One TTL was costing a publisher 30% of its hits

One TTL was costing a publisher 30% of its hits

A travel publisher in Madrid set a single 1-hour cache lifetime for everything. Hot articles re-fetched needlessly; the homepage served stale prices.

We tiered the TTLs by content volatility:

— Article bodies and old posts: 7 days at the edge, they never change
— Homepage and category pages: 60s edge cache plus stale-while-revalidate 600s
— Price and availability widgets: pulled into a separate fragment with 30s TTL
— Static assets with hashed filenames: 1 year, immutable
— CSS/JS without hashes: fixed the build to add hashes first, then 1 year

The homepage now serves instantly from a 60s cache while revalidating in the background, so readers never wait. Article hit-ratio jumped because 7-day TTL stopped pointless re-fetches. Map volatility before you set numbers — TTL is per-content-type, never global.

The number that mattered: 31% more edge hits from splitting one TTL into five.