FraudLensAI
Sign inGet started

Explainable fraud-risk screening for finance teams

Surface the records worth a second look, with the evidence attached.

FraudLens AI screens transaction exports and invoices for duplicate, unusual and inconsistent entries, ranks them for review, and keeps a clear record of every decision your team makes.

Create an account

The demo uses synthetic data in an isolated workspace. No bank connections, no real records.

Findings · Q2 vendor payments

205 rows analyzed · 15 findings

Illustrative · synthetic data
  • 35

    Duplicate transaction reference

    DUP-001 · TXN-1042 · Brightline Office Supply · 612.40 USD

  • 40

    Unusually large amount

    STAT-001 · TXN-2009 · Fulcrum IT Consulting · 98,500.00 USD

  • 20

    Possible repeat payment

    DUP-002 · TXN-2002 · Apex Logistics LLC · 2,450.00 USD

  • 8

    Vendor name variation

    VEND-001 · TXN-2010 · Northwind Tradres · 845.00 USD

Scores are review priorities from triggered rules, not fraud probabilities.

Manual review doesn't scale

Teams scan thousands of rows in spreadsheets, and spot checks vary from reviewer to reviewer.

Subtle signals get missed

Repeat invoices, near-identical vendors and out-of-pattern amounts hide in ordinary-looking exports.

Alerts without context stall

A bare anomaly score doesn't say what changed or what to verify next, so it gets ignored.

How it works

From raw export to documented decision

  1. 01

    Upload and map

    Drop in a CSV export. FraudLens suggests column mappings, shows a preview and reports invalid dates, amounts and currencies before anything runs.

  2. 02

    Run transparent checks

    Versioned rules look for duplicates, near-duplicates, data-quality gaps, unusual amounts and vendor look-alikes. The same input always gives the same result.

  3. 03

    Review the evidence

    Every flag shows the rule, the matched rows, the baseline it was compared against and how its review priority was calculated.

  4. 04

    Decide and document

    Move findings to In review, Escalated or Dismissed, add notes, and export a report. Every action is recorded in the audit trail.

Capabilities

Deterministic checks first. AI only where it helps.

The core engine is rule-based and runs entirely without an AI provider, so every result can be reproduced from the stored input and rule version.

Duplicate detection

Exact duplicate references and invoice numbers, plus same-vendor, same-amount payments within a configurable date window.

Robust amount outliers

Median and MAD baselines per currency, with minimum sample sizes. Groups too small to judge are skipped, never guessed.

Vendor name variations

Highlights near-identical supplier spellings so look-alike or duplicated vendor records can be checked.

Data-quality validation

Missing fields, unparseable dates and invalid currency codes are surfaced as low-severity signals, not accusations.

Your own policy rules

Optional amount thresholds and watched categories reflect your approval policy. They are off unless you turn them on.

Invoice field extraction

Pull vendor, number, date and totals from PDFs and images with visible confidence. You confirm every field before checks run.

Review workflow and audit

Statuses, notes and a full history for every finding, scoped to your account and exportable for follow-up.

Optional AI summaries

Plain-language summaries of the existing evidence, off by default. They cannot create findings or change a score.

Principles

Built to support reviewers, not replace them

Explainable first

No black-box verdicts. Each finding links to the exact rule, version and source rows behind it.

Review priority, not probability

Scores rank what to look at first. They are heuristics, not a measure of how likely fraud is.

Humans decide

FraudLens never blocks payments, freezes accounts or labels anyone as fraudulent.

Privacy by design

Synthetic demo data, owner-only access, deletion you control, and no data sent to an AI provider unless you opt in.

See it on a synthetic dataset

The demo loads 205 synthetic transactions with planted duplicates, outliers and data-quality issues.

Create an account