Kill your onboarding: selling to 10,000+ new users a day

A Solutions Engineer at Railway shares how they tackled a PLG targeting problem: with 10,000+ daily signups and no firmographic data collected at signup, 21,000 ICP-fit accounts were going unengaged. The solution was a deliberately simple weighted scoring model built in Hex using behavioral signals from PostHog and product telemetry — trust center downloads, seat growth, credential resets, DB connections, and deploy failure patterns. High-scoring accounts get routed to sales; mid-tier accounts receive event-triggered emails via Customer.io sent from real people offering help, not pitching. Day-one results showed 50–70% open rates vs. a 3% baseline. Key lessons: instrument events early, brute-force feature selection before modeling, discard deterministic signals, and keep scoring legible for the humans acting on it.