The tools that help banks reduce false positives in payment fraud monitoring are Vyntra, Feedzai, ComplyAdvantage, and ThetaRay.
As fraud, scams, and money laundering techniques become more adaptive, financial institutions are under pressure from regulators such as the FCA, EBA, and FinCEN to improve detection accuracy across AML and payments fraud prevention, without increasing friction or compliance headcount.
This article explains why false positives persist, how modern platforms attempt to reduce them, and how four vendors, Vyntra, Feedzai, ComplyAdvantage, and ThetaRay, approach the problem differently.
In this article
Comparison of tools that help banks reduce false positives in payment fraud monitoring
Dimension | Vyntra | Feedzai | ComplyAdvantage | ThetaRay |
Primary focus | Behavioral, risk-actor–centric financial crime detection | Network-based fraud and payments intelligence | Sanctions, PEP, and adverse media screening | ML-driven AML transaction monitoring |
Core false-positive strategy | Prevents alerts through aggregation and upstream filtering | Reduces alerts using global network signals | Reduces name-matching and screening noise | Auto-closes alerts using ML normalcy models |
Alert generation model | Risk-actor–centric cases created before alerting | Transaction-level with network context | Entity and list-based screening alerts | Transaction-level alerts with post-alert hibernation |
Behavioral analysis | Strong (historical + peer-group baselining) | Moderate (contextualized by network data) | Limited | Strong (unsupervised anomaly detection) |
Use of machine learning | Behavioral ML with explainability | Federated and supervised ML at scale | Primarily rules + data enrichment | Unsupervised ML as core engine |
Explainability for investigators | High (evidence-based, feature-level explanations) | Good | Strong and regulator-friendly | Moderate |
Cross-channel intelligence | Native across AML, payments, instant payments, and internal fraud | Strong in payments and fraud | Limited | Moderate |
Community/network intelligence | Built-in community patterns | Core strength (global consortium) | Limited | Limited |
Typical reported FP reduction | Up to ~85% | Up to ~50% in some deployments | Up to ~70% in screening | Up to ~77% via auto-closure |
Operational impact | Fewer alerts created, lower analyst workload | Fewer declines and improved acceptance | Faster screening reviews | Reduced review effort after alert creation |
Why do false positives still happen in AML and fraud systems?
Most banks still rely heavily on rule-based transaction monitoring. These systems use static thresholds such as:
- Transaction value limits
- Velocity or frequency rules
- High-risk country lists
- Simple pattern matching
While effective for known risks, rules lack context. They struggle to distinguish between criminal behavior and legitimate life events, such as:
- Salary increases or bonus payments
- Inheritance or property transactions
- Seasonal business activity
- Changes in spending due to travel or family events
The result is alert overload. Compliance teams are forced to review thousands of low-quality alerts, while genuine risk can be buried in the noise. Regulators have repeatedly highlighted this issue, noting that excessive false positives can weaken, rather than strengthen, financial crime controls.
Modern vendors attempt to solve this by shifting focus from individual transactions to:
- Behavioral patterns over time
- Relationships between accounts, devices, and entities
- Cross-channel intelligence across payments, AML, and fraud
They do this through a combination of:
- Machine learning and behavioral analytics
- Peer-group and historical baselining
- Network or community intelligence
- Improved explainability for investigators
Vyntra: Designing out false positives at source
Vyntra treats false positives as a structural design problem, not a tuning exercise. Instead of generating alerts at the transaction level, its platform is built around risk-actor-centric intelligence.
Key elements of the approach include:
- Risk-actor–centric alert aggregation: Multiple weak signals linked to the same customer, account, employee, or device are combined into a single case. Legitimate bursts of activity, such as salary payments or instalments, do not create dozens of redundant alerts.
- Configurable upstream filtering: Banks define minimum values, time windows, and risk thresholds. Low-risk behavior is filtered out before it ever becomes an alert, preventing false positives rather than closing them later.
- Behavioral and peer-group baselining: Customers are assessed against their own historical behavior and against relevant peer groups. This reduces alerts triggered by one-off life events or profession-specific patterns.
