Business Central Expense Agent to Flag Natural-Language Policy Issues for Approvers

The detail that matters most is what the feature leaves out. Microsoft does not describe it as auto-approval. It describes a second reviewer that adds context, while a person still makes the final decision.

What the roadmap entry says​

The roadmap listing says that once an employee submits an expense report, Expense Agent checks the whole report and each expense line. It looks for possible policy violations, suspicious patterns, or anything else that may need a closer look. The output is extra context for the approver, and Microsoft says the human approval step stays in place.

The listing gives these details:

Treat those dates as estimates. Microsoft's roadmap page says it gives estimated release dates and descriptions for commercial features and that all of it can change. When we retrieved the public roadmap page, its readable text did not show item 573255, so the fields above come from the roadmap listing itself and have not been confirmed against a rendered live page.

Microsoft's examples are policies that numbers alone can't capture:

• Business-class travel: when it is allowed
• Business meals: what counts as appropriate
• Hotels: when it is acceptable to book outside preferred standards

The listing also limits the first release. Because policy checks run after submission, the initial version can apply controls at the level of a single expense line. Read that as the clearly stated scope for launch, not as a promise of complex rules that compare several lines.

Section summary: This is a planned, in-development feature. It flags issues after submission and does not approve anything, and Microsoft's dates are targets, not commitments.

This feature has been on the books for a while​

The roadmap entry is not the first time this feature has appeared. Microsoft's 2026 release wave 1 plan for Business Central already had an item called "Expense Agent supports approval process," and its feature description matches the new roadmap text almost word for word. That release-plan entry says it validates submitted reports against company policies written in natural language. It also claims the feature reduces review effort, improves consistency across approvers, strengthens compliance, and gives finance teams better visibility into policy exceptions and suspicious submissions without removing human oversight.

The reason it now shows up on the AI at Work roadmap is a change in how Microsoft publishes plans. A notice on Microsoft's release-plan pages says Release Plans stop being published in September 2026. From that month on, new Dynamics 365, Power Platform and Dataverse capabilities go on the AI at Work roadmap, and the old release plans stay up for historical reference.

The new entry does give Business Central admins something the old plan didn't: specific preview and GA months. If you've followed this feature since wave 1, October 2026 is the preview date to plan around.

Rules and policies are different things​

The new feature makes more sense once you see that Expense Agent already does two kinds of checking. Microsoft Learn's prerelease page on policy compliance spells it out:

• Rule validation (already present): Administrators define rules in Expense Management Rules in Business Central, and the agent applies them automatically. When you create or edit an expense, the app checks it against the rules your organization has set up. If an expense doesn't meet a rule requirement, you see a notification right away rather than finding out after you submit.
• Soft warnings and hard blocks: A soft warning lets the employee add a justification and keep going. Hard blocks prevent you from submitting until you fix the issue. For example, if a required receipt is missing or an expense category isn't allowed, you must correct the expense before you can submit the report.
• Policy validation (what this feature covers): Microsoft says rule validation is performed proactively for individual expenses, while policy validation is executed after the expense report has been submitted. The same page explains why they are separate: some controls work on a single line, some on the whole report, and some depend on the relationship between multiple expense lines.

In practice, rules are the strict, numeric checks that can stop a submission. The new natural-language policy layer is the judgment-based check that runs after submission and informs the approver without blocking anyone. One page says employees catch problems "before you submit" and another says the new feature works "after submission." Both are true, because they describe two different layers.

Section summary: Hard limits stay in Expense Management Rules and are enforced before submission. Policy validation adds a natural-language review step after submission and does not replace those rules.

Where it fits in the approval flow​

Microsoft's prerelease approver guide shows how approvals work today. An approver goes to the For My Approval tab and checks the report lines, amounts, categories, receipts and the policy compliance summary. From there they choose one of two actions:

1. Approve: The report moves to Approved and the employee gets a notification.
2. Send Back: The approver writes comments under Request clarification in the Add comments field. The report goes back to Draft, and the employee can fix it and resubmit.

The same guide says approval workflows are set up and managed in Business Central itself. The policy-validation feature should add its findings to this review. Microsoft has not published screenshots or exact UI details for item 573255, though, so don't plan training material around a specific layout yet.

One possible point of confusion: the earlier wave 1 release plan for Expense Agent as a whole says the agent can submit a report for approval automatically when approvals are turned on and no rule violations are found. That covers routing a report into approval. It is not the new policy feature approving reports on its own, and nothing in the roadmap entry suggests it will.

How the agent operates​

Microsoft's technical overview of Expense Agent gives some background on how it works. It says the agent interacts with Business Central in a similar way as a user. It receives natural-language instructions that define how to handle expenses and uses page metadata (captions, tooltips, and other properties) along with its own instructions to determine each processing step. The same documentation says all financial data, audit trails, and posting logic are processed within Business Central, so existing rules, permissions, and workflows continue to apply.

Admins should also note Microsoft's language caveat. It warns that the feature can be used in unsupported languages but might not function as intended. Language quality might vary based on the user's interaction or system settings, which might impact accuracy and the user experience. If your expense policies aren't written in English, test with care during preview.

Questions to ask before preview​

The idea is sound. Business Central MVP Ian Grieve of azurecurve put it this way: rules-based enforcement cannot easily handle nuanced situations like "business class is acceptable for flights over six hours". Still, using AI to judge expense claims raises questions that Microsoft hasn't answered publicly:

• False positives and misses: How often will it flag legitimate claims, or overlook real violations? The roadmap doesn't say how errors are measured or handled.
• Explainability: The approver gets "context," but in what form and how much? Without published UI details, nobody outside Microsoft knows yet.
• Audit trail: Can auditors see the agent's findings later, and what happens if an approver overrides a flag? Not yet documented for this feature.
• Policy authoring: Natural language is flexible, but a vague policy will likely give vague guidance. This is general industry experience with language-model tools, not a documented Microsoft requirement.

For a business, the human-in-the-loop design is the selling point. The approver still makes the decision and is still accountable for it. The AI's job is to point out what deserves attention so approvers don't have to dig through every receipt.

What Business Central admins should do now​

1. Separate your rules from your policies. Put clear numeric limits (receipt thresholds, banned categories) in Expense Management Rules, where they can block submission. Keep the judgment-based guidance for the natural-language policy layer.
2. Rewrite your travel and expense policy. Replace phrases like "use good judgment" with specific conditions and examples, the kind of thing a new employee could actually follow.
3. Check your environment. The listing names the Worldwide (Standard Multi-Tenant) cloud on web. That doesn't guarantee every tenant, country or language will get the preview in October. The broader Expense Agent preview started in English in the US and expanded to Australia, New Zealand and the UK later.
4. Pilot with approvers. In preview, compare the agent's flags with what your experienced approvers would have caught. That will show you how much it actually helps.
5. Expect changes. Microsoft's roadmap dates are estimates, and features can slip or change before GA.

The bottom line​

With Expense Agent, Microsoft plans to take Business Central expense control beyond hard limits into context-dependent policy, the territory approvers have always handled with experience. Microsoft has scoped it cautiously: checks run after submission, the first release works at the line level, and a human makes every final call. If preview arrives in October as planned, finance teams will soon see whether the agent's judgment is good enough to help them.