Skip to content

Three-way match

Every invoice held to its order and its receipt, line by line.

Upload the AP ledger, the purchase orders, the goods receipts and the supplier invoices. Deixio proposes the matching and the tests, code compares every line, and you decide each exception.

Illustrative overview of Three-way match end to end: your intent, tests the AI proposes and you approve, the run, results you decide and the workpaper you sign, then run again next period.
Illustrative · sample data

Three files, two formats, one question per line.

Quantities and prices sit in the order file, the receipt file and the invoices, and the invoices arrive as PDFs and scans. Checking every line by hand is rarely possible, so the work is usually sampled.

From the plan to the findings, every decision is yours.

02Tests · The AI proposes, you approve

Tests drafted from the evidence and your intent.

The AI looks through the evidence and proposes each test with what it proves, its population, its sources and what counts as an exception. It lists what it noticed and asks its questions. Nothing runs until you approve the plan.

  • Billed quantity above received
  • Price against the order price
  • Invoice with no order
  • Invoice with no receipt
  • Duplicate invoices

It asks what it needs to know, for example: “What unit price tolerance should apply?”

Illustrative example: tests for three-way match drafted by the AI, with one added by the auditor, and an approve control.
Illustrative · sample data

At a glance: Three-way match

What you upload

  • The AP ledger
  • Purchase-order lines and goods receipts (CSV or XLSX)
  • Supplier invoices and credit notes (PDF, scans, Word and more)

Tests the AI may propose

  • Billed quantity above received
  • Price against the order price
  • Invoice with no order
  • Invoice with no receipt
  • Duplicate invoices
  • Currency
  • Extension errors
  • Items in one source only

The AI proposes tests from your evidence and intent; these are examples, not a fixed list.

What you get

  • Results per test
  • Findings with the compared values side by side and their numbered sources
  • The workpaper, with its exports

What the auditor sees.

Illustrative example: a findings list where one finding is confirmed by a reviewer with a date, and others wait for a decision.
Illustrative · sample data

Every decision is a person’s.

Findings wait for a decision: confirm, reject or ask the client. Signing is blocked while any finding is open, and each decision records who and when.

Illustrative example: a figure opened beside its source page, with the box where it is printed, the AI’s confidence and who accepted it.
Illustrative · sample data

Every figure explains itself.

Open a figure to see the page it was read from, the box around it, how sure the AI was and who accepted it.

Watch one procedure run, figure by figure.

In 30 minutes we run a procedure end to end on sample data and show you where every figure comes from.

  • A source for every figure
  • Human in the loop
  • Append-only, hash-chained audit trail

Questions about three-way match

Does it match every line, or a sample?

Every line the approved tests cover. Sampling only happens if you add it as an explicit step.

Can one invoice match several receipts?

Yes. Matching handles one-to-one, one-to-many and many-to-one, within a date window you approve.

Who sets the tolerance?

You do. The AI asks, for example “What unit price tolerance should apply?”, and Deixio applies your answer to every row and shows its working.

What about invoices in another currency?

Amounts in two currencies are never compared or added without an instruction you approve.

What if an invoice is a poor scan?

Where code cannot read a figure itself, a stronger model reads it from the page picture and the figure is labelled so. Below your firm’s confidence line, it waits for your approval.