An organisation cites Netflix as justification for adopting microservices. What is missing from the comparison?

What is being tested

Whether you can identify what makes the style survivable, and whether you reason by analogy carefully.

What is missing: the platform

The decomposition is the visible part. What made it work was everything built to support it:

• Client-side service discovery and load balancing, with deliberately stale-but-usable registry data so a registry outage does not become a total outage.
• Standardised resilience libraries, so every caller had timeouts, bulkheads and circuit breakers by default rather than by remembering.
• Deep request tracing, without which a distributed failure cannot be diagnosed.
• Deliberate failure injection, so degradation paths were exercised before customers found them.

Adopting the style without that platform means every team reinvents resilience badly, and nobody can diagnose a cross-service failure.

What is also missing: the driver

The decomposition was not motivated by "microservices are better". It followed a specific, painful event — a database corruption that halted the business for days — which reframed the requirement as no single failure should be able to stop everything.

That is a driver. "We would like to be modern" is not, and the difference determines whether the cost is worth paying.

The reasoning error underneath

Reasoning by analogy from an organisation whose constraints differ. Their scale, team count, existing estate and history are not yours. Ask of any published architecture: what scale, what team size, what constraint drove this, and does that constraint apply here?

What to ask instead

• What specifically is the current architecture preventing? Independent deployment? Independent scaling? Technology choice? Each has a cheaper route than a full decomposition.
• Do we have the operating model? Functional teams plus microservices produces a distributed monolith — every cost, no benefit.
• Do we have a platform team? If not, the platform work lands on delivery teams.
• Does the cognitive load fit? Twelve services across four engineers will be badly operated regardless of design quality.

The likely recommendation

A modular monolith with enforced internal boundaries, plus extraction of the specific components with a forcing reason — a different resource profile, availability class, or compliance boundary.