The Optimization We Didn't Build

Thursday, 15 October, 2026 - 11:35–11:55

Leila Vayghan, Shopify

Over-engineering can be one of infrastructure’s most expensive habits. This talk tells the story of automating database fleet scaling for flash sales at a large-scale e-commerce platform, and the decision to reject a sophisticated, data-driven design in favor of a simpler fixed-target approach. We’ll walk through the technical tradeoff: a tempting per-shop, per-region scaling model based on historical QPS would have saved at most about $6K/month, while adding long-lived data complexity and operational risk. The simpler design decoupled flash-sale safety from baseline provisioning, reduced manual toil, and unblocked a separate right-sizing initiative that delivered $1.5M-$1.8M in annual infrastructure savings. Attendees will leave with a concrete case study in saying no to complexity, using cost analysis to guide architecture, and recognizing when the best engineering decision is the one you choose not to make.

Leila is an infrastructure engineer at Shopify focused on reliability and cost efficiency for large-scale cloud platforms. Her work often sits at the intersection of SRE practice, cross-team coordination, and pragmatic engineering tradeoffs.

Presentation Video