Choose a fraud score cut-off in a minute
Enter fraud and good customer counts by fraud score band, then set what each mistake costs. See the cut-off that saves the most and what it catches or flags.
Band a scored file
The file is read in this browser and never uploaded. Only the band counts go into the link and into the email if you ask for it.
Cut-off that saves most
Flag from [500,600) saves the most, €605,400, but flags more than your 3% capacity. Within it, Flag from [600,700) saves €567,900.
Try another cut-off
| Fraud | Good | |
|---|---|---|
| Flagged | 711 | 2,300 |
| Passed | 89 | 63,700 |
At this cut-off you catch 711 of 800 fraud, stop 2,300 good customers and save €605,400.
Money saved by cut-off
The dark line shows money saved, the orange line shows catch rate on the right-hand scale, and the dot marks the best cut-off.
Cut-off results
| Cut-off | Flagged | Fraud caught | Good flagged | Precision | Catch rate | False-positive rate | Share flagged | Money saved |
|---|---|---|---|---|---|---|---|---|
| None flagged | 0 | 0 | 0 | – | 0.0% | 0.0% | 0.0% | €0 |
| Flag from [900,1000) | 296 | 226 | 70 | 76.4% | 28.2% | 0.1% | 0.4% | €202,350 |
| Flag from [800,900) | 666 | 416 | 250 | 62.5% | 52.0% | 0.4% | 1.0% | €370,650 |
| Flag from [700,800) | 1,156 | 556 | 600 | 48.1% | 69.5% | 0.9% | 1.7% | €491,400 |
| Flag from [600,700) | 1,851 | 651 | 1,200 | 35.2% | 81.4% | 1.8% | 2.8% | €567,900 |
| Flag from [500,600) ◀ | 3,011 | 711 | 2,300 | 23.6% | 88.9% | 3.5% | 4.5% | €605,400 |
| Flag from [400,500) | 5,246 | 746 | 4,500 | 14.2% | 93.3% | 6.8% | 7.9% | €603,900 |
| Flag from [300,400) | 9,768 | 768 | 9,000 | 7.9% | 96.0% | 13.6% | 14.6% | €556,200 |
| Flag from [200,300) | 18,782 | 782 | 18,000 | 4.2% | 97.8% | 27.3% | 28.1% | €433,800 |
| Flag from [100,200) | 36,792 | 792 | 36,000 | 2.2% | 99.0% | 54.5% | 55.1% | €172,800 |
| Flag from [0,100) | 66,800 | 800 | 66,000 | 1.2% | 100.0% | 100.0% | 100.0% | −€270,000 |
The copy pastes into a decision table as two columns, with the score band as the condition and pass or review as the result.
Get the result by email
Set the cut-off as a rule in the decision flow, with the review queue behind it, then test each change on past applications. Start free
Show formulas
Precision = Fraud caught ÷ Flagged Catch rate = Fraud caught ÷ All fraud False-positive rate = Good flagged ÷ All good customers Money saved = Fraud caught × Loss per missed fraud − Good flagged × Cost per good customer flagged
These figures use past labels only. Fraud found later, or deterred by the friction, is not included.
Fraud score threshold calculator
A fraud score cut-off sets which cases are flagged for review. This calculator uses the fraud and good customer counts in each score band, along with the costs of missed fraud and flagged good customers.
The cut-off changes how much fraud is caught and how many good customers are flagged. Compare the money saved at each cut-off and see which options fit your review capacity.
Fraud score questions
How do I choose a fraud model cut-off?
Compare the money saved at each cut-off using the fraud and good customer counts and the costs of each mistake. The calculator shows the cut-off that saves the most on those counts and the best cut-off within your review capacity.
What do precision and recall mean for a fraud score?
Precision is the share of flagged cases that are fraud. The catch rate, also called recall, is the share of all fraud that is caught.
How do I read a confusion matrix for fraud?
It compares cases flagged or passed with whether they were fraud or good customers. It shows fraud caught, good customers flagged, fraud missed and good customers passed.
How do I weigh fraud losses against false positives?
Set the loss for a missed fraud and the cost of flagging a good customer. The calculator uses those costs and the counts in each band to work out money saved at each cut-off.
How does review capacity affect a fraud cut-off?
Review capacity sets the most cases the team can take as a share of all cases. The calculator shows the best cut-off that fits within that capacity.