Fraud Scoring for High-Risk Merchants: Tools, Thresholds & Best Practices

The Gap Between Fraud Prevention and Conversion Is Where Merchants Lose

Every high-risk merchant faces the same impossible-feeling tension: block too little fraud and chargebacks destroy your merchant account. Block too much and you’re declining legitimate customers, killing revenue, and handing business to competitors.

Fraud scoring is the mechanism that sits between those two failure modes. When implemented correctly, it lets you make granular, data-driven decisions about every transaction, not blunt blocks based on geography or card type.

In 2026, fraud scoring has become more sophisticated, more accessible, and more necessary than ever. Card-not-present fraud losses globally are expected to surpass $35 billion, and high-risk merchants in sectors like digital goods, subscription SaaS, gaming, nutraceuticals, forex, and travel bear a disproportionate share of that exposure.

This guide breaks down how fraud scoring works, which tools lead the market, how to set thresholds that balance protection with conversion, and the best practices that separate merchants with 0.3% fraud rates from those sitting dangerously close to card network termination thresholds.

What Is Fraud Scoring?

Fraud scoring is a risk assessment process that assigns a numerical score to each transaction based on dozens, sometimes hundreds, of data signals. The score reflects the estimated probability that a transaction is fraudulent.

Scores are typically expressed on a scale of 0 to 100, where:

  • 0–30 = Low risk → auto-approve
  • 31–70 = Medium risk → additional review or authentication step (e.g., 3DS2 challenge)
  • 71–100 = High risk → auto-decline or manual review queue

The specific thresholds vary by merchant, vertical, and average order value, but the principle is consistent: replace binary block/allow decisions with a graduated risk framework that routes transactions intelligently.

How Fraud Scores Are Calculated

Modern fraud scoring models combine multiple layers of signals, including:

  • Device intelligence: device fingerprint, OS, browser, screen resolution, and whether the device has been associated with previous fraud
  • Behavioral analytics: how the user navigated the site, typing cadence, mouse movement patterns, time spent on each page
  • Network and IP data: IP geolocation, VPN/proxy/Tor detection, IP velocity (how many orders from the same IP in a given period)
  • Card and transaction data: BIN country vs. shipping country mismatch, AVS match result, CVV match, card type, and previous transaction history on that card
  • Identity signals: email address age and reputation, phone number validity, name/address consistency
  • Order patterns: basket size vs. historical average, number of items, whether the shipping address differs from billing address, and express shipping selection (a classic fraud signal)
  • Velocity checks: how many transactions from the same card, device, email, or IP have occurred within a defined time window

Each signal is weighted within the scoring model. Machine learning models continuously recalibrate these weights based on confirmed fraud outcomes, meaning a well-tuned fraud scoring system improves over time.

Why Fraud Scoring Is Non-Negotiable for High-Risk Merchants

A standard low-risk retailer might accept a simple AVS + CVV check as sufficient fraud control. For high-risk merchants, this approach is reckless.

Here’s why the stakes are categorically different:

Higher average order values mean each fraudulent transaction carries more financial damage. A $400 chargeback in digital goods is fundamentally different from a $15 physical retail dispute.

Digital goods and instant-access products can’t be recalled after delivery. Once a fraudster downloads software, accesses a subscription, or receives a gift card code, the loss is unrecoverable.

Cross-border transaction exposure is far greater in high-risk verticals. Merchants in LATAM, the UK, and Canada selling internationally face elevated fraud rates from specific regions, without proper scoring, broad geo-blocks become the only alternative, and broad geo-blocks cost legitimate revenue.

Chargeback ratio sensitivity means even a modest uptick in fraud can push a high-risk merchant account into Visa’s Dispute Monitoring Program (VDMP) or Mastercard’s Excessive Chargeback Merchant (ECM) program. At 0.9% chargeback ratio, you’re in monitoring. At 1.8%, you’re facing account termination.

Fraud scoring compresses the number of fraudulent transactions that become chargebacks — which directly protects your ratio and your processing relationship.

Leading Fraud Scoring Tools for High-Risk Merchants in 2026

The fraud prevention technology market has matured significantly. Here are the categories of tools high-risk merchants should evaluate:

Enterprise-Grade Fraud Platforms

Kount (an Equifax company) remains one of the most widely deployed fraud scoring platforms for high-risk eCommerce and fintech. Its AI-driven scoring engine pulls from a global network of transaction data across thousands of merchants, enabling consortium-based fraud signal sharing. Particularly strong for merchants processing high volumes in North America and the UK.

Signifyd specializes in eCommerce fraud protection with a guaranteed chargeback model, meaning they absorb chargeback losses on transactions they approve. This makes their threshold decisions skin-in-the-game, which aligns incentives well. Strong for mid-to-large eCommerce operators.

Riskified uses machine learning models trained on behavioral and device data, with a similar chargeback guarantee model. Known for strong performance in fashion, electronics, and high-ticket digital goods verticals.

Payment-Native Fraud Tools

Stripe Radar is deeply integrated into the Stripe payment stack and uses machine learning trained on Stripe’s global transaction volume. Well-suited for SaaS and subscription businesses already on Stripe. Offers customizable rules and score thresholds with no additional integration overhead.

