AI & agents

What is Exception Desk?

An exception desk is a review queue where an AI accounting agent surfaces transactions, matches, or filings it could not confidently handle, so the human clears only the exceptions rather than reviewing everything.

Example

Out of 200 bank transactions this month, the AI confidently processed 187. The remaining 13 appear on your exception desk: three with ambiguous suppliers, five with unusual amounts, and five with no matching invoice. You resolve these manually.

How Exception Desk Works in Practice

The exception desk is the human touchpoint in an AI-native accounting workflow. Instead of reviewing every transaction (the traditional model), you review only the items the AI could not handle with sufficient confidence.

This inverts the traditional workflow. In a manual system, every transaction needs human attention. With an exception desk, the AI handles the majority automatically and surfaces a curated list of items that genuinely need human judgment. This might be 5-10% of transactions rather than 100%.

The exception desk also serves as a feedback mechanism. When you resolve an exception, the AI learns from your decision. Over time, the same type of exception occurs less frequently as the agent builds company-specific context. The exception desk gets shorter as the system learns your business.

Step by Step

The AI agent processes each incoming item (transaction, document, payment) and assigns a confidence score to its classification and matching decisions. Items above the confidence threshold are processed automatically. Items below the threshold are routed to the exception desk.

The exception desk presents each item with context: what the AI knows about it, what it tried to do, why it was not confident, and suggested resolutions. The human can accept a suggestion, override it, or provide additional information.

Resolutions feed back into the agent's knowledge. If you consistently resolve a particular type of exception the same way, the agent learns to handle that pattern automatically in future. This creates a virtuous cycle where the exception desk becomes more efficient over time.

Practical Tips

Set a recurring weekly reminder to clear the exception desk so items do not pile up
Resolve exceptions with as much context as possible to help the AI learn your preferences
Use exception patterns to identify process improvements: if the same type of exception recurs, consider whether bank rules or vendor settings could prevent it

Common Mistakes to Avoid

Ignoring the exception desk: unresolved exceptions can affect report accuracy and period-end close
Over-reviewing by checking items the AI already processed confidently: this defeats the purpose of the exception-based workflow
Not providing context when resolving exceptions: a quick resolution helps close the item but a detailed one helps the AI learn

Frequently Asked Questions

How many exceptions should I expect?

In the early weeks, perhaps 10-15% of transactions. As the agent learns your business patterns, this typically drops to under 5%. The rate depends on how varied your transactions are and how consistently you resolve exceptions.

What happens if I do not resolve exceptions?

Unresolved exceptions remain in the books as uncategorised or unmatched items. They affect the accuracy of your reports and can block period-end close. Review them regularly, ideally weekly.

Related Terms

Agentic Accounting

Agentic accounting is accounting performed by AI agents that take multi-step actions with tools (reading bank feeds, matching invoices, posting journals, preparing returns) rather than just answering questions or making suggestions.

Continuous Close

Continuous close is an accounting approach where the books are kept substantively up to date at all times through continuous processing, so period-end close becomes a short review rather than a weeks-long catch-up exercise.

Human-in-the-Loop Filing

Human-in-the-loop filing is a workflow where an AI agent prepares regulatory filings (VAT returns, annual accounts, confirmation statements) and the human reviews and confirms before electronic submission. The agent never files silently.

AI-Native Accounting

AI-native accounting is an accounting platform built from the ground up around AI agents and a living ledger, rather than adding a chatbot to traditional accounting software.

Confused by accounting jargon?

AccountsOS explains everything in plain English. Ask any question about your books and get a clear, jargon-free answer.

Try Free for 14 Days

AI accounting glossary View full glossary