A Guide to Automate Controls with an Invoice Verification Software

Invoice errors and fraud: a massive and underestimated cost

Supplier billing errors and fraud are among the most underestimated sources of financial leakage in corporate operations.

At Koncile, we see it every day, helping businesses detect these issues and implement automated workflows to eliminate them.

According to professional associations, between 0.5% and 5% of all payments made by companies contain errors.

This is especially true in industries with decentralized purchasing or complex pricing structures, such as construction, logistics, energy, or maintenance.

These errors fall into two broad categories:

Human errors

• Double invoicing or duplicate entries
• Price discrepancies between PO and invoice
• Quantity mismatches
• Incorrect tax application
• Misunderstanding of rate cards or contractual terms

Fraudulent behaviors

• Fake invoices issued by malicious insiders
• Tampered or modified documents
• Impersonation of legitimate suppliers
• Manipulation of metadata or supplier identity

These two issues, error and fraud, are often treated as separate problems.

But both can now be tackled through a common foundation: automated invoice verification.

Kiabi: the $100m fraud

In 2023, Kiabi uncovered a large-scale internal fraud case involving its former treasurer, who allegedly issued fictitious invoices to legitimate suppliers while modifying the bank details so that payments totalling over €100 million, were redirected to her own accounts.

No external intrusion was involved, only the absence of a robust, automated control process capable of verifying invoice authenticity and banking data.

An invoice verification software could have flagged these anomalies early on by:

• Detecting duplicates, for instance by spotting unusually high amounts or repeated descriptions for similar services;
• Cross-checking supplier bank details against the master vendor database to detect unauthorized changes.

This case illustrates how easily manual checks and superficial 3-way matching can miss fraudulent activity hidden among thousands of supplier invoices.

What is an invoice verification software?

The term “invoice verification software” can mean several things depending on your context. There are two approaches:

Embedded verification in P2P or ERP systems

In large corporations, [invoice checking] is often built directly into Procure-to-Pay (P2P) workflows.

Here’s how it works:

1. A Purchase Order (PO) is created within the ERP or procurement tool.
2. When goods or services are received, a Delivery Note (DN) is issued.
3. The supplier sends an Invoice, which is matched against the PO and DN.

This “3-way match” ensures that:

Invoice total = PO amount = Delivered quantity × Agreed unit price.

If the values don’t match, the invoice is automatically blocked.

This setup works perfectly in structured environments,  manufacturing, retail, or logistics, where every item is predefined in catalogs and pricing agreements.

However, it’s far less effective in organizations where:

• Purchase Orders are open-ended,
• Pricing is negotiated dynamically, or
• Data formats vary (paper POs, PDFs, emails, Excel rate sheets).

In these contexts, rigid ERP validation rules often lead to false positives, manual rework, and approval bottlenecks.

Independent verification software

An invoice verification software like Koncile provides a more modular layer of control.

Instead of replacing your ERP, it connects via API or even via email forwarding, and performs automated checks across all incoming documents:

• Invoices (PDF, XML, images)
• Purchase Orders
• Rate grids or price lists
• Delivery receipts

If no existing tool is in place, Koncile can automatically fetch invoices from a shared inbox or cloud folder.

From there, the system ingests, analyzes, and reconciles data, detecting discrepancies and even producing ready-to-send credit note requests.

Types of invoice controls that can be automated

Automated verification isn’t limited to price matching.

A robust invoice control system can perform several categories of checks, from simple duplicates to advanced fraud detection.

1. Duplicate detection

An profit recovery firm has recently identified that 80% of its Fortune 500 companies clients have significant double invoicing problem, despite solid P2P system.

They occur when:

• The same invoice is submitted twice under slightly different file names,
• The same PO is billed multiple times, or
• Two subsidiaries receive and pay the same supplier invoice.

An automated verification software detects duplicates by analyzing:

• Invoice numbers and supplier IDs,
• Amount and date similarities,
• Even visual or structural matches using OCR fingerprints.

2. Fraud detection

Organizations that haven’t yet transitioned to e-invoicing are particularly exposed to fraud.

Some software solutions include features to detect fraud based on PDF metadata analysis (timestamp inconsistencies, missing signatures), or even subtle pixel-level modifications and forged seals.

However, many invoices are scanned or lack digital traces.

At Koncile, we’ve found that the most effective way to detect invoice fraud is through contextual analysis.

It means:

• Internal cross-checks within the document — simple but powerful, verifying internal consistency by summing line items against total amounts.
• External cross-checks — comparing invoice data with other invoices to detect unusual patterns or anomalies.
• Supplier validation — cross-matching supplier references (company name, address, VAT number, IBAN) against the official vendor master file to flag unknown or suspicious entities.
• Statistical analysis — identifying anomalies in the number, frequency, or amount of invoices based on historical payment patterns.

3. Legal and fiscal compliance

In the United States, there is no single federal law that dictates a mandatory invoice format or content. However, invoices are still expected to include key details such as the seller and buyer information, invoice number, dates, item descriptions, totals, and payment terms to remain legally valid and auditable. Some states or local tax authorities may impose additional requirements, for instance, including specific tax information or registration numbers, and federal contracts or export transactions can require extra data fields.

In practice, following standard invoicing conventions ensures compliance across all U.S. jurisdictions. The following items are therefore commonly included:

• Legal name and address of supplier and buyer
• Unique invoice number and date
• Purchase order or contract reference
• Employer Identification Number (EIN) or Tax Identification Number (TIN)
• Description of goods or services provided
• Subtotal, tax amount, and total
• Payment terms and currency

In Europe, countries have national requirements on mandatory mentions on invoices, including

Automated systems can instantly detect missing or invalid fields and alert AP teams — a huge help when managing thousands of suppliers across multiple countries.

