Test fraud velocity rules for automated lending decisions
Drop in past applications with their time, outcome and identifiers. Choose a window and count to see, in a minute, what a velocity rule on each identifier would catch and stop as the page replays the applications in time order.
Applications file
2,903 applications were read, including 127 known bad and 2,322 known good, and 0 rows without a readable time were left out.
The file is read in this browser and never uploaded. Only counts go anywhere, and only if you ask to get the result by email.
Identifier columns: email, phone, device_id, ip.
The rule fires on the application that brings the count for the same value within the window to this number, so only earlier applications can make it fire.
These are 3,000 generated demo applications over 60 days, with bursts on shared devices and phones planted among the bads and a busy shared IP among the goods.
Known bads caught
A rule that fires at 3 applications with the same detail within 1 day would have caught 29.9% of known bads and stopped 102 known goods, equal to 4.4% of known goods.
Bads caught and goods stopped
Each line is one identifier, from a count of 2 at the top right to 10 at the bottom left, and the large dot is the chosen count. Up and to the left is better.
Results by identifier
| Identifier | Flagged | Bads caught | Goods stopped | Catch rate | Precision | False positive rate |
|---|---|---|---|---|---|---|
| 0 | 0 | 0 | 0.0% | – | 0.0% | |
| phone | 35 | 24 | 5 | 18.9% | 82.8% | 0.2% |
| device_id | 33 | 22 | 5 | 17.3% | 81.5% | 0.2% |
| ip | 128 | 4 | 95 | 3.1% | 4.0% | 4.1% |
| Any identifier | 178 | 38 | 102 | 29.9% | 27.1% | 4.4% |
Email your result
The same velocity counts can run on every application at the moment of the decision. Start free
How rules are tested
Count = Applications with same value (Window) Rule fires = Count ≥ Count to fire Catch rate = Bads caught ÷ Known bads Precision = Bads caught ÷ (Bads caught + Goods stopped) False positive rate = Goods stopped ÷ Known goods
The applications are replayed in time order, so each rule sees only earlier applications. Values match exactly after cleaning, and a match is a reason to look. It does not prove fraud.
What velocity rules show
A velocity check counts how often the same identifier appears within a chosen time window. This tool replays past applications in time order and tests whether the chosen count makes a velocity rule fire.
The results show known bads caught and known goods stopped for each identifier. Compare the catch rate and false positive rate, then use the results to choose a count to test before the rule goes live.
Common questions
What are velocity checks in fraud prevention?
A velocity check counts applications that share an identifier within a chosen time window. The rule fires when the count reaches the chosen number.
How do I set a velocity limit for a device or phone?
Choose a window and count, then replay past applications with the tool. Compare the known bads caught with the known goods stopped for each identifier.
How many applications per device in a day is too many?
Test a count on past applications with a one-day window. The results show how many known bads it catches and known goods it stops.
How can I test a velocity rule on past applications?
Load a CSV file or paste your past applications, then choose the time, outcome and identifier columns. The tool replays the applications in time order and shows the results for each identifier.
How do I balance fraud caught against good customers stopped?
Compare the catch rate with the false positive rate for each identifier and count. A match is a reason to look at an application, not proof of fraud.