advanced 2 min answer

A business sponsor asks for a real-time data platform because "the competition has one". Reporting is currently a nightly batch that lands at 06:00 and nobody has complained. How do you handle this?

streamingbatchrequirementscost
Show the full answer Hide the answer

Do not answer the technology question

"Real-time platform" is a solution, and it has arrived without a problem attached. Answering it directly leads either to an expensive build with no measurable benefit or to a refusal that reads as obstruction. Both are avoidable.

Ask what decision changes

For each candidate use case, two questions: what decision does this data support, and what would be done differently if it were fresher?

The answers sort quickly:

Use case Decision Freshness that changes it
Fraud screening Decline this transaction Milliseconds — genuinely streaming
Inventory shown to shopper Whether they buy Minutes
Stock replenishment Today's purchase order Hourly is ample
Executive weekly review Steering Nightly is ample
Regulatory return Nothing Monthly

Usually one or two rows justify streaming and the rest do not. That is a far more useful artifact than a yes or a no, because it makes the scope small and defensible rather than making the whole platform contested.

The most common real requirement

"Real-time dashboards" very often means current when opened, not sub-second. A micro-batch on a five-minute cadence serves that at a fraction of the cost and with none of the operational burden. It is worth testing this explicitly, because it resolves a large share of these requests.

Be honest about the cost

Streaming's expense is not licences. It is always-on compute, stateful operators that must be reasoned about, harder correctness (watermarks, late data, exactly-once), harder reprocessing, and a much smaller pool of engineers who can operate it at three in the morning. None of that appears in a vendor comparison, which is why streaming platforms are frequently approved on a business case that omits their main cost.

What to propose

  1. Build the one case that justifies it — fraud screening — as a genuine streaming path, and let it prove the capability.
  2. Move the batch to micro-batch where freshness has any value. Cheap, low risk, and it addresses most of the perceived gap.
  3. Design so the choice is reversible. Land events into a log regardless. Batch consumers read from it in windows, streaming consumers read continuously. Adding a streaming use case later then does not require a new platform.

If overruled

Accept it and shape it. Ask for the first use case to be one with a measurable outcome, a defined success metric, and a review at six months that includes the operational cost. A streaming platform delivered with one real use case and honest running costs is a reasonable outcome; one delivered as a platform in search of users is the failure mode to avoid.