USE CASES
Web scraping
Failed collection is rarely a parsing bug. Usually you sent too many requests from one IP. Spreading the exit is the cheapest fix.
Three real reasons you get blocked
Group your failures by status code and you can usually tell which layer stopped you.
- The exit is a datacenter range: the ASN reads as IDC, and many sites downgrade that traffic or serve a challenge
- Too many requests from one IP: you crossed a rate threshold
- The request does not look like a browser: missing Accept-Language, mismatched TLS fingerprint, odd header order
Match rotation to the target
More rotation is not better. Stateless read-only requests are happiest with a fresh IP each time. But once a flow carries pagination cursors, carts or search sessions, switching IPs mid-flow looks like session hijacking and gets cut off sooner.
Stateful flows want sticky sessions: one exit for the whole session, then rotate. Session duration is set in the console.
Where the savings are
Residential traffic is billed by the gigabyte, so the lever is not unit price — it is not downloading bytes you never read.
- Block images, fonts and media; most jobs only need HTML and JSON
- Prefer an API or sitemap over rendering full pages — that is a tenfold difference
- Enable gzip or brotli; response size usually drops to a third
- Cache repeats locally: pagination jobs repeat more than you would guess