A design review presents a new event-driven platform. What cost questions do you ask before approving it?

What the interviewer is testing

Whether cost is part of your architecture review or an afterthought handled by finance later, and whether you know the specific line items that surprise people.

The questions

1. "What is the cost per unit of business value at expected volume?" Not total spend — cost per order, per user, per transaction. This is the only figure comparable to revenue, and it is the one that says whether the design is economically viable at scale. If nobody has computed it, the review should stop here.

2. "Model it at 1×, 10× and 100× volume." Costs in event-driven systems rarely scale linearly. Broker throughput tiers, per-message charges, per-invocation charges and storage retention all have step functions. The cheapest design at launch volume is frequently not the cheapest at scale, and the crossover point is worth knowing before building rather than after.

3. "Where does data cross a zone or region boundary?" The most commonly missed line item, because it does not appear on the diagram. An event-driven platform is a large generator of internal traffic: producers to broker, broker replication across zones, broker to every consumer. Multiply by the replication factor and the consumer count. A chatty design that happens to span zones can produce a transfer bill larger than its compute bill.

4. "What is the retention policy, and what does retained data cost?" "Keep events for 30 days" at some volume is a storage number somebody should have calculated. Infinite retention is a decision, and it should be made deliberately.

5. "What does the observability cost?" An event-driven system generates far more telemetry than a monolith — every hop, every consumer, every retry. This is routinely 20–40% of the platform cost and is almost never in the estimate.

6. "What is idle cost?" Managed brokers, provisioned throughput and reserved capacity cost money at 3 AM when nothing is happening. For a workload with a low duty cycle this can dominate, and it is the argument for serverless messaging.

7. "What is the cost of the non-production environments?" Three environments of the same platform is three times the idle cost. Usually the answer is smaller tiers and scheduled shutdown, but it needs to be an answer.

What I would require before approval

A one-page cost model with per-component estimates, a unit cost at expected volume, the same at 10×, and at least one costed alternative — because a single absolute number is not reviewable, and a comparison is.

What a strong answer adds

Insisting on cost allocation being designed in: tags applied at provisioning so the platform's spend is attributable per team and per feature from day one. Retrofitting allocation onto a running platform is tedious, and without it the next cost investigation starts with 40% of spend unattributed.