What it is
A system design review is a structured, independent assessment of a software system’s architecture. It looks at how the system is built, how its parts interact and how it behaves under load and failure, then compares that with what the business needs from it. The review can cover a system already in production or a design that has not yet been built.
When you need it
- Usage is about to grow sharply and you are unsure where the system will break.
- Incidents keep recurring and fixes do not seem to stick.
- A rewrite or major investment is being proposed and you want an objective view first.
- A new architecture has been drafted and you want it checked before development starts.
- You have acquired or inherited a system and need to understand its condition.
How we approach it
We start by agreeing on the questions the review must answer and the quality attributes that matter most, such as scalability, availability, security, maintainability or cost. Priorities differ: a payments platform weighs correctness and security above all, while an internal reporting tool may care most about maintainability.
Next we gather evidence. We interview the engineers who build and run the system, read architecture documents if they exist, inspect the code structure and infrastructure definitions, and review incident history, monitoring dashboards and cost reports. From this we draw current-state diagrams, which are often useful on their own.
The analysis walks through critical user journeys and asks what happens when each dependency is slow, down or wrong. We check for single points of failure, unbounded queues, missing timeouts, chatty data access, weak trust boundaries and tight coupling that makes change risky.
Findings are written up with the evidence behind them and ranked by impact and effort. Quick wins sit beside larger structural changes, each with a clear recommendation.
Common pitfalls to avoid
- Reviewing diagrams only. Diagrams often describe intent, not reality. The code and infrastructure tell the true story.
- Unranked findings. A long list of issues with no priority leads to paralysis.
- Ignoring team capacity. Recommendations must fit the people who will carry them out.
- Treating the review as an audit. The goal is improvement, so engineers should feel safe sharing problems.