Which Fraud Pattern Deserves the First Control?
Build frequency, loss and exposure-rate views; preserve a severe-event override; and choose a causal next step without claiming that a leading category is a root cause.
28 minutes
Scenario
A payment provider can fund one fraud-control experiment this quarter. Operations wants the most frequent category, finance wants the largest loss, and risk wants every severe event escalated. A detection rule changed halfway through the month, so raw counts are not directly comparable over time.
- Your role
- Fraud strategy analyst
- Method
- Pareto Analysis
Evidence pack
Account takeover
40 confirmed cases; €160,000 loss; 80,000 exposed remote-card transactions.
Merchant non-delivery
260 cases; €104,000 loss; 400,000 exposed purchase transactions.
Duplicate charge
190 cases; €19,000 loss; 500,000 exposed purchase transactions.
Cash withdrawal
12 cases; €144,000 loss; 40,000 exposed cash withdrawals.
Other
98 cases; €49,000 loss. The exposure population is mixed and 31% of records have no analyst note.
Detection change
A new account-takeover rule started on day 16; confirmed detections per day doubled while average loss per detected case fell 35%.
Risk threshold
Internal risk appetite requires immediate review of any single event above €25,000, regardless of category frequency.
Capacity
The experiment team can investigate one category deeply and add one low-effort containment control.
Constraints
- Frequency, total loss, loss per case and exposure rate answer different questions.
- Pre- and post-rule periods must be separated for account takeover.
- The Other category cannot be treated as homogeneous.
- A Pareto category is not a cause.
Case steps
Work through each prompt using the evidence pack. These guided cases support self-directed practice; server-scored attempts are not available yet.
Choose the primary measure for this quarter's control investment and define two secondary views that protect against a misleading ranking.
Calculate loss per case and cases per 10,000 exposed transactions for e1–e4. Do not calculate a rate for Other.
Explain how e5–e7 limit a basic Pareto chart and specify the minimum data repair or stratification before committing the experiment.
Recommend one category for deep investigation and one low-effort containment control. Show why a reasonable analyst could choose differently.
Design the causal investigation and post-action measurement. Include a comparison group or period and a condition that would stop or redirect the experiment.