False Negative Rate

The false negative rate (FNR) is the proportion of genuinely bad cases that a detection system fails to catch. In financial crime work, a false negative is a fraudulent payment, sanctioned party or money laundering pattern that passes through without an alert, so the FNR shows how much real risk is being missed. It is calculated as false negatives divided by the total number of actual positive cases.

The FNR is the counterpart of the false positive rate, which measures how often legitimate activity is wrongly flagged. Banks and payment firms usually find false positives easy to count, because every alert is reviewed, but false negatives are hard to see, because a missed case produces no alert. The measure comes from statistics and is used in fraud detection, sanctions screening and transaction monitoring alike.

Think of a fishing net meant to catch large fish while letting small ones swim through. Some large fish will still slip out through a torn patch or around the edge. If 100 large fish swam into the area and the net caught 90, ten were missed; the false negative rate is the share that got away, here one in ten.

A detection system is that net for bad payments. Every fraud or laundering case it does not flag is a fish that got away, and the firm often does not know it escaped until much later, when a customer complains or the police call. That is why measuring this number takes extra work: you have to go looking for the fish you did not catch.

The false negative rate is:

• $FN$ (false negatives) is the number of bad cases the system did not flag.
• $TP$ (true positives) is the number of bad cases the system did flag.
• $FN + TP$ is the total number of actual bad cases.

The FNR is the complement of the detection rate, also called recall or sensitivity:

For example, suppose a card issuer later confirms 500 fraudulent transactions in a month. Its fraud engine had blocked or alerted on 420 of them and missed 80. The false negative rate is $80 \div (80 + 420) = 80 \div 500 = 0.16$, or 16%, and the detection rate is 84%. If the engine also flagged 9,580 legitimate transactions, those are false positives, which affect customer experience and workload but do not enter the FNR calculation.

The difficulty is knowing the true number of bad cases. Firms use several sources:

• Confirmed fraud reports. Customer disputes, chargebacks and fraud claims reveal card and payment fraud that the system did not stop, often weeks later.
• Below-the-line testing. In transaction monitoring tuning, analysts review samples of activity just under current thresholds to see how much suspicious activity sits there.
• Back-testing against known cases. Customers who later became the subject of a suspicious activity report through other routes, such as staff referrals or law enforcement requests, are checked to see whether any transaction monitoring scenario should have fired.
• Test data. Sanctions screening tools are tested with lists of known names and deliberate variations (spelling changes, transliterations) to measure missed matches.

Because AML outcomes are rarely confirmed by a court, the FNR for money laundering monitoring is usually an estimate with wide uncertainty, while card fraud FNR can be measured more directly.

Lowering a threshold catches more bad cases, reducing the FNR, but it also flags more legitimate activity, raising false positives and the risk of an alert backlog. The right balance depends on the cost of each error. A missed sanctions match can lead to severe penalties, so screening tools are usually set to accept many false positives. In fraud, a missed fraud costs the loss amount, while a false decline costs a sale and annoys a customer, so firms often weigh both in money terms. Detection quality also depends on human review: an alert raised but wrongly closed in alert disposition is a miss for the program as a whole, even if the software worked.

Regulators rarely set a numeric FNR target, but they expect firms to know whether their controls catch the risks they have identified. New York's Part 504 rule and US model risk guidance call for ongoing testing of detection logic, and the UK's Financial Conduct Authority has criticised firms that measure alert volumes without testing what they miss. A low false positive rate reported without any estimate of misses is treated as an incomplete picture of an AML program or fraud risk management framework.

• New York State Department of Financial Services
• Financial Conduct Authority