You have 14 systems exchanging data through bespoke point-to-point pipelines. Adding a system means building integrations to six others. Propose an alternative.

What the interviewer is testing

Whether you recognise the n² integration topology and know why a log rather than a bus.

The diagnosis

With n systems exchanging data point-to-point you tend towards n² integrations, each with its own format, failure modes, backfill story and owner. Adding a system means building connections to everything it needs; changing a producer means finding every consumer, and nobody has the list.

The integration count is the problem, and improving individual pipelines does not address it.

Why a durable log

Producers write once; consumers read at their own pace and track their own position. The count drops from n² to n.

Three properties do the work:

Durability with retention, so a consumer added later can read history it was not present for. This is the property that most convinces people, because it makes adding a consumer trivial.

Independent consumer offsets, so a slow or failed consumer does not affect producers or other consumers.

Ordering within a partition, which is what makes the log usable for state replication rather than only for notifications.

Why not the alternatives

An enterprise service bus centralises transformation logic, which means the integration logic lives in a component owned by neither producer nor consumer — a shared bottleneck that becomes the thing everyone waits for. The topology improves; the ownership gets worse.

Direct API calls couple availability: every consumer's uptime now depends on every producer's, and there is no history, so a consumer that was down loses data.

What a strong answer adds

Naming what the log does not fix: consumers still depend on the meaning of the events, so semantic coupling remains and needs schema contracts, a registry with a compatibility policy, and published domain events rather than raw internal records.

And the operational obligations that arrive with it: partition and key design, retention sizing, consumer lag monitoring in time rather than records, and the replication slot or offset retention risk where a stalled consumer causes upstream storage to fill.

Common weak answers

A bus with central transformation. Adopting the log without a schema registry, which reproduces the coupling in a new place.