Athena — vision.md

Athena — Vision

Superseded scope notice. This document describes the original, narrower v1 vision (liquide-middelen only: kiting, unsupported manual journal, restricted cash/ G-rekening, lapping). The owner's data-audit-engine-requirements-source.md (2026-09-01) explicitly supersedes/expands that scope into a general ledger–wide "Data Audit Engine" (population analysis, JET, revenue, margin, debtors, sampling, workpaper, findings). The authoritative, numbered requirements are now in srs.md, sequenced for delivery in roadmap.md. srs.md §3 documents exactly what carried forward from this document (the four-layer pipeline idea, the determinism/traceability principles, the assist-not-certify stance — all still valid) versus what was dropped (the four liquide-middelen finding types themselves, and the bank-statement-PDF/bank-confirmation-DOCX ingestion paths that supported them — none of these has a corresponding row in the new MoSCoW table). Read this document only for that carried-forward groundwork, not as a still-current scope statement.

What Athena is

Athena is an AI-assisted audit tool focused on the liquide middelen (cash and cash equivalents) balance sheet area for Dutch financial statement audits. It ingests the heterogeneous set of documents an auditor already collects for a cash audit — general ledger bank mutations, XAF-style auditfiles, trial balances, bank statements (PDF), bank confirmation letters (DOCX), bank reconciliations (XLSX), and accounts-receivable receipt registers (CSV) — and cross-references them automatically to surface a specific, well-defined set of audit-relevant patterns that today require an auditor to manually line up dates, amounts, and accounts across multiple files.

Athena does not "audit" a company. It performs the mechanical cross-referencing work that precedes an auditor's judgment call, and it produces flagged, traceable findings for a human auditor to review, investigate, and conclude on.

Who it's for

  • Dutch accountants and auditors performing liquide-middelen substantive procedures, typically at mid-size accounting firms (MKB-gericht kantoren) or internal audit functions of larger entities.
  • The primary user is the field auditor or audit senior doing the fieldwork on the cash balance, not the partner signing the opinion. Athena's output feeds into that auditor's working papers; it is not itself a working paper or an opinion.
  • Users are expected to already understand Dutch audit methodology (NV COS) and professional skepticism norms — Athena assumes an audit-literate user, not a layperson.

Core value proposition

A cash/liquide-middelen audit routinely requires cross-referencing facts that live in different documents and different formats: a bank mutation booked in the general ledger on one date, the same movement clearing at the bank on a different date, a manual journal entry with no counterparty bank confirmation, a balance in the trial balance that is restricted per a bank confirmation letter's wording, or a sequence of customer receipts that were re-applied to the wrong invoices. Finding these patterns by hand means an auditor manually re-keying or eyeballing rows across a CSV, an XML auditfile, two PDFs, a Word document, and two spreadsheets.

Athena's value is doing that cross-referencing automatically and consistently: parsing each source format into a common transaction/balance model, then running a fixed set of detection procedures against that model, and returning findings that state exactly which source documents and which fields produced the flag. This turns a manual, error-prone reconciliation exercise into a repeatable, auditable first pass — reducing the time an auditor spends finding the needle, while leaving all judgment about materiality, disposition, and audit opinion impact to the auditor.

Non-goals for v1 (explicit)

Because "AI tool that produces audit findings" touches professional liability directly, the boundaries below are deliberate and load-bearing, not just scope trimming:

  • Athena is not a full financial-statement audit tool. V1 covers only the liquide-middelen balance sheet area. It does not touch revenue, inventory, fixed assets, or any other audit area, and it makes no claim about the financial statements as a whole.
  • Athena does not replace auditor judgment. Every output is a flagged item for review, not a conclusion. Athena never states that a finding is confirmed fraud, confirmed misstatement, or requires a specific accounting treatment — it states what pattern was detected, in which source documents, with which supporting data points, and leaves evaluation, materiality assessment, and disposition to the auditor.
  • Athena does not certify, sign off, or produce an audit opinion. No output from Athena should be presented to a client or regulator as an audit conclusion. Its output is intermediate working-paper material.
  • Athena does not assert completeness. A "no findings" result from Athena means no patterns were detected by the procedures it currently runs — it is not a statement that the cash balance is free of misstatement, and must not be communicated as such.
  • Athena is not a general bookkeeping, ERP, or import/export tool. It reads audit evidence documents to produce findings; it does not modify source data, does not post journal entries, and does not integrate as a system of record.
  • Athena does not perform legal or regulatory determination. Where a finding touches a legal concept (e.g., a G-rekening's restricted status, a bank guarantee's terms), Athena surfaces the extracted facts; it does not render a legal or NV COS-compliance judgment.
  • Areas visible in the broader internal audit methodology this project originated from — cash counts, foreign currency revaluation, analytical review of cash flow trends, and automated management-letter drafting — are explicitly out of scope for v1. They are not evidenced in the current training/ground-truth data and are not committed to as v1 deliverables (see scope.md and architecture.md for what is in scope, and the note on the source methodology document for what may follow in later phases).

Reacties

Nog geen reacties