Metric Definition
Fraud catch rate
Track from
Card fraud detection rate
Card fraud detection rate is the percentage of fraudulent card transactions that your fraud controls correctly identify and stop, out of all fraudulent transactions attempted. It tells you how much of the fraud aimed at your card programme you actually catch. A high detection rate protects revenue and cardholders, but only matters when read alongside the false positives it creates.
7 min read
What is card fraud detection rate?
Card fraud detection rate is the percentage of fraudulent card transactions that your fraud controls correctly identify and stop, out of all fraudulent transactions attempted. If 1,000 fraudulent transactions were attempted in a month and your systems caught 940 of them, your detection rate is 94 percent. It is the single clearest measure of how well your fraud defences are working.
The metric matters because it ties a number to an outcome teams care about. Every percentage point you fail to detect becomes a chargeback, a refund, or a loss the business absorbs. But detection rate never travels alone. Pushing it higher almost always raises the number of legitimate transactions you wrongly decline, so it has to be read next to your false positive rate and approval rate to mean anything.
Definition note
Card fraud detection rate should count only confirmed fraud, not everything your rules flag. A transaction your model blocks but a cardholder later confirms as genuine is a false positive, not a catch. Counting flags instead of confirmed fraud inflates the rate and hides how much good business you are turning away.
How to calculate card fraud detection rate
The formula divides the fraud you caught by all the fraud that was attempted, then multiplies by 100. The hard part is not the arithmetic, it is the denominator. You only know about missed fraud after the fact, once a chargeback or a cardholder dispute lands, so the true detection rate for a given month firms up over the following weeks as disputes settle.
Most teams report a provisional rate at month end and revise it as the dispute window closes. Be explicit about which version a number is, because a 96 percent provisional rate can drift to 91 percent once late chargebacks arrive.
- 1
Fraudulent transactions caught
Count every transaction your controls declined, held, or flagged that was later confirmed as fraud. This is your numerator.
- 2
Fraud missed
Count fraud that completed and surfaced later as a chargeback, dispute, or written-off loss. These are the cases your controls let through.
- 3
Total fraudulent transactions attempted
Add caught and missed fraud together. This denominator is only complete once the dispute window for the period has closed.
- 4
Divide and convert to a percentage
Divide caught fraud by total attempted fraud and multiply by 100. Report whether the figure is provisional or final.
Card fraud detection rate in a metric tree
A headline detection rate of 92 percent tells you where you stand but not what to do. Decomposing it into a metric tree shows which part of the defence is leaking. Detection rate is the product of how well each layer of control performs, from the data feeding your models to the analysts reviewing edge cases.
The gap between a dashboard that reports 92 percent and a decision about which control to tune is exactly the gap a metric tree closes. KPI Tree breaks the rate into its causal drivers, attaches RACI ownership to each branch so the fraud, data, and risk teams each own their node, and pushes an alert to the accountable owner when their branch moves. When detection slips, you see whether the cause is stale model features, a rule that needs retuning, or a review queue that is backed up.
Metric tree insight
A falling detection rate usually traces to one branch, not the whole system. When detection drops three points but model recall holds steady, the cause is often a backed-up review queue letting flagged fraud settle while analysts catch up. The tree points the fraud team at the queue, not at retraining the model.
Card fraud detection rate benchmarks
Detection rate benchmarks depend heavily on transaction mix, geography, and how aggressively a programme trades approvals for safety. Card-not-present volume is harder to police than card-present, so a card-not-present heavy programme will sit lower. Read these ranges as starting points, and always pair detection rate with the false positive rate it costs you.
| Detection rate | Reading | Typical trade-off |
|---|---|---|
| Below 85 percent | Underperforming controls | High fraud losses, models or rules likely stale |
| 85 to 92 percent | Common range | Workable balance, room to tune by segment |
| 92 to 97 percent | Strong performance | Watch false positives climbing alongside it |
| Above 97 percent | Very high catch rate | Often paid for with declined genuine transactions |
How to improve card fraud detection rate
Improving detection rate is rarely about one big change. It comes from tightening the layers that feed it, retraining on recent fraud, closing the gap between a chargeback and the model learning from it, and giving analysts the time to clear edge cases properly. Every gain has to be weighed against the genuine transactions it might block.
Retrain on recent fraud
Fraud patterns shift fast. Retrain models on the latest confirmed cases so recall does not decay against new attack methods.
Shorten the feedback loop
Feed confirmed chargebacks back into the model quickly. The faster the loop, the sooner detection adapts to emerging fraud.
Tune rules by segment
Set thresholds per channel and geography rather than one global rule. Card-not-present and high-risk regions need tighter controls than low-risk in-person spend.
Resource the review queue
A backed-up manual queue lets flagged fraud settle before anyone looks. Staff it so high-risk cases are cleared inside the settlement window.
Common mistakes when tracking card fraud detection rate
- 1
Reading detection rate in isolation
A 99 percent detection rate that declines a tenth of genuine transactions is a bad trade. Always pair it with false positive rate and approval rate.
- 2
Treating provisional as final
Late chargebacks lower the true rate. Reporting a month-end provisional figure as settled overstates how well controls performed.
- 3
Counting flags as catches
Flagged is not the same as confirmed fraud. Counting every flag inflates the rate and hides false positives.
- 4
Ignoring channel mix
A blended rate hides a weak channel. Card-not-present fraud can drag the programme down while card-present looks healthy.
Related metrics
Cart abandonment rate
Checkout drop-off
Operations Metrics
Metric Definition
Cart Abandonment Rate = (1 − Completed Purchases / Carts Created) × 100
Cart abandonment rate measures the percentage of online shopping carts that are created but not converted into completed purchases. It is one of the most impactful e-commerce metrics because it represents revenue that was within reach but lost at the final stage of the buying journey.
Checkout conversion rate
E-commerce metric
Ecommerce & Marketplace MetricsMetric Definition
Checkout Conversion Rate = (Completed Purchases / Checkout Starts) x 100
Checkout conversion rate measures the percentage of users who begin the checkout process and successfully complete their purchase. It isolates the final stage of the buying funnel, from the moment a shopper initiates checkout to the order confirmation page. This metric is critical for e-commerce businesses because the checkout is where purchase intent is highest, and any friction at this stage directly destroys revenue that was nearly captured.
Escalation rate
Customer Support Metrics
Metric Definition
Escalation Rate = (Escalated Tickets / Total Tickets Handled) x 100
Escalation rate measures the percentage of support tickets that are transferred from one tier or team to a higher tier or specialist group for resolution. It reflects the gap between the issues customers raise and the ability of frontline agents to resolve them, making it a key indicator of agent readiness, process maturity, and product complexity.
Customer satisfaction score
CSAT
Product Metrics
Metric Definition
CSAT = (Satisfied Responses / Total Responses) × 100
Customer satisfaction score measures how satisfied customers are with a specific interaction, product, or experience. Unlike NPS which measures loyalty, CSAT captures satisfaction at a moment in time, making it ideal for evaluating specific touchpoints in the customer journey.
Metric decomposition
Metric Definition
Break the card fraud detection rate into its underlying drivers so you can see which transaction segments or detection rules move the catch rate.
Metric trees for fintech companies
Metric Definition
See how fintech teams place the card fraud detection rate alongside the other risk and payment metrics it trades off against.
Build card fraud detection rate as a metric tree
Stop reading detection rate as a single number on a dashboard. Decompose it into model performance, rules, review capacity, and data quality in KPI Tree, give each branch an accountable owner, and get an alert the moment a layer starts leaking.
