Sponsored-product advertising on e-commerce platforms typically uses static, per-campaign click bids that ignore the variable value of ad slots across contexts such as user intent, device, placement, and time. We present a production dynamic bidding system that adjusts CPC (Cost-Per-Click) bids in real time based on predicted post-click conversion probability. The system is built around three design choices: (i) it models post-click conversion probability, aligning the prediction target with the CPC charging event; (ii) it normalizes predictions at the campaign-SKU level, preserving each advertiser’s spending intent as the average operating point; (iii) it calibrates heterogeneous placement-specific signal to a shared CTR interface, enabling a single bidder to serve across incompatible surfaces. We provide a step-by-step derivation of the bid multiplier equation, detail the complete system architecture (model, feature pipeline, low-latency serving, and normalization design), and present results from production deployment on a large home-furnishings marketplace. Production experiments show positive and statistically significant gains in attributed conversions per click, first on the initial placements and later after expansion to additional placement types, with neutral customer-level conversion impact, further supporting the effectiveness of the approach.