A marketplace launches in a new city. Supply and demand are both near zero. What architectural support does this need?

What the interviewer is testing

Whether you recognise that a business problem imposes architectural requirements, rather than treating launch as a deployment.

The architectural requirements

Per-market configuration, not global constants. Matching radius, pricing parameters, wait-time thresholds, incentive levels and quality thresholds all need different values in a market with twelve suppliers than in one with twelve thousand. If these are constants in code, every market launch is a deployment and every adjustment is an engineering request.

They should be configuration with a defined schema, changeable by the market operations team within guardrails, versioned and auditable.

Models that degrade sensibly with no data. ETA prediction, demand forecasting and ranking all rely on historical data that does not exist. Each needs an explicit fallback: a simple heuristic, a model transferred from a similar market, or a global prior — chosen deliberately rather than producing whatever a model outputs on an empty feature vector.

Metrics that are meaningful at low volume. A conversion rate over forty sessions is noise. Dashboards and alerts calibrated for mature markets will either fire constantly or never, and both erode trust in the instrumentation.

Experimentation that accounts for tiny populations, since standard A/B tests will not reach significance for a long time.

The observation worth making

The system must handle several orders of magnitude of scale simultaneously — a market with a dozen participants and one with hundreds of thousands, running on the same code with the same infrastructure.

That argues for configuration-driven behaviour and against anything that assumes a volume floor: sampling that discards everything at low volume, batch jobs sized for large markets, or alerting thresholds expressed as absolute counts.

What a strong answer adds

Treating launch as a product capability rather than a project. If launching a market is an engineering exercise each time, the expansion rate is bounded by engineering capacity. If it is a configuration exercise with a checklist and self-service tooling, it is bounded by the business.

That reframing is what turns a series of launches into a platform.

Common weak answers

Treating it as a deployment and capacity question. Hard-coding market-specific behaviour with feature flags per city, which accumulates unmanageably.