Track alerts, investigations, and confirmed fraud outcomes easily. See detection rate trends across reporting periods. Download results to share with audit and leadership teams.
Use this sample to understand how the metrics connect.
| Scenario | T | A | I | TP | FP | FN | Case detection rate | Precision |
|---|---|---|---|---|---|---|---|---|
| New ruleset rollout | 120,000 | 3,600 | 2,400 | 320 | 2,080 | 110 | 74.42% | 13.33% |
| Mature model baseline | 120,000 | 2,100 | 1,900 | 410 | 1,490 | 60 | 87.23% | 21.58% |
| High-risk merchant spike | 60,000 | 3,000 | 2,700 | 520 | 2,180 | 240 | 68.42% | 19.26% |
Fraud programs should track case-based detection using true positives and false negatives. If TP=410 and FN=60, detection reaches 87.23%, matching the sample baseline. Many teams set quarterly targets above 80% for mature models, while new rule sets may sit near 70%. Monitor week-over-week variance; a five point drop often signals data drift, new attack paths, or delayed labels. Include confidence intervals when counts are too small.
Precision shows workload quality. In the baseline example, TP=410 and FP=1,490 yields 21.58% precision. Raising precision by five points can remove hundreds of unnecessary reviews when investigations exceed 1,900 monthly. Track false positive rate when TN is available; a 1% FPR at 120,000 transactions produces 1,200 noisy cases. Use segmented metrics by channel, merchant, or device to find concentrated noise. Review queues daily to calibrate.
Operational rates connect detection to capacity. Alert rate equals alerts generated divided by total transactions; 2,100 alerts on 120,000 transactions is 1.75%. Investigation rate equals investigated divided by generated; 1,900 of 2,100 is 90.48%. Investigation yield equals TP divided by investigations; 410 of 1,900 is 21.58%. A yield under 10% usually indicates overly broad rules or weak scoring thresholds. Capacity plans should reflect peaks.
Value coverage adds a financial lens. If detected fraud value is 185,000 and total confirmed fraud value is 212,000, coverage is 87.26%. Teams often prioritize high value recovery even when case detection is stable, because a small number of large incidents can dominate loss. Compare prevented loss to investigation cost; if analyst time is 9.5 minutes per case at 18.75 per hour, investigation cost is 5,640.
Use balanced scorecards for governance. Pair detection rate with precision and cost per confirmed fraud to prevent gaming one metric. When TP rises but FN also rises, re-check labeling lag and dispute pipelines. If precision rises while coverage value falls, thresholds may be excluding high-risk outliers. Export CSV for audit trails and store PDFs with model version, feature changes, and approval notes for every reporting period. Document overrides and exceptions.
At minimum, enter true positives and false negatives. The calculator then reports case detection as TP divided by TP plus FN. Add total transactions to derive true negatives and compute false positive rate and accuracy.
Detection focuses on missed fraud, while precision reflects review quality. If alerts are broad, FP grows faster than TP, pushing precision down even when detection stays high. Tuning thresholds and segmentation can improve precision.
The tier is a simple benchmark based on case detection. Strong is 90% or higher, Moderate is 75% to 89.99%, Watchlist is 60% to 74.99%, and Weak is below 60%. Use it as a discussion starter.
Leave TN blank and provide total transactions, TP, FP, and FN. The tool will derive TN as total transactions minus TP minus FP minus FN, when totals are consistent. If totals conflict, TN is left empty.
Investigation cost equals investigated alerts times average minutes per investigation divided by 60, multiplied by the hourly analyst rate. This produces a labor estimate and supports cost per confirmed fraud calculations.
Use value coverage when losses are uneven, such as card-not-present fraud spikes or takeover events. It highlights how much confirmed fraud value was detected, not just how many cases. Track both to avoid missing high-impact outliers.
Important Note: All the Calculators listed in this site are for educational purpose only and we do not guarentee the accuracy of results. Please do consult with other sources as well.