Accounting Accruals Agent: How to Test AI Accrual Proposals

To test the Accounting Accruals Agent, treat every proposal as untrusted until you have checked it. Trace the policy rule the agent cites, rebuild the amount from source data yourself, and prove that each accept, adjust and reject path behaves the way your close control requires. Before any of that, confirm which release and edition your tenant runs, because the sources disagree on availability.

Start with the release and edition check

The agent is an S/4HANA Cloud feature, not an on-premise one. A vendor blog says the agent and the 2602 features are available exclusively to SAP S/4HANA Cloud Public Edition customers. SAPinsider describes it the same way, saying the agent works through Joule in SAP S/4HANA Cloud Public Edition and is currently under Restricted Availability. Joule itself is a separate dependency. One third-party analysis notes that classic on-premise S/4HANA customers currently have no roadmap commitment for native Joule access, so do not assume a Private Edition or on-premise system can run this.

The status is less clear than the marketing suggests. An SAP Community post by SAP says the agent was a beta version in SAP Cloud ERP 2602, with general availability planned for that release. A later SAP post says the agent is going to be made generally available in SAP Cloud ERP 2602.1. Those dates are not the same thing as what is switched on in your tenant. The same SAP Community post series says the dedicated Manage Accrual Proposals app is planned for release by end of September 2026. Today is 1 November 2026, so that date has passed. I could not confirm from the sources I checked that the app has shipped. Check your tenant's Fiori launchpad for the app, then check the release notes and the feature scope description for your exact version. Write the version into your test evidence.

Also check what the feature scope actually covers. The SAP feature scope document describes the agent as reading an uploaded accruals policy handbook and preparing proposals from it and from historical data. It also lists notification of accountants when proposals are ready. Your test scope should match that list, not the marketing summary.

What changed from ECC and from the classic Accrual Engine

In ECC, most manual accruals lived in spreadsheets and were keyed in by hand. S/4HANA added the Accrual Engine, which is a rules-driven calculator. One SAP Community guide notes that the Accrual Engine postings are stored only as line items of the universal journal entry, in table ACDOCA. The engine handles manual accruals and purchase order accruals through accrual objects, job templates and configured item types.

The Accounting Accruals Agent is a different design. The classic engine calculates from configured methods. The agent reads a policy document and proposes amounts, with reasoning attached. That means your test base changes. You are not only testing configuration; you are testing whether an interpretation of prose policy produces the same result an accountant would reach. Keep the classic engine in your regression scope too. If your company already runs purchase order accruals through it, the two will coexist in the same period close.

Step 1: Test the policy handbook as an input

The agent is only as good as the policy text it reads. Start by listing every accrual rule in the handbook, and give each one an ID. For each rule, record the expected accrual type, the company code, the accounts, the calculation method and the materiality threshold. The vendor guidance also says you should review materiality thresholds so you know which open items get a proposal at all.

Then write tests in three groups:

• Positive cases: one item per rule that should produce a proposal, including the boundary amounts either side of each threshold.
• Negative cases: items that look similar but should not be accrued, such as an invoice already received or an item below materiality.
• Ambiguous cases: rules with two plausible readings. Record how the agent reads them, and have the policy owner decide the correct answer before you sign anything.

SAP provides example templates for utilities and audit accruals as a starting point. Use them to see how the policy should be written. Do not treat them as your policy.

Step 2: Recalculate every proposal outside the agent

This is where most of the evidence comes from. For each test item, calculate the expected accrual in a controlled spreadsheet or query from the source data: open purchase orders, service entries, contract schedules or prior-period postings, depending on the rule. Compare your figure to the proposal. Record the difference and the tolerance you agreed with the controller.

The agent is designed to explain itself. Each proposal should show the policy rule used and an explanation of how the amount was calculated. Test that the explanation is correct, not only that it exists. If the text cites rule 4 but the number follows rule 6, that is a defect even when the amount is right.

It helps to compare against the classic engine's review logic. In the Accrual Engine, a proposed amount below a lower threshold is not processed, an amount between thresholds is auto-approved, and an amount above the upper threshold needs review. SAP Learning describes the approval flow with the upper and lower thresholds, and it is set per company code, accrual item type and ledger group. If the proposed accrual amount is above the upper threshold, review or approval is required. If it sits between the lower and upper threshold, approval is granted automatically. Ask the product owner whether the agent has an equivalent gate. Do not assume it does. Test whether it does.

Step 3: Test duplicates, periods and company codes

Duplicate postings are the most expensive failure in an accrual run. The agent is documented to check whether accruals have already been posted for the period to prevent duplicates. Test that claim directly:

• Generate a proposal for a period, then generate it again for the same period and company code. Confirm that the second run does not create a second set of proposals.
• Post one proposal, then regenerate. Confirm the posted item is excluded.
• Run the agent for a period that has already closed, and confirm that the behavior matches your close calendar.
• Run two company codes in one request, and confirm that each proposal lands on the right company code and ledger.

The agent is meant to work for a particular period and company code. Your test cases should prove the boundary, not just the happy path.

