the legacy connector - which works !

[jjpreston291]

Also, the 'createdby' in the aimodel table is the Environment GUID. If you want the Azure AD ID, then bring in the 'systemusers' table. Then create a relationship between aimodel/createdby and systemusers/systemuserid

[jk_one]

Hi @jjpreston291 

Thanks for trying it out and providing feedback (and additional information)!

I was trying this today using the new dataverse connector but i could not see the msdyn_aievent table

So have switched to the legacy connector - which works !

---

Yes, fixing that is in our backlog, I'll keep the thread updated once I have more info.

Thanks,
JK

[jjpreston291]

@jk_one Hi, Do you have any ideas how to identify the flows that trigger the AI models ? Your report has helped me identify the top models. If these model owners do not take action to reduce the usage, then I would like to stop the Flows from running  , or at least have visibility of them. Asking this from an Admin point-of-view.

[jk_one]

Hi @jjpreston291 

Unfortunately, there is no way to link a flow and an AI Builder predict today.

The closest you can come is to identify the user whose connection is used for the predict, using the "Created by"  in the AI Event table (aievent/createdby) : either they own the flow that consumes credits, or they have shared their flow with someone who is using their connection.

Hope that helps,

JK