Set fraud score cut-offs by segment
Use this tool to set fraud score cut-offs for decision automation. Drop in scored cases with outcomes and segments to compare one cut-off with the best per segment at the same review capacity, and see the gain for your channels in a minute.
Scored cases
20,000 cases with 351 fraud in 4 segments were read, and 0 rows without a score or a known outcome were left out.
The file is read in this browser and never uploaded. Only counts go anywhere, and only if you request the result by email.
Both the single cut-off and the segment cut-offs must fit this capacity, so the comparison uses the same review effort.
These are 20,000 generated cases in four channels, where the score runs higher for one channel and lower for another, for good and fraud alike.
Net gain
At ≥ 675, one cut-off catches 209 fraud and flags 591 good cases. Within the same 5% capacity, cut-offs per segment catch 283 and flag 705, a gain of €64,890.
Net by segment
One cut-off Segment cut-off
Segment results
| One cut-off | Segment cut-off | ||||||||
|---|---|---|---|---|---|---|---|---|---|
| Segment | Cases | Fraud rate | Flagged | Fraud caught | Net | Cut-off | Flagged | Fraud caught | Net |
| Online | 6,940 | 2.7% | 730 | 165 | €140,025 | ≥ 675 | 730 | 165 | €140,025 |
| Store | 6,112 | 0.5% | 27 | 16 | €14,235 | ≥ 643 | 44 | 19 | €16,725 |
| Partner | 4,002 | 2.8% | 16 | 15 | €13,485 | ≥ 527 | 171 | 85 | €75,210 |
| Phone | 2,946 | 0.7% | 27 | 13 | €11,490 | ≥ 643 | 43 | 14 | €12,165 |
| Total | 20,000 | 1.8% | 800 | 209 | €179,235 | – | 988 | 283 | €244,125 |
Get the result by email
The next step is the cut-offs as a decision table on segment and score, with every change tested on past cases. Start free
Cut-off method
Net = Fraud caught × Loss per fraud − Good flagged × Cost per good case One cut-off = max Net, Flagged ≤ Review capacity Segment cut-off = max Σ Net, Σ Flagged ≤ Review capacity Every cut-off, single or per segment, comes from the same list of 50 score values at equal steps across the cases. The gain comes from segmenting because the search uses the same cut-off list.
The segment cut-offs are chosen together, sharing review capacity across cases where flagging gains the most. They are fitted to this file, so results on new cases will be a little worse, especially for small segments.
Fraud score cut-offs by segment
A fraud score cut-off flags cases at or beyond a chosen score. This tool compares one cut-off for the whole book with cut-offs for each segment, using the score, outcome and segment in your file.
A score can behave differently across channels or products, so one cut-off may flag good cases in one segment and miss fraud in another. Compare fraud caught, good cases flagged and net within the same review capacity, then use the result to consider cut-offs for each segment.
Common questions
How do I set a fraud score cut-off?
The cut-off flags every case at or beyond the chosen score. This tool compares candidate cut-offs using fraud caught, good cases flagged and net.
Should the fraud threshold differ by channel or product?
It can differ when scores behave differently across segments. This tool compares one cut-off for the whole book with a cut-off for each segment at the same review capacity.
How does a segmented cut-off compare with one cut-off?
The tool compares fraud caught, good cases flagged and net for each approach. Both use the same list of candidate cut-offs and the same review capacity.
How do I choose cut-offs within review capacity?
Set the review capacity as the share of cases that may be flagged. The tool chooses the cut-offs with the best net within that capacity.
Why does one fraud score mean different risk across channels?
The demo shows scores running higher in one channel and lower in another for good cases and fraud alike. The tool compares cut-offs by segment using the outcomes in your file.