Step 4: Test every review path and the posting

The documented workflow lets accountants accept, adjust, or reject proposals before posting. Test all three. Accept should post the entry as proposed. Adjust should post the changed amount and keep the reasoning on record. Reject should post nothing and leave the item available for the next run.

Then test who can do each step. One vendor blog says the new Fiori app requires appropriate business role assignments, and the same post recommends defining who should review and approve proposals with your Basis team. Build a role test with at least three users: the preparer who triggers runs, the GL accountant who reviews, and a controller who approves over threshold. Confirm that a preparer cannot approve their own proposal. Use the same discipline you would apply to any Fiori role test. For the broader authorization approach, see our Joule agent authorization testing guide.

1. Policy inHandbook version and rule IDs loaded and locked
2. Proposal runOne period, one company code, logged with the run ID
3. Independent checkYour recalculation matched within agreed tolerance
4. Review pathAccept, adjust and reject tested, with the right role per user
5. Post and reconcileJournal entries tied back to proposals and to the reversal plan

Do not stop at the posting. Reconcile the posted entries to the accrual population, and check reversal behavior in the next period. A proposal that posts correctly and then reverses to the wrong account is still a failed close.

Step 5: Test repeatability, because the output is not fixed

An AI proposal can change when the same input is run again, when the policy wording changes or when the data changes. Run each positive test case at least three times on the same data, and record any variation. Repeat after any policy edit. Then decide with the controller what variation is acceptable. A zero-variance rule is easy to explain. A variation that changes which entries post is not acceptable without review.

Our AI testing course covers evaluation sets and sign-off for non-deterministic features in more depth. Start with the free course on testing Joule agents and AI features, then the course on how SAP's agents work and how to evaluate them. For a wider view of how Joule fits into an S/4HANA program, read SAP AI for Practitioners. If you are still deciding which Joule features your edition supports, see Joule in S/4HANA Private Edition by line of business.

Where the other finance agents fit in your plan

Accruals are not the only embedded finance AI in play. SAP's H2 2025 innovation guide describes a Cash Management Agent that gathers and reasons over daily bank statements to support reconciliation. Intercompany matching is on SAP's roadmap, but the dates in the sources conflict. One SAP Community post lists Intercompany Matching (GA March 2027), while SAP's own release-level posts should be the reference. Treat any roadmap date as a planning assumption until it appears in the release notes for your tenant. Use the same test approach for each agent: a documented policy or rule set, an independent expected result, a role test, and a repeatability check.

The test design also matters for the SAP Joule acceptance work. Our SAP Joule and embedded AI feature acceptance checklist can serve as the starting point for the sign-off form. Add a line for each accrual rule ID so every proposal can be traced back to the policy it used.

Frequently asked questions

Is the Accounting Accruals Agent available in S/4HANA on-premise or Private Edition?

The sources I checked tie the agent to S/4HANA Cloud Public Edition. The SAPinsider article says it works through Joule in Public Edition, and a vendor blog says it is exclusive to that edition. For Private Edition or on-premise, ask SAP through your account team and check the release documentation for your version before planning any test.

Does the agent post journal entries on its own?

The documented workflow has the accountant accept, adjust or reject proposals before posting. SAP's feature scope describes the agent as handling preparation and posting end to end, and the SAP sources do not settle the question for every configuration. Test your own configuration to see whether any step posts without a human action.

How is this different from the Accrual Engine I already test?

The Accrual Engine calculates accruals from configured methods, thresholds and accrual objects. The agent reads a policy handbook and proposes entries with reasoning. Keep both in regression scope. The Accrual Engine remains the reference for purchase order accruals and periodic posting runs, and the agent is an additional source of proposals that must be reconciled to them.

What should I do if the Manage Accrual Proposals app is missing in my tenant?

Do not assume the feature is live. Confirm the release version in your tenant, check the app in the Fiori launchpad, and read the release notes for that version. If the app is not there, record it as a scope gap and plan the test against the release that delivers it. Recheck the SAP Help documentation and your SAP account contact before the next close.

Sources

1. Automating Month-End Accruals with the SAP Accounting Accruals Agent, SAP Community, 2026
2. Feature Scope Description for Accounting Accruals Agent, SAP Help Portal
3. SAP Releases Accounting Accruals Agent for Month-End Close, SAPinsider, 2026
4. Finance in SAP Cloud ERP 2602, SAP Community
5. My highlights of the SAP Cloud ERP 2602 release, SAP Community
6. SAP S/4HANA Cloud 2602 Accounting Accruals Agent, SAVIC Technologies (third-party), 2026
7. Schedule Accruals Jobs, SAP Help Portal
8. Accruals Management - Purchase Order Accruals, SAP Learning
9. Manual Accrual Engine step-by-step configuration in SAP S/4HANA, SAP Community
10. 7 SAP AI Use Cases for Finance That Cut Close Time, Cremencing (third-party), 2026
11. SAP Innovation Guide H2 2025, SAP
12. Agentic AI and Intelligent Assistants: The New Change Agents in S/4HANA, SAP Community