Adyen RevenueProtect is the equivalent for merchants on the Adyen platform, a rule and model-based fraud scoring engine with global reach. Particularly effective for merchants with high LATAM transaction volumes given Adyen’s strong regional acquiring relationships.

Braintree’s fraud tools (built on Kount’s data network) provide solid scoring for PayPal/Braintree merchants, though they offer less customization than standalone platforms.

Specialist and Emerging Tools

SEON has gained significant traction in 2025–2026 for its real-time digital footprint analysis, enriching transactions with email reputation, social media presence, and phone number validity data. Particularly effective for fintech onboarding and account creation fraud, not just payment fraud.

Sardine specializes in behavioral biometrics and device intelligence for fintech and crypto merchants, with strong coverage for LATAM market fraud patterns.

NS8 and NoFraud serve mid-market eCommerce merchants with accessible pricing and solid integration support for platforms like Shopify, WooCommerce, and Magento.

Setting Fraud Score Thresholds: The Balancing Act

One of the most consequential decisions in your fraud scoring setup is where to draw the threshold lines. Set them too aggressively and you decline legitimate customers. Set them too loosely and fraud bleeds through.

A Recommended Starting Framework

Score Range Risk Level Recommended Action
0–25 Very Low Auto-approve, no friction
26–50 Low-Medium Approve with passive monitoring
51–70 Medium-High Trigger 3DS2 authentication challenge
71–85 High Route to manual review queue
86–100 Very High Auto-decline

 

This is a starting point, not a universal standard. Your thresholds should be calibrated based on:

  • Your vertical’s fraud base rate: digital goods merchants typically see higher fraud rates than physical goods and should set tighter thresholds
  • Your average order value: higher AOV justifies more friction at lower score ranges because the loss per fraudulent transaction is greater
  • Your false positive tolerance: declining a legitimate $2,000 order has real revenue consequences; your threshold should reflect what that miss costs vs. what the fraud costs
  • Geographic patterns: if a meaningful share of your fraud is originating from specific IP ranges or BIN countries, tighten thresholds for those segments specifically rather than globally

Threshold Review Cadence

Fraud patterns shift constantly. A threshold calibration that was optimal in Q1 2026 may be significantly under- or over-blocking by Q3. Build a monthly threshold review into your operations, pulling confirmed fraud data, false positive rates, and chargeback patterns to recalibrate.

Best Practices for Fraud Scoring in 2026

Layer Your Signals – Don’t Rely on Any Single Indicator

No single data point is a reliable fraud indicator in isolation. An IP from Brazil is not fraud. A prepaid card is not fraud. A mismatched billing and shipping address is not fraud. But an order with all three signals, combined with a newly registered email address, a device seen in previous fraud cases, and express shipping on a high-value digital goods purchase, that pattern demands action.

Build your scoring rules around signal combinations, not individual flags.

Integrate 3DS2 as a Dynamic Step-Up Authentication Layer

3D Secure 2.0 should not be applied to every transaction, that adds friction that reduces conversion. Use your fraud score to trigger 3DS2 selectively for medium-risk transactions (score range 51–70). This way, the authentication step is reserved for transactions that actually warrant it, and liability shifts to the card issuer for those that pass the challenge.

For UK and EU merchants, selective 3DS2 triggering also aligns with PSD2 Strong Customer Authentication (SCA) exemption logic, allowing low-risk transactions to benefit from transaction risk analysis (TRA) exemptions without challenging every customer.

Build a Manual Review Queue – and Staff It

Auto-approve and auto-decline handles the clear ends of the risk spectrum. The middle band, typically 15–25% of flagged transactions, benefits from human review. Build a structured review queue with defined criteria for approval and decline, and document your decisions. Over time, this data feeds back into model improvement.

For high-risk merchant accounts with lean teams, outsourcing manual review to your fraud platform vendor’s review service is a viable option.

Monitor False Positive Rates as Closely as Fraud Rates

False positives, legitimate customers incorrectly declined, are silent revenue losses. Unlike chargebacks, they don’t appear on any report your processor sends you. Build explicit tracking: sample declined transactions, attempt to classify them as fraud or legitimate, and calculate your false positive rate monthly. Industry benchmarks suggest a false positive rate below 2–3% is achievable with well-tuned scoring.

Share Data Across Your Stack

Fraud scoring works best when it has access to as many signals as possible. Integrate your fraud scoring platform with your CRM (to identify returning customers who can bypass friction), your customer support system (to flag accounts with dispute history), and your chargeback alert service (to feed confirmed fraud signals back into the scoring model in near real-time).

The 2026 Fraud Scoring Imperative for High-Risk Merchants

The technology available for fraud scoring in 2026 is more capable, more accessible, and more important than at any prior point. Machine learning models trained on global transaction data, real-time behavioral biometrics, and consortium fraud networks give even mid-market high-risk merchants tools that were previously reserved for enterprise-level payment operations.

But the technology alone isn’t enough. The merchants who genuinely protect their high-risk merchant accounts are those who treat fraud scoring as an ongoing operational discipline, calibrating thresholds, reviewing false positives, feeding confirmed outcomes back into their models, and staying ahead of evolving fraud patterns in their target markets.

Your fraud score is only as good as your last calibration.

Looking for verified fraud prevention tools and high-risk payment processing solutions? Explore our curated listings at TheFinrate — the fintech industry’s trusted comparison and discovery platform.