Planner

A fraud score is worth what its alerts stop. Pick how fast a confirmed alert blocks the card and how much to review.

purchases · compromised cards · Nov–Dec 2010

Share of purchases sent to an analyst
Planning values; replace them with your program's own
Net value, Nov–Dec 2010

Response speed

    Net value at the same review budget and costs. Blocking sooner stops more of each run of fraud.

    Fraud dollars stopped

    Cards blockedFraud total

    Analyst workload

    Alerts a dayTotal alertsAnalysts

    Customer friction

    False alertsWere fraud

    Savings

      Net value after costs

      Operating costs

        Costs as a share of savings

        Budget needed to reach a target

        Smallest review budget that stops the chosen share of fraud dollars

        Response speedBudgetAlerts a dayNet valueApply

        What the score is

        A boosted-tree model over 60 point-in-time variables: how much each card, merchant and pairing has spent, how often and how suddenly, over the last 0–30 days.

        Why the block matters

        Fraud comes in runs on a compromised card. A confirmed alert blocks the card and every later fraud on it is declined. That, not the flagged purchase, is where the value is.

        Why the numbers hold

        The model and cutoffs were chosen on Jan–Oct. Everything here replays Nov–Dec, which no choice touched. With 16 cards the range is wide; the report quantifies it.