AI Overview
The AI Overview answers the question that otherwise arrives with the Azure invoice: what are the platform's AI features consuming, which feature is spending it, and is this month running away.
In-context help (F1)
Press F1 to open this page. Every figure on the overview can be clicked and leads to the calls it was computed from.
Open the page
- Via search: AI Overview
- Or from the merchantCENTRAL AI Setup through the AI Overview action
Tokens first, money second
Azure bills by token — roughly, parts of words. The price per token depends on the model, the region and your Azure agreement, and it changes. merchantCENTRAL does not guess that rate: a figure that looks like money but rests on a guessed rate is worse than no figure at all — people plan against it.
Instead you enter your own prices in the AI Setup. Only then do the cost figures appear on this page, and everywhere they are labelled as an estimate. Without prices the overview shows tokens only, exactly as before.
The tokens are counted by the real tokenizer, not estimated: Business Central provides the cl100k_base encoding, which is what GPT-3.5 and GPT-4 deployments are billed in. Newer models (gpt-4o and after) bill in o200k_base, which the platform does not expose, so on those deployments the figure is close rather than exact. For "which feature is costing us this" the difference does not matter; the invoice itself is in the Azure portal.
Key figures
| Tile | Meaning |
|---|---|
| Tokens This Month | Consumption since the first of the month, with a trend line for the last seven days. |
| Calls This Month | Number of AI calls, with a trend line as well. |
| Tokens Per Call | Average consumption of a single answer. The figure that rises when a feature starts sending more data — in a monthly total that looks exactly like "more calls". |
| Budget Used | Share of the monthly budget, as a percentage. With a token and a cost budget both set, the tile shows the one further along — it will stop the features first. With no budget set, it shows consumption since setup instead. |
| Approx. Cost This Month | Only visible once both prices are entered in the AI Setup. Below it stands what a single call cost on average. |
Cards
- Consumption by Feature — AI features ranked by tokens this month, biggest spender first. This is the answer to "on what", which a single total cannot give.
- Calls This Month — how the calls ended: Answered, Failed, Not sent. The distinction matters for cost: a failed call may already have been billed by Azure, a call that was not sent never left Business Central.
- Consumption by Model — tokens this month per deployment, once there are more than the standard one (see Model per Feature). Different models cost different amounts per token; this is where a month that got more expensive without more calls explains itself.
- Budget and Resource — monthly budget, remainder, reset date, the Azure deployment in use and total consumption since setup.
Below that, one line per AI feature that has been used, with its consumption this month and its state: Active / Switched off / No longer installed. A feature that runs on a deployment of its own shows that deployment's code next to its name.
Monthly budget
The budget is a stop, not a warning. Once it is reached, the AI features refuse with a message naming the figure and where to change it. On the first of the next month they resume by themselves.
A brake, not a guarantee
It is checked before each call against what other sessions have already committed. Two runs starting at the same moment therefore both pass it and cross the line together. The overshoot is one round of calls, not a runaway month.
Why a stop and not a warning
A warning is read by whoever happens to be at the screen — on a running batch, nobody. Only a stop makes sure the invoice surprises no one.
The budget is off by default. A value of 0 means no limit. It is set in the AI Setup under Monthly Token Budget, in money under Monthly Cost Budget; whichever is reached first applies. Budget Warning % decides when the overview turns amber (80 % by default).
Blocked calls are logged. A month in which the budget stopped forty explanations is visible on the overview — otherwise the feature gets reported as broken.
On the dashboard
Once the AI is set up, the merchantCENTRAL Dashboard carries an Artificial Intelligence line among the modules, with the month's consumption, the estimated cost and the budget state. Clicking it opens this overview. Without a configured resource there is no line; setting it up happens through the merchantCENTRAL AI Setup page only.
Once the monthly budget is reached or close to it, a row also appears under Waiting for you, next to orders stuck in error — an exhausted budget stops every AI feature of the platform.
Actions
| Action | Effect |
|---|---|
| Usage Log | Opens every recorded call, newest first. |
| AI Setup | Opens the setup with the resource and the budget. |
| Copilot & AI Capabilities | Opens the Business Central page where each AI feature can be switched off by name. |
| Clean Up Log | Removes log entries older than the retention that is set up. |