Skip to content

AI-assisted internal-audit fieldwork

Compare the right values against an explicit rule.

Extract, Reconcile and Checks are capabilities inside Tests. The Recipe defines the comparison; a Run records the execution.

Book a demo

Illustrations explain the product model. Confirm your procedure, formats and review requirements during evaluation.

Tests / Extract, Reconcile and Checks

Choose the sources and matching rules.

An invoice-to-PO comparison uses a purchase-order reference and per-field source authority. The product model describes exact, normalised, fuzzy and semantic matching, with ambiguous cases escalated to people.

One-to-one, one-to-many and many-to-one matching need explicit grouping rules. A three-way procedure adds goods-receipt evidence; the illustrated subtotal check does not establish delivery or payment.

₹5,000 beyond the approved PO.

Expected ₹1,20,000; actual ₹1,25,000; absolute difference ₹5,000; tolerance ₹1. This is an Amount mismatch under Synthetic Procurement Policy v3 §4.2.

A separate supplied-ledger loan example compares ₹35,00,000 sanctioned with ₹36,00,000 disbursed. It is not a P2P screen or a regulatory assertion.

Illustrative example · invented data · not a product capture

P2P — Q1 FY27 · DEMO-0042

P2P amount check

Fail
Expected₹1,20,000
Actual₹1,25,000

Amount mismatch · Awaiting reviewer confirmation

DEMO-0042.pdf · Page 1 · subtotal region
Role / version
Primary evidence · v1
Original → normalised
1,25,000.00 → ₹1,25,000
Method / corrections
Illustrative extraction · None in illustration
purchase-orders.xlsx · Orders!H43
Role / version
Source of truth · v1
Original → normalised
120000 → ₹1,20,000
Method / corrections
Illustrative cell read · None in illustration
Synthetic Procurement Policy v3 §4.2

Compare invoice subtotal with approved PO amount. Absolute difference must not exceed ₹1. If the approved PO is unavailable, report Insufficient evidence.

Effective 2026-04-01 · Recipe v3

Invented source excerpts; no original files are downloadable. Confidence and exact page regions need evaluation against actual evidence.

A check can use one source.

A configured check can compare a population with a formula, threshold, date rule or reference list. Explain failed results with a typed reason: amount, date/cut-off, duplicate usage, party variance, tax-field mismatch, missing document or unsupported evidence.

Missing evidence is reported separately unless the configured procedure explicitly defines its absence as a failure.

Illustrative Run snapshot · DEMO-RUN-018

180 of 190 eligible records completed.

One check per record · Q1 FY27. Completion does not resolve human review.

Received
200
Eligible
190
Execution completed
180
Excluded
10
Awaiting evidence
6
Processing failures
4
  • Pass157
  • Fail17
  • Insufficient evidence6
  • Needs human review6
  • Not applicable0
  • Processing error4

Outcomes total 190 eligible records; excluded records are separate. Missing evidence is reported separately unless the configured completeness procedure explicitly defines its absence as a failure.

Recipe v3 · Inputs v1 · illustrative engine demo-1 / model fixture-1. Completed Run history is immutable. A new Recipe version creates new work; it does not rewrite this snapshot.

Draft / review incomplete · Finding: Not raised · 0 confirmed exceptions.

Does this example confirm current availability?

The example explains the product model with synthetic data. Confirm the procedure, formats and review behaviour you need in an evaluation; a labelled illustration does not replace product evidence.

Your next procedure

Start with the work you need to review.

Discuss the sources, rules and output your team needs. Please keep confidential audit evidence out of the enquiry form.

Book a demo