When an Agent Works the Exception Queue, Your Exception Rate Becomes a Capacity Forecast
In Accounts Payable Without the Data Entry, we argued that invoice capture removes the typing, and that the design of the exception path decides whether accounts payable actually improves. We recommended baselining exception rate, broken down by type, before anything is configured. That post looked at invoice capture in Dynamics 365 Finance. The same logic applies to agents, and we use the Business Central Payables Agent as the example here.
That advice matters even more once an agent is working the queue. At DAX Software Solutions, we think that when an agent drafts the routine invoices, your exception rate stops being only a quality metric. It becomes the number that tells you how many review hours, how many Copilot Credits and how much daily throughput your AP process needs.
What work does an AP agent leave for people?
The exceptions, plus a review of everything else. Microsoft’s Payables Agent FAQ states that the agent “stops processing and requests your assistance when it encounters uncertainty, such as an unidentified vendor or an unclear line classification.” When it proposes purchase order matches, Microsoft states that the warnings are “deterministic Business Central validations, not AI confidence,” and that matching doesn’t support Charge (Item) lines or order lines with prepayments.
Those stops are your exceptions. Microsoft’s guidance on supervising agent tasks states that tasks needing attention appear at the top of the Tasks pane, with a notification badge showing how many are waiting.
Routine invoices still get a human look. Microsoft states that users should “review the draft before the agent finalizes it,” and then review the resulting unposted invoice before a user posts it. The agent doesn’t post purchase invoices.
Why does the exception rate become a capacity forecast?
Because exceptions become the part of the human work that moves. Before the agent, AP effort was spread across keying, checking and resolving. After the agent, keying largely moves to the agent, and the human workload has two parts: a short review of each draft and a longer resolution for each exception. That makes the forecast simple to write down:
Monthly review hours = (invoices x draft review time) + (invoices x exception rate x exception resolution time)
As an illustrative example, not a client result: 2,000 invoices a month, two minutes to review each draft, a 15% exception rate and 12 minutes per exception gives about 67 hours of draft review plus 60 hours of exception handling. If a vendor format change pushes the exception rate to 25%, exception handling rises to 100 hours. The exception rate is the input that moves most, so it is the one to watch.
What else does the exception rate need to sit beside?
Credits and daily limits. Microsoft’s Business Central billing guidance states that the Payables Agent uses 50 Copilot Credits to process an invoice and 5 for each invoice line, and that emails without a valid invoice attachment don’t incur any cost. Using the same illustrative 2,000 invoices with three lines each, that is 130,000 Copilot Credits a month.
The FAQ also states that the agent processes up to 100 emails and up to 500 invoices per day. A team that receives most of its invoices in the last days of the month should check that peak days fit within that limit.
Finally, Microsoft states that when the Copilot Credit quota is depleted, the AI capability is unavailable until more capacity is added. If that happens at month end, the work returns to people, so the forecast needs a fallback.
How should you use exception data to plan?
Break it down, watch it weekly and fix causes upstream.
- Baseline by type: split exceptions into capture, matching, approval and master data, as set out in our earlier post
- Map agent escalations: match the agent’s stop conditions, such as unidentified vendors and unsupported line types, to your categories
- Time two kinds of review: a standard draft review and the resolution of each exception type
- Forecast low, expected and peak months: for review hours, Copilot Credits and daily throughput
- Watch the leading indicator: Microsoft states you can view agent KPIs from the Tasks pane, so track the Needs Attention count and exception rate weekly
- Fix upstream: repeated vendor exceptions point to master data, and repeated matching exceptions point to purchasing or receiving
- Plan the fallback: decide who processes invoices manually if credits run out at month end
This connects to who approves the overage and why agent license savings rarely reach the invoice: the people who review exceptions are the capacity you keep.
Frequently asked questions
When does the Payables Agent ask for help? Microsoft states it stops and requests assistance when it encounters uncertainty, such as an unidentified vendor or an unclear line classification.
Does the Payables Agent post invoices? No. Microsoft states it doesn’t post purchase invoices. Users review the draft and the unposted invoice.
How is the Payables Agent billed? Microsoft states 50 Copilot Credits to process an invoice plus 5 per invoice line. Emails without a valid invoice attachment don’t incur any cost.
Is there a daily limit? Microsoft states the agent processes up to 100 emails and up to 500 invoices per day.
DAX Software Solutions: Governance Before Scale
DAX Software Solutions implements and supports Business Central and designs exception handling and human-in-the-loop operating models for AP agents. If you want your exception data to drive capacity planning, talk to our team.
Figures in the example above are illustrative. Microsoft’s product and billing guidance is informational and subject to change. Confirm current rates, limits and behavior for your environment before acting.