Single transaction scoring
Enter indicators, tune settings, then calculate a consistent score.
Batch scoring (CSV)
Upload up to 200 rows and receive scored output with actions.
Example data table
Use these rows to validate your workflow and thresholds.
| Scenario | Amount | Typical | Tx 24h | Acct age | Geo mismatch | IP risk | Category | Expected action |
|---|---|---|---|---|---|---|---|---|
| Returning customer, normal | 75 | 85 | 1 | 540 | No | 10 | Low | Approve |
| New account, fast activity | 260 | 90 | 7 | 12 | Yes | 55 | Medium | Review / Step-up |
| High-risk signals stacked | 1200 | 110 | 15 | 3 | Yes | 92 | High | Block |
Expected action depends on your tuned thresholds and weights.
Formula used
1) Normalize each indicator to a 0–1 risk scale.
- Ratio signals (amount, velocity): normalize by comparing to typical values, cap extreme ratios, then map to 0–1.
- Bounded signals (IP risk, device reputation): convert to 0–1 using percentages.
- Binary signals (geo mismatch, time anomaly): map No→0, Yes→1.
- Age signal: younger accounts score higher risk, scaled over 365 days.
2) Compute a weighted average risk.
Let ni be each normalized factor and wi its weight. The base risk is: R = (Σ wi · ni) / (Σ wi).
3) Apply protective offsets for verified controls.
Verified email, phone, payment method, and two-factor reduce risk by fixed offsets. The adjusted risk is R′ = clamp(R − P, 0, 1).
4) Map adjusted risk to a 0–100 score.
A logistic function improves separation near decision boundaries: Score = 100 / (1 + e−k(R′ − m)), where k is steepness and m is the midpoint.
How to use this calculator
- Enter current transaction amount and a realistic typical baseline.
- Fill behavior signals: velocity, failed logins, geo mismatch, and timing.
- Add account history: account age and chargeback rate.
- Select verifications enabled to reflect actual control strength.
- Open advanced settings to tune weights and thresholds for your policy.
- Press Submit to see the score above the form.
- Download CSV or PDF for audit trails and reporting.
- For scale, upload a CSV and download scored results.
Risk signals and normalization
Fraud exposure rises when transaction behavior deviates from an established baseline. This calculator compares amount and 24‑hour velocity to typical values, caps extreme ratios, and converts each signal to a 0–1 scale. For example, an amount that is 3× typical maps higher than 1.2×, while anything at 5× is treated as maximum abnormality.
Weighting and governance
Weights translate business priorities into a repeatable model. If chargebacks are costly, increasing the chargeback weight shifts the score upward for accounts with elevated dispute rates. Governance teams can document each weight with evidence, such as historical loss contribution, and review changes quarterly to prevent drift and maintain consistent approvals across channels. Run back‑tests on at least 10,000 historical events, then lock parameters with formal change control and sign‑off records.
Thresholds and decision routing
Operational actions come from thresholds: review, step‑up, and block. Setting review at 45 and step‑up at 65 can route moderate cases to analysts while reserving stronger friction for higher scores. When false positives increase, a small threshold lift of 3–5 points can reduce manual workload without removing controls entirely. For digital goods, consider a lower block threshold and stronger step‑up during spikes.
Control strength adjustments
Verified controls lower risk because they raise attacker cost. Email and phone verification reduce the adjusted risk, while verified payment methods and two‑factor provide stronger offsets. In practice, a fully verified returning account can drop several score points, keeping low‑risk traffic flowing while still flagging stacked anomalies like geo mismatch plus high IP risk.
Monitoring performance metrics
Use outcomes to measure precision and recall. Track approval rates, review hit‑rate, and post‑decision chargeback percentages by score band. A healthy program might show chargebacks under 0.5% in low bands, 1–2% in moderate, and over 3% in critical. Recalibrate weights if bands stop separating outcomes.
Operational reporting and audits
Exports support evidence‑based control narratives. The CSV details raw inputs, normalized factors, and weighted contributions, enabling explainability during audits. The PDF summary is suitable for case files, listing the score, recommended action, and top drivers. Use these artifacts to standardize investigator notes and demonstrate consistent treatment across teams.
FAQs
1) What does the Fraud Risk Score represent?
It is a 0–100 indicator derived from weighted, normalized signals. Higher scores mean stronger abnormality and higher expected loss probability, given your current tuning and control offsets.
2) How should I choose typical amount and typical velocity?
Use a stable baseline: customer history, segment averages, or recent rolling medians. Update baselines regularly so ratio signals reflect true behavior changes, not seasonality.
3) Why does verification reduce the score?
Verified email, phone, payment method, and two-factor increase attacker cost and reduce takeover likelihood. The calculator applies small protective offsets to the adjusted risk before scoring.
4) Can I tune weights without breaking consistency?
Yes. Document the rationale, test against labeled outcomes, and change weights under governance. Apply the same configuration across channels, then monitor band outcomes to confirm separation.
5) What thresholds work for review and step-up?
Start with review around 40–50, step-up around 60–70, and block around 80. Adjust using your fraud rates, analyst capacity, and customer friction tolerance.
6) How do CSV and PDF exports help operations?
CSV supports analytics, model monitoring, and audit trails by listing inputs and factor contributions. PDF provides a compact case summary with score, action, and top drivers for investigator notes.