AI-Powered Transaction Matching Assistant for Bank Reconciliation

Oracle has introduced the Transaction Matching Assistant, an AI layer in the bank reconciliation workflow that uses OpenAI GPT 4.x to resolve ambiguous matches between imported bank transactions and GL transactions. This is not a SuiteScript or REST API change — it is a feature-flag-gated addition to the existing Match Bank Data page.

What changed

• New matching stage. After system rules, custom rules, and enrichment-based matching (if enabled) have run, the Transaction Matching Assistant can evaluate remaining candidates when multiple GL transactions still match a single bank transaction. It scores on payee/payor name, memo content, and date proximity.
• One-to-one matches only. The assistant does not handle many-to-one or one-to-many scenarios. It picks the single most likely GL match from the remaining candidate set.
• No learning. The model evaluates each transaction in isolation. It does not train on your historical accept/reject decisions and will not improve over time within your account.
• Results surface on the Review subtab of the Match Bank Data page, marked with a message indicating AI assistance. Matches are not auto-submitted — a user must explicitly confirm.
• Undo behavior is unchanged. Undoing an AI-recommended match returns both the bank and GL transaction to the To Be Matched subtab.

Feature enablement

• Navigate to Setup > Company > Enable Features > Accounting subtab.
• For eligible accounts created after May 14, 2026: enabled by default.
• For eligible accounts created before May 14, 2026: disabled by default — must be manually enabled.
• Eligibility depends on region. Oracle does not specify which regions in this note — verify your account's eligibility in the Enable Features page or contact support.

Permissions

No new permissions are required. The assistant uses the same permission set as the Match Bank Data page. See Oracle's Permissions for Banking Features documentation for the full list.

What to do

1. Evaluate your reconciliation workflow. If your team already uses system rules, custom rules, and enriched bank data, this adds a final AI tiebreaker step. Decide whether that fits your internal controls and audit requirements.
2. Review data privacy implications. The feature sends transaction details (payee/payor names, memos, dates, amounts) to OpenAI GPT 4.x. Confirm this aligns with your organization's data handling and vendor policies before enabling.
3. Enable the feature at Setup > Company > Enable Features > Accounting if your account was created before May 14, 2026, and you want to opt in.
4. Test in Sandbox first. Enable the feature in a sandbox account and run reconciliation against a representative bank statement to evaluate match quality before rolling out to production.
5. Train reconciliation users. Matches recommended by the assistant appear on the Review subtab with an AI indicator. Users must understand that these are suggestions, not confirmed matches, and that the model does not learn from corrections.

Gaps in the release note

• Oracle does not specify which regions are supported or what makes an account "eligible."
• There is no mention of governance impact, SuiteScript hooks, or workflow action triggers tied to AI-recommended matches.
• It is unclear whether the assistant's recommendation metadata (e.g., confidence score) is exposed on the transaction record or available via N/search or N/query. If you need to report on AI-assisted matches programmatically, investigate the banktransaction and reconciliation record types for new fields.
• No details on rate limits or throughput constraints when processing large volumes of ambiguous matches.