Freshness Requirement
How stale data may be before the decision it supports degrades — the only question that justifies streaming over batch.
Streaming is materially more expensive than batch: always-on compute, stateful operators to reason about, harder correctness, more difficult reprocessing, and a smaller pool of engineers who can operate it. That cost is worth paying when freshness genuinely changes an outcome.
The test is to name the decision and what changes. Fraud declines a transaction in milliseconds — streaming, unarguably. Inventory availability shown to a shopper changes whether they buy — minutes matter. A weekly executive report does not improve by being five seconds fresh, and a monthly regulatory return improves not at all.
Two answers are common and both mean batch. "Real-time" as an unexamined requirement usually survives one question about what decision it supports. And "the business asked for real-time dashboards" usually means they want the dashboard to be current when opened, which micro-batch at five-minute intervals serves at a fraction of the cost and complexity.
The honest position is that a large proportion of streaming platforms are serving requirements that hourly batch would meet, and the cost is paid in operational burden rather than in licence fees, which makes it invisible in the business case.