Posting QA Agent
Check every posting - before it hits the general ledger.
Checks every posting for formal completeness, plausibility, account consistency and tax code correctness, detects duplicates and calculates an anomaly score for risk-based escalation.
Score Dashboard
What This Agent Does
The Posting QA Agent is the quality filter for the general ledger. Every posting passes through a multi-stage check before it is finally posted. The check ranges from formal completeness (document, account, amount, date) to plausibility (amount within normal range) to consistency (does the VAT code match the posted account).
The Decision Layer breaks the quality check into eight decision steps. Formal check, account consistency and duplicate detection are fully rule-based. The plausibility check uses both thresholds (R) and historical comparison (A). The anomaly score is ML-based and determines whether a posting is auto-approved or escalated for human review.
The result: a 30-40% reduction in correction postings. Errors are caught before posting, not at month-end. The agent is the central quality gate for the entire general ledger and processes the highest daily transaction volume in the GL domain.
Micro-Decision Table
Check formal completeness Are document, account, amount and date present? Rules Engine
Checklist of mandatory fields per posting type
Decision Record
Challengeable: Yes - rule application verifiable. Objection possible for incorrect data or wrong rule version.
Plausibility check Is the amount within normal range for this account? Rules Engine
Threshold check rule-based (R), historical comparison by ML (A)
Decision Record
Challengeable: Yes - rule application verifiable. Objection possible for incorrect data or wrong rule version.
Check account consistency Do debit and credit accounts match? Rules Engine
Double-entry bookkeeping - deterministic check
Decision Record
Challengeable: Yes - rule application verifiable. Objection possible for incorrect data or wrong rule version.
Check tax code consistency Does the VAT code match the posted account? Rules Engine Auditor
Mapping table of account to permitted tax codes
Decision Record
Challengeable: Yes - rule application verifiable. Objection possible for incorrect data or wrong rule version.
Challengeable by: Auditor
Period check Is the posting to the correct accounting period? Rules Engine
Date comparison: document date vs. open accounting periods
Decision Record
Challengeable: Yes - rule application verifiable. Objection possible for incorrect data or wrong rule version.
Duplicate detection Does an identical or similar posting already exist? Rules Engine
Pattern match on amount, account, date and reference
Decision Record
Challengeable: Yes - rule application verifiable. Objection possible for incorrect data or wrong rule version.
Calculate anomaly score How likely is an error or unusual posting? AI Agent
ML-based score from historical patterns and context data
Decision Record
Challengeable: Yes - fully documented, reviewable by humans, objection via formal process.
Decide routing Is the posting approved or escalated for review? Rules Engine
Score threshold determines the escalation path
Decision Record
Challengeable: Yes - rule application verifiable. Objection possible for incorrect data or wrong rule version.
Decision Record and Right to Challenge
Every decision this agent makes or prepares is documented in a complete decision record. Affected parties (employees, suppliers, auditors) can review, understand, and challenge every individual decision.
Prerequisites
- ERP system with posting interface (SAP FI, DATEV, Sage or equivalent)
- Chart of accounts with mapping table (account to tax code)
- Historical posting data for ML-based plausibility check (min. 12 months)
- Defined thresholds and escalation rules
Governance Notes
No human decision in the standard flow (0H / 6R / 2A). The agent checks and escalates - the final decision on escalation rests with the clerk. UStG (tax code consistency), HGB (double-entry bookkeeping) and GoBD (completeness, accuracy, timely recording) as direct legal bases.
GoBD-compliant: every check is logged with result and applied rules. Escalations are documented with anomaly score and escalation reason. The agent reduces the risk of incorrect tax reporting, which is directly relevant during tax audits. Paragraph 203 StGB relevant: posting data contains complete business transactions.
§203 StGB-relevant data is encrypted end-to-end and never passed to AI models in plain text.
Process Documentation Contribution
Infrastructure Contribution
The Posting QA Agent is the central quality gate for the general ledger. Its anomaly score pattern is reused by the Fraud Detection Agent and all agents with risk-based escalation. The tax code consistency check is the foundation for the VAT Return Agent. The duplicate detection is used by all agents that create postings. The escalation logic (score threshold determines path) is the reference pattern for all quality gates in Finance. Builds Decision Logging and Audit Trail used by the Decision Layer for traceability and challengeability of every decision.
Related Agents
Provisions Agent
Calculate, value and post provisions - from leave to litigation costs.
Accruals Agent
Automate period-end accruals - prepaid expenses, deferred income and reversal.
Revenue Recognition Agent
Revenue recognition per IFRS 15 - from contract analysis to disclosure note.
Frequently Asked Questions
How is the anomaly score trained?
The score is based on the company's historical posting data. After a learning phase of at least 12 months, the model recognises industry-specific patterns. The score is continuously calibrated based on confirmed errors and false positives.
Does the check slow down the posting process?
No. The check runs in real time and typically takes under one second per posting. Only on escalation is the process interrupted - affecting fewer than 5% of postings after the introduction phase.
Can the agent also check mass postings (e.g. from payment processing)?
Yes. The agent processes individual and mass postings equally. For mass postings, each line item is checked individually. The check scales linearly - even thousands of postings per day are no problem.
Implement This Agent?
We assess your finance process landscape and show how this agent fits your infrastructure.