Absolutely, caching the per‑user personalization data in a fast store can shave milliseconds off every request. Just key it by user ID, maybe something like user:123:pref, and give it a short TTL so the cache stays fresh. Keep the TTL short enough that when you update a preference you can invalidate or update the key immediately—no stale surprises. Use pipelining or a Lua script to atomically update a user’s score and preferences so you don’t get half‑updated state. Just a heads‑up: keep an eye on memory usage; a naive approach can lead to evictions that slow you down again. I’ll make the data stay sharp, but the marketing brain still has to decide what that data means.