4. Price and quantity verification

One of the most financially impactful controls is price verification.

Here, the system compares each invoice line against:

• The corresponding Purchase Order,
• The rate card (Excel grid listing product codes and unit prices),
• Or historical benchmark prices paid by other business units.

This allows the company to:

• Detect overbilling,
• Confirm that negotiated discounts are applied,
• Identify unusual price variations across subsidiaries or suppliers.

For example, if one plant pays €4.20 per liter of lubricant and another pays €4.60 for the same product, the system flags the inconsistency — creating opportunities for renegotiation or correction.

Automating invoice control with rate grids

A price grid is an Excel file listing items with their product codes, descriptions, packaging details, and unit prices.

The challenge is to reconcile these grids with invoices automatically.

Here’s how Koncile automates this process.

Step 1: Extract data from invoices

Every incoming invoice is processed through an intelligent OCR (Optical Character Recognition) engine.

The OCR detects and restructures all information:

• Supplier name and details
• Invoice number, date, and total
• Line items: product description, product code (if any), quantity, unit price, total line amount

If invoices are electronic (XML, UBL, EDI), the data is ingested natively, no OCR needed.

This results in a structured dataset that can be used for automated comparisons (and also to build a catalog of all purchased items).

Step 2: Import and normalize the rate grid

The next step is to ingest your rate grid, typically an Excel or CSV file listing products, codes, and prices.

But not all grids are straightforward.

Some include:

• Tiered pricing based on volume,
• Conditional discounts (duration, location, client type),
• Or composite items (bundled services).

Koncile automatically restructures these grids into a normalized, machine-readable table.

In cases of very complex rate logic, specific parameters can be configured manually during onboarding.

Step 3: Match invoices to suppliers

Before comparison, each invoice must be linked to the right supplier.

This is done through AI-based categorization.

There are two main methods:

1. Small vendor base → use a Large Language Model (LLM) to interpret supplier names, aliases, and contexts.
2. Large vendor base → use a hybrid “RAG + Embedding” system.

The embedding technique converts all supplier data into vector form, allowing semantic search (e.g. “SUEZ ENVIRONNEMENT” ≈ “SUEZ FRANCE SAS”).

Koncile then re-ranks top results using a lightweight LLM to confirm the most likely match.

Method 1 can’t be used when the list is too large, as it becomes slow and expensive to process due to the high number of lines that need to be checked.

Step 4: Line-by-line matching

Once supplier and rate grid are identified, the software proceeds to line-item reconciliation.

Case 1: Product codes available

Direct 1:1 matching based on SKU or article number.

Case 2: Product codes missing

Semantic matching on text descriptions using embedding technology.

Here, the system uses embeddings to compute similarity scores between the invoice line (“Bituminous mix 0/10”) and rate grid entries (“ENROBE A CHAUD 0/10”).

It selects the top 5 matches and submits them a LLM to select the right match.

Step 5: Compute price deviations

Once lines are matched, each difference is quantified:

\text{Delta} = \text{Quantity} × (\text{Unit Price in Grid} - \text{Unit Price on Invoice})

This yields a detailed breakdown of potential overcharges per line, per invoice, per supplier — exportable to dashboards or accounting reports.

Step 6: Generate credit note requests

The final step is to produce a credit note request document.

This report consolidates all detected discrepancies and presents them clearly for supplier communication:

SupplierInvoice #ProductQtyUnit Price (Invoice)Unit Price (Grid)DifferenceTotal Delta

This standardized output can be sent automatically by email, attached to your AP workflow, or integrated via API into your ERP.

Automating invoice control with Purchase Orders (POs)

When companies already operate with structured Purchase Orders, the same principle applies — with a slightly different workflow.

Step 1: Extract POs and invoices

Both POs and invoices are captured via OCR or through direct system integration (SAP, Oracle, etc.).

Each document is normalized to ensure comparable fields: product code, quantity, price, and total.

Step 2: Identify supplier

Same as before: supplier identification is handled through AI matching using embeddings and LLM reasoning.

Step 3: Match invoice lines to PO lines

For each invoice, the system attempts to match every line with a PO line.

1. First, by exact product code or PO reference.
2. If that fails, by semantic similarity of item descriptions.
3. If still uncertain, the software flags the line for human review.

Step 4: Flag anomalies

The tool identifies:

• Price deviations
• Quantity mismatches
• Missing or invalid PO numbers
• Invoices referencing closed or already billed POs

For small volumes, a full LLM prompt can be used to reason about the relationship between PO and invoice content.

For larger datasets, embedding + re-ranking remains the most scalable and cost-efficient approach.

Benchmarking and performance analytics

Once invoice and rate data are structured, companies can go much further than just error detection.

• Benchmark pricing across subsidiaries or suppliers.
• Identify systematic deviations (e.g., one site consistently overpaying).
• Detect process inefficiencies (e.g., PO not referencing correct rate grid).
• Build supplier performance dashboards tracking compliance and accuracy.

In short, automated verification transforms invoice control from a reactive process into a strategic intelligence tool.

Tangible benefits: up to 5% direct savings

Across industries, automated invoice verification yields measurable results:

For CFOs and procurement directors, the ROI is immediate: even a few detected overcharges can justify the investment in weeks.

The future of invoice verification

With the rise of AI and regulatory frameworks like mandatory e-invoicing in the EU, the landscape of invoice verification is evolving rapidly.

Tomorrow’s systems will combine:

• Real-time e-invoice validation (as data streams, not documents)
• Self-learning AI models that adapt to supplier formats
• Cross-company benchmarking to detect outliers globally
• Automated dispute workflows that generate, send, and track credit note requests