Fraud Score

A fraud score is a numeric value assigned to a transaction, or sometimes an account, representing how likely it is to be fraudulent based on a combination of behavioural, transactional and device-level signals. Rather than a simple approve-or-decline decision, a fraud score gives a graded sense of risk that systems can act on with more nuance.

What Actually Feeds Into the Number

A fraud score typically draws on dozens of inputs at once, including transaction amount relative to a customer's typical spend, device and location consistency, time of day, and how the current transaction compares to known fraud patterns. No single input drives the score, and the EBA and ECB's joint report on payment fraud shows exactly why this layered approach beats reliance on any one indicator.

Why a Score Beats a Binary Decision

Treating every transaction as either clearly fine or clearly fraudulent misses the large middle ground where most genuinely ambiguous cases actually sit. A graded score lets a business route low-risk transactions straight through, block clearly high-risk ones, and send the uncertain middle to manual review rather than guessing in either direction.

How Scores Connect to Detection Systems

Fraud scores are usually generated as part of a broader fraud detection pipeline, feeding into automated decisioning rules that determine whether a transaction is approved, declined or escalated. The score itself is rarely the final word; it's an input into a larger decision process alongside merchant-specific rules and thresholds.

Related but Distinct From Risk Score

Fraud score and risk score are related concepts but not identical. A fraud score typically focuses narrowly on the likelihood of fraudulent intent in a specific transaction, while a broader risk score might also factor in things like a merchant's industry category or overall business risk profile.

Watching Scores Over Time, Not Just Per Transaction

Individual fraud scores matter, but tracking how average scores shift over time can reveal broader patterns, such as a sudden increase in flagged transactions from a particular region or device type. Real-Time Payment Analytics & Reporting makes that kind of trend visible far sooner than reviewing individual transactions ever could.

Setting Thresholds Is Where the Real Work Happens

The fraud score itself is only half the picture; deciding what threshold triggers a decline, a manual review, or a velocity check is where a business's specific risk tolerance actually comes into play. Two merchants using the same underlying scoring model can end up with very different outcomes depending on where they set those thresholds.

Why the Same Transaction Can Score Differently Elsewhere

Because fraud scoring models are typically trained on a provider's own historical transaction data, the same purchase can receive a noticeably different score depending on which provider evaluates it. This isn't a flaw so much as a reflection of different data sets and different fraud patterns each provider has actually seen across its own merchant base.

Combining Fraud Scores With Business Context

A fraud score works best when it's read alongside business-specific context rather than treated as an absolute verdict. A high score on a large first-time order from a new customer might warrant review, while the same score on a modest repeat purchase from a long-standing customer might reasonably be approved automatically, since the surrounding context changes how much weight the score should carry.

Explaining Scores to Customers Who Get Declined

A declined transaction based on a fraud score can understandably frustrate a genuine customer who has no idea why their payment failed. Some merchants build lightweight fallback options, like asking for an alternative payment method or offering manual verification, specifically to recover legitimate sales that a fraud score alone might otherwise have blocked outright.