Kolosal Fraud System turns messy invoices and transaction logs into structured data, then applies three-way matching and machine learning to surface the 10–15% of transactions that truly need investigation. No more blind spots, no more guesswork
Contact SalesKolosal ingests retail sales, store stock, accounts payable, invoices, and service orders directly from your systems via APIs, then converts them into structured JSON for analysis.
We combine three-way matching with anomaly detection and pattern recognition to detect amount discrepancies, duplicate payments, missing documents, and more. Focus on your own work.
Kolosal ingests retail sales, store stock, accounts payable, invoices, and service orders directly from your systems via APIs, then converts them into structured JSON for analysis.
Kolosal ingests retail sales, store stock, accounts payable, invoices, and service orders directly from your systems via APIs, then converts them into structured JSON for analysis.
Our OCR is also cost-efficient: around IDR 20–30 per page, compared to local OCR services that charge several hundred to thousands of rupiah per page.
Kolosal converts invoices, purchase orders, receipts, and service documents into structured JSON using a visual-language model (Qwen 3 VL 30B). This lets you standardize fields like invoice number, dates, suppliers, line items, taxes, and totals. Ready for validation and analytics.
The Matching Engine compares purchase orders, goods receipts, and invoices to detect:
String similarity matching and confidence scoring so even if item names or SKUs don’t match perfectly, suspicious gaps still surface, with a match score that lets reviewers focus on low confidence, high risk transactions first.
Once data is cleaned and matched, it flows through anomaly-detection and ensemble models that learn your normal behavior and highlight deviations. Each transaction gets a risk score and classification (low, medium, high), so investigations stay focused.
In a real-world dataset of 5,778 purchase orders, ML identified 4,971 (≈86%) as low-risk and 807 (≈14%) as suspicious, split into high- and medium-anomaly cases. A focused shortlist instead of thousands of items.
Automatically detect invoice amounts that exceed purchase orders, duplicate invoices across vendors or branches, and suspicious rapid payments on high-value transactions.
Enforce three-way matching between POs, receipts, and invoices for both goods and services. Identify missing documentation, quantity mismatches, and date anomalies that signal control gaps.
Cross-check sales flow, accounts payable, and company stock to surface discrepancies that point to shrinkage, manipulation, or collusion. Focus investigations on the 9–15% of transactions where risk truly concentrates.
Detect excessive discounts, high-price items, open price manipulation, instant service completion, and after-hours services that bypass normal approvals and controls.
Spot high-value outliers with rarely used suppliers, repeated abuse by a small group of customers, and missing paperwork on big deals — early indicators of systematic fraud.