Skip to main content
Evidence-first financial review

Find hidden financial leaks. See the evidence.

LeakHunt AI analyzes invoices and expense data to surface potential duplicates and supplier price changes — with the source records and calculations behind every finding.

Core detection works without an AI provider.

Illustrative exampleHigh priority
Potential duplicate invoice
Same supplier, invoice number, amount, and currency.
SupplierNorthwind Office Supplies

Record A

INV-2031

12 Mar 2026

$1,248.00 USD

Record B

INV-2031

18 Mar 2026

$1,248.00 USD

Supporting evidence

  • Supplier name matches exactly
  • Invoice number matches exactly
  • Amount and currency match exactly
  • Line items are 96% similar
No separate payment record is linked, so this is a duplicate invoice finding, not a duplicate payment.

Estimated potential impact

$1,248.00 USD

Awaiting review

Potential only — a user must confirm the issue before it counts as confirmed savings. Synthetic data for illustration; not a customer record.

Exact decimal arithmetic
Evidence attached to findings
Human-reviewed findings

Where money leaks

The leaks rarely announce themselves.

They hide in spreadsheets, repeat across systems, and arrive as small price changes. LeakHunt turns those patterns into reviewable findings.

Duplicate invoices

The same supplier, document number, currency, and total entered more than once. Exact and probable matches are shown separately.

Invoice record duplication, not an automatic payment claim.

Quiet price increases

The same item costs more than last time, but no report flags it. Unit prices are compared across invoices.

A price increase is a signal to investigate, not a recorded loss.

Unusual expenses

An amount far outside the same supplier's own history, with the quartile fence, median, and comparison shown.

An anomaly is not proof of fraud or an overpayment.

How it works

From a messy file to a defensible decision.

No black-box score. You see what was compared, why it was flagged, and what remains uncertain.

  1. 01

    Bring your records

    Import invoices and expenses from CSV or XLSX files. Invalid rows are shown for correction, never silently dropped.

  2. 02

    Inspect the evidence

    Open a finding to see the source records, the amounts compared, the rule that fired, and the confidence level.

  3. 03

    Decide what is real

    Resolve or dismiss a finding after reviewing its evidence, or keep investigating. Every review action is a human decision, and a reviewer can separately record what they accept as real and what was recovered.

Product

Built for the follow-up question.

A finding is not a verdict. LeakHunt is designed to help you ask the right next question with the evidence in hand.

Explore all features

Duplicate invoice detection

Exact matches use supplier, document number, currency, and total. Probable matches use supplier, amount, and currency within three days, with different document numbers.

Supplier price changes

Previous price, current price, absolute and percentage change, and comparison dates. Impact needs quantity data, which is not stored.

Unusual expense flags

Amounts above the supplier's upper quartile fence and at least twice its median, once at least 8 confirmed rows exist for that supplier and currency.

Potential savings with assumptions

Open exact duplicate invoices produce per-currency estimates with the formula and assumptions visible. Separately, a reviewer can record the exposure they accept as real and, later, what was actually recovered.

Source-level evidence

Every finding links back to the invoices, transactions, and calculations that produced it.

Scoped by organization

Row Level Security, server-side authorization, and organization-scoped queries protect records. Original uploaded files are not retained.

Trust

Financial data deserves a high bar.

LeakHunt is built around reproducibility, evidence, and explicit uncertainty — not confident-sounding guesses.

The math is exact

Amounts are stored and compared as decimals, never binary floating point. Calculations are reproducible.

Nothing is invented

Findings are grounded in stored records. Missing evidence is stated, not filled in with guesses.

You decide

No finding is marked confirmed by software. An authorized person reviews the evidence and acts.

Designed for isolation

Every customer record belongs to an organization. Hiding a button is never treated as security.

Common questions

What LeakHunt can and cannot tell you.

Is LeakHunt AI available to use today?

Accounts, isolated workspaces, CSV and XLSX imports, deterministic detection including unusual-expense rules, finding review, recorded savings, CSV exports, and the grounded analyst are available now. PDF and image OCR are not available yet.

Does a duplicate invoice mean I paid twice?

No. A duplicate invoice record shows that two similar records exist. LeakHunt only flags a possible duplicate payment when separate payment evidence supports that conclusion.

Do I need a paid AI provider?

No. The core detection engine is rule-based and works without an AI provider. An optional grounded analyst explains stored findings; it runs locally by default, and a server-side external provider can be configured. Neither mode can confirm fraud or savings.

How is potential savings calculated?

Open exact duplicate invoices produce an estimate of the stored amount multiplied by the number of extra copies, kept separate by currency. These are estimates, not confirmed savings. A confirmed amount and a realized recovery are only ever what a reviewer records on the finding, and a recorded finding leaves the estimate pool so nothing is counted twice.

What file types are supported?

CSV and XLSX are supported now. PDF and image OCR are not available yet.

Bring the invoices. Find the questions worth asking.

Start with the records you already have. LeakHunt shows you where the evidence points — and where it does not.

Early-access build. Accounts, imports, detection, and review are available; payments and billing are not connected yet.