Engineering

Real-time vs batch: what actually matters

Not every metric needs sub-second latency. Here is how to decide what to stream and what to aggregate overnight.

Jonas WeberHead of EngineeringMar 5, 20266 min read

Real-time is not always better. It is more expensive, harder to debug, and often unnecessary for decisions that happen weekly, not hourly.

Stream what drives action

Signup funnels, error spikes, and checkout drop-offs benefit from immediate visibility. If someone can fix the problem in the next hour, stream the event.

Batch what informs strategy

Cohort retention, LTV modeling, and quarterly planning rarely need live updates. Nightly rollups keep costs down and queries fast without sacrificing insight.

The best analytics stacks mix both. dumbledor gives you streaming by default, with rollups you can tune per event type.