As volume grows, is fraud rate staying flat?
№ 105 · Fraud rate vs volume
Definition
MetricFraud losses as a share of processed volume, plotted against volume growth
Unitbasis points of processed volume
Scatter with volume on x and fraud rate on y. Goal is to scale volume without proportional fraud. Build by aggregating fraud incidents vs volume per period. Curve drifting up-right means you're losing on scale. Example: $4M volume at 0.08% fraud; $14M volume at 0.04% fraud (improving).
Benchmarks
| Bottom 30% | Median | Top 30% |
|---|---|---|
| Above 15 bps, the level seen on US debit including issuer and merchant losses, and higher again on card-not-present weighted books | 6 to 7 bps of total card volume worldwide | Below 5 bps of card volume, achievable on domestic card-present weighted books |
Gross card fraud losses across issuers, merchants and acquirers, expressed in cents per USD 100 of card volume. Global card industry 2022 to 2024, with US debit as a separately measured 2023 sub-population. · The two headline benchmarks are not the same measurement. Nilson's 6.43 cents per USD 100 in 2024 is worldwide across all card types and all loss-bearing parties. The Federal Reserve's 17.6 bps for 2023 covers US debit and prepaid transactions at covered issuers only, and it counts losses to all parties. The gap is composition, not contradiction, and any single fintech should benchmark against the sub-population it actually resembles. US card fraud is disproportionate: 41.87% of global losses on 26.31% of global volume in 2024. Note also that the Fed series has risen steadily from 7.8 bps in 2011 to 17.6 bps in 2023, so a flat fraud rate in a growing US debit book is a genuine achievement rather than a neutral outcome.
| Category | Bottom 30% | Median | Top 30% |
|---|---|---|---|
| All card types worldwide, all loss-bearing parties | about 6.8 bps | about 6.6 bps | about 6.4 bps |
| US-issued cards | card-not-present concentrated, above the US blended rate | 41.87% of global losses on 26.31% of global volume in 2024, implying roughly 10 bps | not separately published |
| US debit and prepaid, covered issuers, issuer plus merchant losses | rising trend from 7.8 bps in 2011 | 17.6 bps in 2023 | not separately published |
When it looks bad
The fraud line bends upward at the exact point the volume line accelerates, so the two curves diverge instead of running parallel, which means the fraud controls were calibrated for the old volume and the new volume is arriving through a channel they were not built for.
Volume grows 18% quarter on quarter and fraud moves from 7 bps to 19 bps in the same quarter. On EUR 240m of quarterly volume that is EUR 288k of incremental loss, which is more than the incremental take rate revenue on the growth.
What to do about it
- Split the chart by card-present and card-not-present before touching the model. US fraud is overwhelmingly a card-not-present problem, so a blended line moving up is usually a channel mix shift rather than a control failure, and the fix is channel-specific rather than global (Nilson Report 2024).
- Treat media mix as a fraud control, not just a performance one. In European finance apps, affiliate networks carried the highest fraud exposure while self-reporting networks and DSPs delivered more reliable traffic, so reallocating spend away from affiliates cuts fraudulent signups upstream of any transaction monitoring (AppsFlyer with Sensor Tower, 2026).
- Benchmark against the right sub-population and set the internal ceiling accordingly. A US debit-weighted programme should hold against the 17.6 bps Fed figure, not the 6.43 bps global Nilson figure, and using the wrong one understates the required reserve by roughly two thirds (Federal Reserve 2023 debit report, Nilson Report 2024).
- Price fraud into the take rate by transaction type rather than carrying it as a central cost. If CNP volume carries 20 bps of loss against a 90 bps take rate, the contribution is still positive, but a flat-priced product will cross-subsidise the worst segment until it stops growing.
Sources
- The Nilson Report US cards carried 41.87% of worldwide fraud losses on 26.31% of global card volume in 2024, with losses overwhelmingly card-not-present. nilsonreport.com ↗
- The Nilson Report (press release) Gross card fraud measured 6.43 cents per USD 100 of total volume in 2024, down from 6.58 cents in 2023. finance.yahoo.com ↗
- The Nilson Report (press release via GlobeNewswire) Fraud of 6.58 cents per USD 100 in 2023 against 6.81 cents in 2022, with the US taking 42.32% of global losses on 25.29% of volume. globenewswire.com ↗
- Payments Dive Nilson projects USD 403.88bn of cumulative global card fraud losses over ten years, concentrated in US card-not-present volume. paymentsdive.com ↗
- Board of Governors of the Federal Reserve System Average interchange in 2023 was USD 0.22 per covered dual-message transaction, USD 0.24 per covered single-message transaction and USD 0.52 per exempt transaction; total debit and prepaid interchange was USD 34.12bn. federalreserve.gov ↗
- Board of Governors of the Federal Reserve System Exempt dual-message interchange fell from USD 0.64 in 2021 to USD 0.62 in 2023, while exempt single-message rose to USD 0.27; network fees averaged USD 0.129 per transaction, 0.28% of average transaction value. federalreserve.gov ↗
- ABA Banking Journal, reporting the Federal Reserve biennial debit report Fraud losses to all parties across covered issuers were 17.6 bps of transaction value in 2023, up steadily from 7.8 bps in 2011. bankingjournal.aba.com ↗
- AppsFlyer, with Sensor Tower and Google Ads European finance installs reached 960m in 2025 with 0.4% growth; banking app users spent about 12.5 hours per download in early 2026 against 8.8 hours for investment and money management apps; web-to-app drove 41.8% of owned media conversions; nearly one in two Western European investment app installs was flagged as fraudulent. appsflyer.com ↗