Do I need real-time recommendations for a feed or is batch enough?

It depends on what “good enough” means for your feed and how fast your content/user behavior changes.

Batch is usually enough if:

• Your feed doesn’t need to react instantly to clicks, likes, follows, or purchases
• Freshness can be measured in hours, not seconds
• You’re optimizing for simpler architecture, lower cost, and easier debugging
• Your ranking model mainly uses slower-changing signals like long-term interests, popularity, or editorial rules

Real-time recommendations are worth it if:

• User intent changes quickly and you want the feed to reflect that immediately
• Session behavior matters a lot, like “show more of what they just engaged with”
• Content inventory changes rapidly and stale results are a bad experience
• You need to react to breaking news, live events, trends, or marketplace availability
• You have strong evidence that latency directly improves engagement or conversion

A common middle ground

Many teams do:

• Batch candidate generation / long-term personalization
• Real-time re-ranking or feature updates for recent interactions, trends, inventory, or session context

This often gives most of the benefit of real-time without full complexity.

Rule of thumb

• Start with batch if you’re early-stage or unsure
• Add near-real-time or hybrid only when you can point to a specific user pain or business metric that batch can’t meet

If you want, I can help you decide based on your product type, update frequency, and latency requirements.