Alloy Holdings is considering replacing part of its current fraud decisioning stack in the Alloy Dashboard with a new graph-based risk engine from a third-party vendor. The proposed technology could improve fraud detection for high-risk onboarding flows, but it would require changes across decisioning services, model monitoring, and analyst workflows. You are the Engineering Manager for a 9-person platform team and have been asked to lead the evaluation and, if justified, the first production rollout within one quarter.
The GM of Identity & Fraud wants measurable fraud-loss reduction before the next board review. The Head of Risk Operations wants better analyst explainability and no disruption to manual review queues in the Alloy Dashboard. The Infrastructure Director is concerned about latency, vendor lock-in, and ongoing operating cost. The Account Management team is pushing for a visible launch because two enterprise customers have asked about graph-based fraud controls.
You have 12 weeks, a fixed incremental budget of $180,000, and no additional headcount. The team includes 5 backend engineers, 2 ML/platform engineers, 1 staff engineer shared 30% with another initiative, and 1 TPM. The new engine must integrate with Alloy's decisioning APIs, stay under 120 ms added p95 latency, and meet internal security review by Week 5. A pilot can include at most 8% of onboarding traffic and only 3 design partners.