The forecast that inverted the planning assumption
A 40,000-run Monte-Carlo forecast of a new region's human-review queue. It found that volume tracks operational decisions rather than fraud inflow, with one ring driving 71% of all tasks ever queued.
Business Analyst, AWS Payments & Fraud Prevention
Replaced a point-guess capacity plan with p10/p90 thresholds whose coverage was measured rather than assumed, plus the counter-intuitive, defensible finding that operational decisions, not fraud inflow, drive queue volume.
fraud-labelled sign-ups month-over-month while human tasks halved, absorbed by new automation
A new region needed an 18-month capacity plan for its human investigation queue. The natural assumption, that more fraud means more review work, was about to be baked into headcount planning without ever being tested.
From 28 complete weeks of task history I built a 40,000-run Monte-Carlo simulation producing average, p10, and p90 planning thresholds, with the coverage of that range measured against held-out data rather than assumed. The distributional work surfaced the real driver: one fraud ring accounted for 71% of all tasks ever queued. In one month, fraud-labelled sign-ups rose ~50% month-over-month while human tasks halved, because new automation closed cases before they reached a person. I ran an anchor-stability pass, an adversarial pass, and a ring-excluded companion scenario, then made and argued the conservative call to keep the ring in the baseline so the forecast would not understate capacity needs. Delivered as a stakeholder two-pager, an Excel model, two notebooks, and an animated explainer.
The central finding inverted the planning premise: queue volume is driven by operational decisions such as bulk campaign pushes and automation launches, rather than by fraud inflow. Capacity planning got probabilistic thresholds with measured coverage instead of a point guess, and a defensible, argued baseline choice rather than a silent one.
Headcount gets planned against fraud-inflow growth, the one variable that turned out not to drive the queue, and the next bulk campaign push or automation launch blindsides the plan in either direction.
Every capacity plan hides a theory. This queue’s theory was that review work scales with fraud inflow. Reasonable, unexamined, and about to become headcount.
I tested it with distribution instead of intuition: 28 complete weeks of task history feeding a 40,000-run Monte-Carlo simulation, producing average, p10, and p90 planning thresholds. What makes those numbers usable is that their coverage was measured against held-out reality instead of assumed from the model’s own confidence.
The distribution told a different story than the theory. One ring accounted for 71% of every task ever queued. In one month, fraud-labelled sign-ups rose about 50% month-over-month while human tasks halved, because new automation was closing cases before they ever reached a person. Queue volume tracks operational decisions, bulk campaign pushes and automation launches, rather than fraud inflow.
A finding that inverts the premise needs to be attacked before it is believed, so it got an anchor-stability pass, an adversarial pass, and a ring-excluded companion scenario. Then came the judgment call. With the dominant ring removed the forecast looks very different, but I argued for keeping it in the baseline, since a capacity plan should be conservative and rings recur.
It shipped in the forms people actually use: a two-page stakeholder brief, an Excel model planners can poke at, two notebooks for the record, and an animated explainer for the room.