Financial forensic intelligence for work that has to be right.

Connected risk

Risk does not stay in one row.

When one record is flagged, Perch scores how that risk spreads across the connected entities, the vendor, its invoices, and the payments around it, so you see the cluster that needs attention, not just a single line item.

What it checks

The real controls, run on your files.

These are the procure-to-pay and financial controls a reviewer actually runs. Perch runs them across the documents you point it at and returns the exceptions with their evidence.

Three-way match

Match invoices to their purchase orders and payments, and flag exactly where the three break.

Duplicate detection

Surface likely duplicate invoices and payments before they turn into cash leakage, with the records side by side.

Payment mismatches

Catch payments that do not match the invoice they were posted against, with the variance quantified.

Invoice over approval

Flag invoices that exceed the purchase order they were approved under, even when vendor totals look fine.

Vendor master checks

Find invoices from vendors that are not in the approved vendor master, a common path for leakage and fraud.

Aging and reconciliation

Surface past-due AP exposure and reconcile ledger activity, flagging the entries that do not tie out.

Why it is different

A finding you can stand behind.

A general assistant will describe an audit. Perch runs one. The difference is determinism and evidence on every result.

Computed, not guessed

Totals, matches, and reconciliations run through deterministic computation, not model arithmetic, so every figure can be traced to how it was produced and reproduced exactly.

Every finding carries its evidence

Each exception comes with severity, financial exposure, the source files behind it, and a recommended action. It is a finding you can act on, not a confidence score with nothing underneath.

Multiple agents and real tools

Perch coordinates focused agents that read the files, run the checks, and verify each other's work using real tools, and the work is visible rather than a black box.

You make the call

Perch surfaces the exception and the evidence. The judgment stays with your team. It is built to assist a reviewer, not to quietly act on your books.

How we test it

Measured against known answers.

Perch's financial checks are benchmarked against real-world financial datasets with known-correct answers, including labeled test labs where issues are planted on purpose. A duplicate invoice is in there deliberately. A figure that contradicts its support is in there deliberately. We measure whether Perch finds the real problems and, just as importantly, whether it resists confident wrong answers, which are penalized hardest because a total that looks right but is wrong is the most expensive outcome in financial work.

This is validation against real-world datasets, not a guarantee on your specific books. The only real test is your own files, which is exactly how we prove it with a pilot.

For teams

Managed workflows you can enable.

Beyond the core checks, Perch runs managed, deterministic workflows that teams can turn on: GL reconciliation, statement and earnings audits, month-end close support, and payroll review. Each produces structured, source-backed output your team can review, check, and export, with the same evidence-first discipline as the rest of the system.