Summary of an article published on The Web Scraping Club on 2026-06-02 (free to read). Read the full article
Proxy cost looks like a simple product of price per GB and bandwidth. In practice that bandwidth also pays for failed requests, blocks, timeouts and retries. The bill depends more on how the pipeline behaves than on how much traffic it moves. The same dataset can cost far more through an inefficient pipeline, because one failed request can trigger several follow-up attempts.
We go through five mistakes that push costs up: the wrong proxy type for the target, retry loops, misconfigured rotation, no caching or deduplication, and no cost-aware routing. For each one we explain where the budget leaks and what to change. We also argue for judging a setup by cost per successful request instead of price per GB, and we add a note on browser traffic versus plain HTTP requests.
What’s inside
- Why a timeout deserves a retry while a 403 or a block calls for a different reaction, such as rotating the proxy or fixing headers.
- When sticky sessions make sense, when faster rotation is fine for bulk extraction, and why signal-based triggers beat fixed rotation rules.
- How cost-aware routing starts on cheaper proxies and escalates to higher-trust ones only when status codes, latency and success rates call for it.
- How much traffic a full browser load generates through images, scripts, videos and fonts, and when HTTP requests are enough.
In the full article: The cost per successful request formula, the comparison of residential, mobile and datacenter proxies, the recommendations for each mistake in detail and the closing checklist for auditing your pipeline.