AI Usage Log
One row per AI call, across every merchantCENTRAL app. The detail behind the figures on the AI Overview.
In-context help (F1)
Press F1 to open this page.
Open the page
- Via search: AI Usage Log
- Or by clicking a figure on the AI Overview — the log then opens already filtered to match
What is not logged
Neither the question nor the answer. Prompts carry business data — item costs, margins, customer text. A log that keeps them is a second copy of that data with none of the protections the original has. The numbers are enough to explain a bill; the content is not needed for it.
There is one exception, and it is deliberate: the error message returned by the service is stored. Without it a failure cannot be investigated at all, and the usual ones ("deployment not found", "quota exceeded") carry no business data. A rejected request can make Azure quote part of what was sent back, though — which is why the table is classified as CustomerContent.
Columns
| Column | Meaning |
|---|---|
| Called At | When the call was made. |
| Feature | Name of the AI feature at the time of the call. It stays readable even after the app has been removed. |
| Deployment | Code of the assigned deployment. Empty for the standard model. |
| Pictures | How many pictures the call sent along. Pictures are billed as prompt tokens. |
| Result | Answered (green), Not sent (amber) or Failed (red). |
| Prompt Tokens | Consumption of the instructions plus the data the feature sent along. |
| Completion Tokens | Consumption of the answer. Azure charges more for these than for the question — hence the split. |
| Total Tokens | The sum of both. This is the figure the monthly budget counts. |
| Duration | How long the AI service took. |
| User ID | Who triggered the call. |
Through Choose columns you can also show Deployment Name, Model (the version Azure reports for direct calls), Reasoning Tokens (the part of the completion tokens a reasoning model spent thinking) and Approx. Cost. The cost is filled only for calls to an assigned deployment: they are priced with its prices when they are made. Calls to the standard model are priced by the setup with its current prices.
The factbox on the right shows the full error message for the selected row — in a column it would be truncated.
Telling the results apart
| Result | Cost | Meaning |
|---|---|---|
| Answered | billed | The normal case. |
| Failed | possibly billed | The call reached Azure and went wrong there. |
| Not sent | none | The monthly budget stopped the call before it left Business Central. |
Actions
| Action | Effect |
|---|---|
| This Month | Shows only the calls that count against the current monthly budget. |
| Show All | Removes the period filter. |
Retention
The log grows by one row per call. Log Retention Days in the AI Setup decides how long entries stay; the Clean Up Log action on the AI Overview removes the older ones. Total consumption since setup is untouched by it.