Fraud Detection · Risk Management
A leading US retailer wanted to identify risks such as counterfeit products and litigation before costly incidents occurred. Existing manual reviews and fragmented public data could not reliably surface risks across thousands of marketplace listings and retail locations.
Reported impact
The retailer can proactively score more than a dozen business risks, identify suspicious products and potential litigation hotspots, and provide risk evidence to reviewers. KPMG completed the first implementation in four weeks.
KPMG built a predictive risk-monitoring system on Google Cloud that uses NLP and multiple machine-learning models to analyze public data, including product listings, reviews, pricing, and community information. KPMG Signals Repository harvests data from more than 55,000 sources and transforms structured and unstructured information into signals; ingestion functions preprocess and cleanse it before it is staged in Cloud Storage and stored in BigQuery. Models trained with Google Compute Engine, Cloud Natural Language, AutoML, and BERT analyze separate risk indicators, including review language and reviewer behavior. The system assigns risk scores, highlights suspicious product descriptions and prices, and provides supporting evidence through personalized Looker dashboards. The workflow also connects with the retailer's case-management system so teams can investigate flagged risks.
Public product, review, pricing, and community data is harvested and preprocessed by KPMG Signals Repository and Cloud Functions, staged in Cloud Storage, and analyzed from BigQuery with NLP and ML models; scored risks and evidence appear in Looker and connect to case-management workflows.
"For each partner product it analyzes, the system provides separate scores for the product’s description, its reviews and its pricing."
"We trained AI models to identify the content that might indicate a higher degree of risk."
Review how this reference was scored and trace each normalized claim back to its documented source.
Scoring methodology
Customer Reference Value measures usefulness as a customer reference. Evidence Confidence measures the quality and traceability of available evidence, not independent verification.
Traceability
Field coverage shows which facts each source supports. Dates are shown only when documented by the source.
Related
Databricks Sample Oct 14, 2024 Simpl Previously hampered by an outdated tech stack and siloed data, which caused frequent failures and lengthy ML model training, Simpl faced substantial scalability and operational challenges. 35% improvement in the analytics team’s turnaround time for reporting tasks Databricks SQL · Delta Lake · Lakeflow Jobs 
In-house Sample Oct 27, 2026 Cybera Scam victims often report fraud only after their money has been transferred and become difficult to trace. Banks need to identify money-mule accounts before they receive or move illicit funds, including before a payment… Cybera runs hundreds of these personas across thousands of conversations every day. Money Mule Intelligence · Recorded Future In-house Sample Oct 27, 2026 Cybera Scam proceeds can be moved through money-mule accounts before victims report fraud, making the funds difficult to trace. Banks also need reliable evidence to identify suspect accounts and intervene without generating a… €60,000 Money Mule Intelligence · Recorded Future 
