Fraud scores help teams assess transaction fraud risk in real time by translating complex behavioral and payment signals into a single, actionable measure. Assessing transaction risk using fraud scores is a growing priority for businesses, given that they lost, on average, 7.7% of annual revenue to fraud from 2024–2025. Understanding how fraud scores work, and what they can and can’t tell you, is necessary for minimizing payment fraud, avoiding unnecessary declines, and making better risk decisions at scale.

Below, we’ll explain how a fraud score is created, how businesses use them, and how to interpret them with clarity and confidence.

What’s in this article?

• What is a fraud score?
  
• How do fraud scores help businesses assess transaction risk?
  
• How do businesses use fraud scores to approve or decline transactions?
  
• What data is used to calculate a fraud score?
  
• How accurate are fraud scores?
  
• Why can the same transaction have different fraud scores?
  
• Are fraud scores proof of fraud or indicators of risk?
  
• How Stripe Radar can help
  

What is a fraud score?

A fraud score is a way to express risk. It’s a signal, often a number or category, that estimates how likely a specific transaction or action is to be fraudulent. Think of it as a summary judgment: based on everything the system can see right now, how risky does this look? Instead of forcing teams to evaluate dozens of data points individually, a fraud score distills them into a single, actionable indicator.

How do fraud scores help businesses assess transaction risk?

Fraud scores are generated in real time, often within milliseconds of a transaction. This is possible because the evaluation process is structured and automated.

Here’s how a fraud score works:

• Signal collection: As a transaction happens, the system gathers relevant data about the payment, customer, device, and context.

• Pattern comparison: Each signal is compared against patterns learned from historical legitimate and fraudulent behavior.

• Risk weighting: Not all signals matter equally. Clear warning signs, such as a location mismatch, carry more weight than minor irregularities.

• Model evaluation: A scoring model, often powered by machine learning, processes the weighted signals and estimates the likelihood of fraudulent activity by comparing the transaction to known outcomes.

• Score generation: The result is a single score on a defined scale that represents relative risk.

• Immediate availability: The score is returned instantly so businesses can approve, block, or review transactions without slowing the customer experience.

• Continuous learning: As the model logs outcomes such as confirmed fraud, successful transactions, and disputes, it will adapt. The ongoing feedback helps keep scores relevant as fraud tactics change.

How do businesses use fraud scores to approve or decline transactions?

Fraud scores don’t make decisions on their own. Businesses define how to act on them based on their risk tolerance and goals. These are some common approaches:

• Risk thresholds: Low-risk scores pass automatically, which keeps checkout fast for most customers.

• Automatic declines: Scores above a defined risk threshold are blocked to prevent likely fraud from becoming chargebacks or downstream losses.

• Manual review queues: Midrange scores are routed to human reviewers when the risk is unclear but worth a closer look.

• Step-up verification: Some scores prompt additional verification rather than an outright decline.

• Policy tuning: Teams regularly adjust thresholds based on outcomes such as fraud rates, false declines, and customer impact.

• Business focus: By concentrating attention on the riskiest activity, fraud scores help teams scale efficiently without reviewing everything.

What data is used to calculate a fraud score?

Fraud scores are built from many small signals that, on their own, might seem ordinary. But when the signals are combined, they’re a powerful metric.

Here’s how payment data is used to calculate a fraud score:

• Transaction details: Amount, currency, item type, timing, and frequency help establish whether the activity fits normal patterns.

• Payment information: Card metadata, issuing country, and billing details are checked for consistency.

• Customer history: Established customers with successful transaction histories generally score lower than new or inactive accounts.

• Account and identity signals: Email address quality, account age, and completeness of profile information help ascertain credibility. Disposable email addresses or recently created accounts often pose a risk.

• Device data: Device identifiers indicate whether a device has been seen before and how it has behaved in the past.

• Network and location signals: Internet Protocol (IP) data helps businesses infer geographic location and network characteristics. Large discrepancies or anonymizing services tend to elevate risk.

• Behavioral patterns: Unusual checkout behavior or repeated credential attempts can influence scoring.

• Speed indicators: Multiple actions such as login and payment attempts in a short time period might signal automation or testing behavior.

How accurate are fraud scores?

Fraud scores are predictive tools, not guarantees. The accuracy of fraud scores depends on how they’re built, how they’re used, and how well they’re tuned to a business’s risk profile.

Keep the following in mind:

• Probability, not certainty: A score reflects likelihood. Reducing fraud often means accepting some false declines, while minimizing friction usually means tolerating more risk.

• False positives and false negatives: No model is perfect. Businesses have to constantly balance blocking fraud with accepting as many legitimate customers as possible.

• Data quality: Rich, current, and accurate data improves performance, while incomplete or outdated inputs decrease reliability.

• Learning from outcomes: Models improve when real-world results such as confirmed fraud, successful transactions, and disputes are used to train the system.

• Business context: Customer behavior, payment methods, and risk tolerance vary widely, which affects how predictive a score is in practice.

• Ongoing recalibration: Regular monitoring and adjustment are necessary to keep scores aligned with developing fraud patterns.

Why can the same transaction have different fraud scores?

It’s normal for the same transaction to receive different fraud scores across systems. That’s because fraud scoring isn’t standardized and depends on who’s evaluating it.

Here’s why the same transaction might have various fraud scores:

• Different models: Some fraud detection systems rely on rules, while others rely on machine learning models trained on different datasets.

• Varying data visibility: One provider might see activity across several companies, while another might see activity only within a single business.

• Different signal weighting: Models prioritize signals differently. Location, device history, or transaction speed might matter more in one system than in another.

• Diverse risk appetites: Some businesses score aggressively to minimize losses, while others score more conservatively to protect conversion.

• Different scales: A “high” score in one system might represent moderate risk in another.

• Varied timing: Scores generated later might incorporate signals that weren’t available at checkout.

Are fraud scores proof of fraud or indicators of risk?

Fraud scores are tools for decision-making. A high score signals elevated risk, not a confirmation of fraud. Legitimate customers can still receive high scores when their behavior appears unusual, and even transactions that appear safe can later turn out to be fraudulent.

Scores are most effective when they’re combined with business rules, human review, or additional checks. Their value lies in helping businesses act early, before fraud becomes a loss or dispute. While individual decisions won’t always be perfect, fraud scores improve overall outcomes across a large volume of activity. For example, Stripe Radar reduces fraud for a company by 38% on average.

How Stripe Radar can help

Stripe Radar uses AI models to detect and prevent fraud, trained on data from Stripe’s global network. It continuously updates these models based on the latest fraud trends, protecting your business as fraud evolves.

Stripe also offers Radar for Fraud Teams, which allows users to add custom rules addressing fraud scenarios specific to their businesses and access advanced fraud insight.

Radar can help your business:

• Prevent fraud losses: Stripe processes over $1 trillion in payments annually. This scale uniquely enables Radar to accurately detect and prevent fraud, saving you money.

• Increase revenue: Radar’s AI models are trained on actual dispute data, customer information, browsing data, and more. This enables Radar to identify risky transactions and reduce false positives, boosting your revenue.

• Save time: Radar is built into Stripe and requires zero lines of code to set up. You can also monitor your fraud performance, write rules, and more in a single platform, increasing efficiency.

Learn more about Stripe Radar, or get started today.