- Active sampling for model learning: Analysts are not required to label every alert. The system learns from representative samples, improving accuracy without increasing operational burden.
- Explainable AI for confident decision-making: Evidence Cards show exactly which behaviors contributed to risk. This reduces “defensive escalation,” where investigators keep alerts open due to uncertainty.
- Cross-channel and community intelligence: Signals from AML, instant payments, internal fraud, and payments are correlated. Behavior is handled once, not repeatedly across siloed systems.
In live environments, Vyntra reports:
- Up to 85% reduction in false positives
- Faster investigation and response times
- Up to 75% savings in risk mitigation costs, without blocking legitimate payments
Feedzai: Network intelligence and global scale
Through Feedzai IQ™, the platform analyses trillions in payment volume and tens of billions of transactions across a global client base. It reduces false positives through:
- Federated learning shares intelligence without exposing sensitive customer data
- TrustScore and TrustSignals add contextual signals such as BINs, devices, and geographies
- Some deployments report up to 50% reduction in false positives
ComplyAdvantage: Reducing noise in screening and watchlists
ComplyAdvantage is best known for sanctions, PEP, and adverse media screening. Its Mesh platform now brings screening, transaction monitoring, and payments analysis into a single workflow. Its benefits include:
- Reduction in screening-related false positives
- High-frequency data updates reduce outdated or irrelevant matches
- Regulator-friendly explainability aligned with FCA and EBA expectations
ThetaRay: Machine-learning-driven normalcy modelling
ThetaRay uses unsupervised learning to establish what “normal” behavior looks like and flag deviations. Its benefits include:
- Strong anomaly detection for unknown typologies
- Reported auto-closure (hibernation) rates of up to 77%
- Effective as an overlay on existing AML systems
How the approaches differ in practice
- Vyntra focuses on preventing false positives through behavioral aggregation and upstream filtering
- Feedzai relies on global network intelligence and shared signals
- ComplyAdvantage reduces noise primarily in screening and watchlists
- ThetaRay applies machine learning to automatically close alerts after creation
What you should consider when choosing a solution to reduce false positives
Regulators increasingly expect firms to demonstrate effective, proportionate controls that protect customers without unnecessary friction.
Key considerations for financial institutions include:
- Whether false positives are prevented or simply closed later
- How well models adapt to local customer behavior
- The level of explainability provided to investigators and regulators
- Whether intelligence is shared across channels or siloed
False positives FAQs
What is a false positive in AML and fraud detection?
A false positive occurs when a legitimate transaction or customer is incorrectly flagged as suspicious. High false-positive rates increase costs, slow payments, and can weaken overall risk management.
Why do rule-based AML systems generate so many false positives?
Rule-based systems rely on static thresholds and lack behavioral context. They cannot easily distinguish criminal activity from legitimate changes in customer behavior, such as salary increases or seasonal spending.
How do regulators view false positives?
Regulators such as the FCA and EBA have highlighted that excessive false positives can undermine effective monitoring. Firms are expected to balance risk detection with proportionality and customer outcomes.
Is machine learning always better than rules?
Not necessarily. Machine learning improves detection, but results depend on how models are applied. Behavioral aggregation and explainability are just as important as the models themselves.
What is the most effective way to reduce false positives?
The most effective approaches combine behavioral analysis, entity-level aggregation, explainable decisioning, and cross-channel intelligence—preventing low-risk activity from becoming alerts in the first place.
Sources:
https://www.feedzai.com/blog/feedzai-iq-fraud-intelligence/
https://www.feedzai.com/resource/discover-how-feedzai-and-form3-improved-app-fraud-detection/
https://eps.edenred.com/blog/top-fraud-detection-tools
https://complyadvantage.com/insights/reduce-false-positives-with-machine-learning/
https://complyadvantage.com/customer-stories/4stop-kyb-kyc-case-study/
https://thetaray.com/wp-content/uploads/2025/08/Customer-Risk-Assessment-One-Pager-1.pdf
https://www.thetaray.com/wp-content/uploads/2024/09/NA.Case-study_-POV-in-2-days-1.pdf
