Background. Headless architectures, where the frontend and backend are decoupled and communicate through chains of API calls, are becoming more common in modern digital commerce. Despite this adoption, it remains unclear how API latency in these systems translates into user-perceived frontend performance, and organizations may not have formally defined performance thresholds to evaluate against. Standardized frontend metrics such as Largest Contentful Paint (LCP) and Interaction to NextPaint (INP) are often used as metrics for user experience, but their ability to capture latency in headless systems has not been empirically examined.

Objectives. We aim to empirically investigate how API response times affect frontend performance metrics in a production headless e-commerce system, and to demonstrate a repeatable method for identifying performance thresholds in the absence of pre-defined Service Level Agreements (SLAs).

Methods. We conduct a single-case study of a production headless e-commerce system, measuring API response times and standardized frontend performance metrics across baseline and simulated network conditions. User behavior context is collected from production analytics data.

Results. Frontend metrics score well under optimal conditions but degrade under different network conditions, while interaction responsiveness remains unaffected. A critical transactional endpoint consistently introduces a multi-second delay that neither metric captures, and this delay proves independent of network conditions, identifying it as a server-side bottleneck.

Conclusions. Standardized frontend metrics alone are insufficient for estimating performance in headless transactional flows. With 43% of users on mobile devices, Fast 4G represents a realistic everyday condition for a large share of users. Yet LCP already falls outside the good threshold under those conditions. The measurement approach demonstrated in this study, baseline measurement, standardized network condition testing, and threshold identification through metric boundary crossings, offers a repeatable method for any organization building headless systems without formally defined performance standards.