What is Loss rate?

--- title: What is Loss rate? source_page: https://www.sardine.ai/fraud-aml-glossary/loss-rate canonical: https://www.sardine.ai/fraud-aml-glossary/loss-rate format: text/markdown description: This is realized fraud or credit losses as a share of volume, the bottom-line measure of what actually got through and cost money. Unlike alert-based numbers, it reflects net losses after recoveries, so it is the truest read on damage done. Alerts can look busy while real losses tell a different story. --- **Quick links:** [Human page](https://www.sardine.ai/fraud-aml-glossary/loss-rate) · [Home](https://www.sardine.ai) · [Customers](https://www.sardine.ai/customers) · [Blog](https://www.sardine.ai/blog) · [Demo](https://www.sardine.ai/demo) --- # What is Loss rate? **Read time:** 4 This is realized fraud or credit losses as a share of volume, the bottom-line measure of what actually got through and cost money. Unlike alert-based numbers, it reflects net losses after recoveries, so it is the truest read on damage done. Alerts can look busy while real losses tell a different story. Loss rate is realized fraud or credit losses as a share of volume, the bottom-line measure of what actually got through and cost money. Unlike alert-based numbers, it reflects net losses after recoveries, so it is the truest read on damage done. What is loss rate, in plain English? Loss rate is the money you actually lost, divided by the volume you processed, usually expressed in basis points. It is the realized figure: not fraud you attempted to stop, not alerts you raised, but the net loss that survived your controls and your recoveries and landed on the books. The word "realized" is what makes it different from most detection metrics. Alerts, flags, and scores describe what your system did; loss rate describes what it cost. It nets out recoveries, the funds you clawed back or won in dispute, so it reflects the true damage rather than the worst case. In the detection stack loss rate is the ground truth that every other metric is ultimately judged against. A model can look busy and precise, but if the loss rate is climbing, something is getting through, which is why this is the number leadership tends to care about most. Loss rate versus alert-based metrics It is easy to confuse activity with outcome. Alerts measure how hard your system is working; loss rate measures what still cost you money. Alerts can look healthy while losses tell a different story. When loss rate rises but alert volume stays flat, a new fraud pattern is usually getting past detection that has not learned it yet. What it looks like in practice In practice A fraud team's dashboards look calm: alert volume is steady, review queues are on time, and the model's scores look stable. Nothing suggests a problem. Then finance flags that fraud charge-offs are creeping up quarter over quarter. The loss rate had been rising for two months while every activity metric stayed flat. A new scam pattern was sailing straight through, generating no alerts because the model had never seen it. The alert dashboards could not show a problem they were not detecting; only the realized loss rate, reconciled back to charge-offs, revealed it. The flat alerts against a rising loss rate was the tell. Why it is the number that counts Loss rate is the truest read on damage because it measures outcomes, not effort. A team can raise alerts, clear queues, and hit precision targets while losses climb, because the fraud that hurts most is the fraud the system never flagged. Reconciling loss rate back to actual charge-offs keeps everyone honest about what is really getting through. The catch is timing. Loss rate lags the fraud event and needs matured cohorts to read accurately, since both recoveries and late-surfacing fraud take time to land. Read it too early and it looks better than reality. It also pays to keep fraud losses and credit losses separate, because they behave differently and mixing them muddies both. Used well, loss rate is the anchor that tells you whether all the other metrics are actually working. What to watch in the data Rising loss rate, flat alerts. The classic signature of a new pattern getting past detection that has not learned it yet. Immature cohorts. Recent periods understate loss because recoveries and late fraud have not settled; read matured cohorts. Fraud versus credit mixing. Blending fraud and credit losses hides the story in each; keep them separate and reconcile to charge-offs. Recovery assumptions. A loss rate that assumes optimistic recoveries can understate true net loss if clawbacks fall through. Segment drift. A steady blended loss rate can hide one product or channel deteriorating fast. Quick questions ### How is loss rate different from fraud rate? Fraud rate may count gross fraud or incidents; loss rate is realized net loss after recoveries. Loss rate is the money that actually stayed lost, which makes it the more conservative, bottom-line figure. ### Why does it lag? Losses are only realized after disputes resolve and recoveries are attempted, which takes weeks or months. So a recent period understates the eventual loss rate until the cohort matures. ### Should fraud and credit losses be combined? Generally no. They have different causes, patterns, and owners. Keeping them separate lets each be managed properly, and reconciling both back to charge-offs keeps the totals honest. ### What does a flat alert count with rising loss mean? It usually means a new fraud pattern is getting through undetected, generating losses but no alerts because the system was never trained or ruled to catch it. It is a strong signal to investigate. ### Is loss rate the same as charge-off rate? They are closely linked; charge-offs are where realized losses ultimately book. Reconciling loss rate to charge-offs is how teams confirm their fraud numbers match what finance actually wrote off. Go deeper FFIEC BSA/AML Examination Manual ↗ — The manual US examiners use to assess BSA and AML programs. FTC Consumer Advice: Scams ↗ — US consumer guidance on current scams and fraud, and how to report them.