---
title: "Invoice Classification + Anomaly Detection | AimRank"
url: https://www.aimrank.io/products/invoice-classification/
description: "Auto-classify invoices to the right GL codes and flag the anomalies - with an audit trail your accountant trusts. AI prototype from €1,290."
language: en
updated: 2026-10-09
---

[Products · #12 · Documents & Language](https://www.aimrank.io/products/)

TIER A BLUEPRINT

# Invoice Classification + Anomaly Detection

Auto-classify invoices to the right GL codes and flag the anomalies - with an audit trail your accountant trusts.

scaffold from **€1,290**

## How it works

01

### Connect the accounting system

i.MAS, Sąskaita123, Rivilė or your stack - read-only. Trains on each client's own historical coding: their past decisions are the labels.

02

### Classify per client taxonomy

Lithuanian-language tuned - generic OCR misreads LT invoice text; this doesn't. Each client's GL/VAT preferences learned separately.

03

### Confidence routes the doubt

High-confidence postings flow through; anything uncertain lands in the review queue with its best guess. It never silently miscodes.

04

### Anomalies caught in real time

Duplicates, VAT mismatches, vendor changes, Benford outliers - flagged on arrival with the reason, not discovered at quarter-end.

## Methodology reference

The published method this build is based on, and what it did in production.

**Intuit** built Rel-Cat to auto-code QuickBooks transactions to the right account, beating its prior production model (top-1 68.7% vs 62.5%).

Relational transaction categorization · as reported by [arXiv ↗](https://arxiv.org/abs/2506.09234)

Referenced for methodology only. AimRank is not affiliated with, endorsed by, or a customer of these companies; each links to its own public source. Sources last checked September 2026.

## The Four Guarantees™ - this build

The scaffold wires all four in. The PoC proves them on your data, as of that day. Only Implementation keeps them true over time: watched, drift-checked, with recourse.

### Measured value

\~70% of manual coding eliminated + anomaly catches (duplicates, wrong VAT, vendor fraud). Classification-accuracy gate per client taxonomy.

### Defensible

A reason for every classification and every flag - an audit trail your accountant, and their auditor, accepts. GDPR Art. 22 per anomaly.

### Self-improving

Calibrated confidence routes low-certainty invoices to review; corrections feed back per client; new-vendor + category drift watched.

### Yours & everywhere

Accounting-system write-back; your cloud, full source. MCP endpoint your accountant agents can call.

## The number, sized honestly

Reference buyer: Lithuanian accounting firm or CFO office managing \~50 SMB clients, \~10,000 invoices/month, \~3 hrs/week per client on coding + checks.

**\~70%** of manual coding eliminated - the agreed number (conservative)

**€80-120k** per year ≈ 3 FTE-equivalents of coding work, at firm scale

**\~2-3 wks** payback at the PoC price, reference volume

**€10-30k** per year in caught duplicates, VAT errors and vendor fraud

## Three ways to own it

| Tier | What you get | Price | |
| ---------------------- | ------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- | -------------------------------------------- | |
| **Scaffolding** ★ | The full repo - classifier + anomaly layer + review queue, LT-language baseline, VAT taxonomy, audit log, MCP server. | **€1,290** | |
| **PoC** ★★RECOMMENDED | Trained on your invoice history and GL codes - coding accuracy and anomaly quality measured against your accountant's decisions. Code + quality report yours. No guarantee at PoC. | from**€3,900** | |
| **Implementation** ★★★ | Production: accounting-system integration, review-queue routing, duplicate blocking live, drift monitoring, retraining, monthly accuracy report - the agreed number (conservative) attaches here. | from **€12,000**\+ from €900/month assurance | |

★ = engagement depth. **PoC is the recommended path**: quality proven on your data before production money. The PoC carries no performance guarantee by design; the agreed number (conservative) attaches at Implementation, informed by the PoC report.

## What we don't promise

**Every client codes differently - and the model has to learn each one.** Expect 2-3 weeks of human-in-the-loop tuning per client before steady-state accuracy; we roll clients in waves, not all at once. Write-back stays gated per client until that client passes the accuracy bar - your firm's reputation posts nothing it isn't sure of.

## Ready to see your own number?

Request the build: within 48h you get a personal reply with the value sized to your volume.

No commitment · reply within 48h · your data stays in the EU