Split Invoice Detection: How ProcureLabs Catches $100K+ in Hidden Fraud
What Is Invoice Splitting?
Invoice splitting — also called purchase splitting or order splitting — is the deliberate practice of dividing a single purchase into multiple smaller invoices to keep each one below an approval threshold. It is one of the most common and hardest-to-detect forms of procurement fraud.
Every organization has approval hierarchies: purchases under $5,000 might need only a department manager's sign-off, while anything over $5,000 requires VP approval, and over $25,000 needs procurement review. Invoice splitting exploits these tiers by ensuring no single invoice triggers the higher level of scrutiny.
The fraud is not in the purchases themselves — the goods or services may be legitimate. The fraud is in the circumvention of controls designed to ensure proper oversight of organizational spending.
Common Patterns
Four splitting patterns that ProcureLabs detects automatically.
Just-Below-Threshold
Multiple invoices from the same vendor consistently land just below the approval threshold. For example, a $5,000 approval limit produces a cluster of invoices at $4,900, $4,850, and $4,950.
Detection Signal
Statistical clustering at 95-99% of threshold value
Example
$5,000 limit produces 12 invoices between $4,800-$4,999 in one quarter
Same-Day Splits
Multiple invoices from the same vendor submitted on the same day for related goods or services that could have been a single purchase order.
Detection Signal
Temporal clustering with vendor + date correlation
Example
Three invoices from Vendor X on March 15: $4,200 + $3,800 + $4,100 = $12,100
Sequential Splitting
Consecutive PO numbers or invoice references for related line items that were deliberately separated into individual purchase orders to stay below thresholds.
Detection Signal
Sequential reference numbers with same vendor and category
Example
PO-2026-1001 through PO-2026-1004, all for office furniture, all under $5,000
Cross-Period Splitting
Spreading a large purchase across budget periods — end of one month and beginning of the next — to avoid triggering cumulative spend alerts.
Detection Signal
Related purchases spanning period boundaries
Example
$4,500 on March 31 and $4,700 on April 1 from the same vendor for the same project
Why Traditional Controls Fail
Approval workflows are designed to evaluate individual invoices. Each invoice passes through the hierarchy independently. A $4,900 invoice gets a manager approval. Another $4,900 invoice gets a manager approval. Neither triggers VP review, even though the combined $9,800 spend clearly should.
The fundamental problem is that traditional controls are transaction-level, not pattern-level. They answer the question “Is this single invoice within policy?” but cannot answer “Is the pattern of invoices from this vendor suspicious?”
Even periodic audits struggle to catch splitting. Auditors typically sample a percentage of transactions. Split invoices, by design, look unremarkable individually. Only when aggregated by vendor, time period, and category does the pattern emerge — and that requires computational analysis, not spot checks.
AI Detection Methods
How ProcureLabs detects split invoices across your entire transaction history.
Statistical Threshold Analysis
Analyzes the distribution of invoice amounts relative to known approval thresholds. Natural purchasing produces a smooth distribution. Split invoices create an unnatural spike just below threshold values.
Temporal Clustering
Identifies invoices from the same vendor that cluster in time windows too short for independent purchasing decisions. Same-day, same-week, and same-period patterns are scored by frequency and amount.
Vendor Behavior Profiling
Builds a baseline of normal invoice patterns per vendor. Deviations — sudden increases in invoice frequency, shifts in typical amounts, or new patterns of just-below-threshold billing — trigger investigation alerts.
Benford's Law Analysis
Applies Benford's Law to the leading digits of invoice amounts. Legitimate invoices follow a predictable distribution (30.1% start with 1, 17.6% with 2, etc.). Fabricated or manipulated amounts deviate significantly from this distribution.
Average value of split invoice schemes before detection.
Detect Split Invoices Automatically
Start your free trial and scan your invoice data for splitting patterns.