AI-powered Anomaly Detection
Use AI/ML to detect abnormal events without rules for transaction monitoring and fraud prevention.
Sumsub Anomaly Detection is designed to help you monitor transactions by detecting anomalies with remarkable accuracy. It utilizes AI and advanced machine learning algorithms like Isolation Forest to identify activities or transactions that deviate significantly from established applicant behavior patterns, indicating potential fraud.
By continuously learning from transaction patterns, the system adapts to new types of anomalies, ensuring your financial operations remain secure.
This means you can trust it to timely spot irregularities and do the following:
- Identify potential fraud activities as they happen, allowing immediate action.
- Prevent significant financial losses by catching fraud early.
- Enhance customer trust by ensuring their transactions are secure.
- Provide tailored protection regardless of business size or industry.
- Let your business meet regulatory requirements related to fraud prevention.
You can use anomaly detection as an independent functionality for depending on suspicious events and transactions or create rules based on the anomaly score.
How it works
Sumsub Anomaly Detection identifies unusual financial activities that may require immediate investigation and sends you real-time alerts or creates a case/rejects/puts the transaction on hold/adds tags, and so on.
Our AI also provides insights into why a particular transaction was flagged. It examines various factors and highlights the specific reasons for its decision, clearly understanding potential issues.
Each event or transaction gets an anomaly score: Low, Medium, or High, depending on the severity of the behaviour deviation.
The severity of the deviation is defined based on the following types of signals: Aggregate, Applicant, Finance, Counterparty, Device fingerprinting, and Other.
Get started
Anomaly Detection is a free feature available to all Sumsub clients. It is intended to analyze financial transactions; each transaction you create will get the anomaly score described above.
You can also use the Anomaly Risk Level as a condition for your rules that put on hold or block the transaction, depending on your choice. For example, you can select to auto-reject all transactions with a high anomaly score:
- Create a rule.
- Set up a rule action as Reject.
- Set up the condition as follows:
If
->Field
->anomalyExpressionData.anomalyRiskLevel
->equals
->Value
->high
. - Proceed with editing and saving your rule.
Review results
The information about the Anomaly Score and the severity of the involved signals is available in the Anomaly Detector section of the target Transaction page.
The score is calculated based on the following signals:
Type of signal | Description |
---|---|
Aggregate | Signals based on calculating and comparing the applicant's transactional patterns and history. This includes calculating a specific counterparty's monthly/quarterly turnover, comparing applicants on the account, etc, to detect anomalies. |
Applicant | Signals based on analyzing customer-provided details with data from databases, documents, and third-party sources to detect anomalies. |
Finance | Signals based on crosschecking transaction data such as transaction method, institution, country, amount, etc, to detect anomalies. |
Counterparty | Signals based on analyzing counterparty data with data from databases, documents, and third-party sources to detect anomalies |
Device fingerprinting | Signals based on device data, such as IP addresses, device fingerprints, statistics, etc., to detect anomalies. |
Other | Signals based on all remaining risk factors not included in the other factors. |
Updated 5 days ago