Decisions · Risk and fraud
Should you use a taxonomy for loyalty fraud
Adopt a fraud taxonomy only when each category changes at least one control decision: block, review, clawback, or ignore. A taxonomy that merely labels incidents for reporting costs more than it returns. If your programme treats most fraud the same way, skip the taxonomy and fix the controls.
The decision
The decision is not whether fraud types exist. It is whether formal categories improve the speed and consistency of intervention. A taxonomy is a control instrument, not a research exercise. Underneath it, you are deciding how many distinct treatment paths your loyalty programme is willing to operate. Each category should force a different next action: suspend the account, limit redemption, reverse the credit, escalate to manual review, or do nothing. If two categories always trigger the same action, merge them. The taxonomy is sound only when the number of categories equals the number of materially different responses you can afford to run.
What it costs you
A taxonomy has three running costs: design time, calibration, and drift correction. Design time includes interviews with fraud analysts, redemption operations, and member support, often four to six weeks of part-time work. Calibration is worse. A new label such as points laundering requires teams to agree on threshold behaviour: how many transfers, to how many newly registered accounts, within what window. Drift correction recurs as loyalty mechanics change, because a category built for transfer abuse may not survive a new gift card redemption path.
An illustrative calculation: a programme with two full-time fraud analysts devoting eight hours per month each to taxonomy maintenance spends roughly 4,600 dollars per year at a loaded cost of 24 dollars per hour. Eight hours times two people times twelve months is 192 hours. At 24 dollars per hour, that is 4,608 dollars. If those same hours could instead be spent reviewing ten high-value redemption cases per month at an average prevented loss of 300 dollars per case, the forgone recovery is 36,000 dollars per year. The taxonomy must improve detection enough to beat that. If it only improves reporting labels, it does not.
How to decide
Apply a two part test to your own programme. First, write down the five most common fraud patterns you observe, without using a taxonomy. For each pattern, describe the intervention you would order if you saw it today. Second, compare the interventions. If every pattern collapses into block and claw back, you do not need a taxonomy. If at least three patterns require genuinely different treatment, such as manual review versus instant block versus member education, a taxonomy earns its keep. Then set a rule: any category that does not change a control must be deleted within one quarter. This keeps the taxonomy a decision tool rather than a filing system.
What would change our mind
The recommendation flips if your programme operates in a jurisdiction or partnership structure where fraud categories carry legal or contractual weight. For example, if a scheme rule requires you to report account takeover separately from promotion abuse, or if a banking partner reimburses only certain labelled fraud types, then a taxonomy becomes a compliance instrument. In that case, even categories that do not change your internal controls may still be mandatory, and the cost of not maintaining them may exceed the overhead. We would also change the recommendation if your fraud volume is high enough that a single taxonomy allows automated routing with clear audit trails, measured by a reduction in manual review time rather than by label count. The reversal condition is narrow, but real: external reporting obligations or audit requirements change the return on classification.