← All articlesArchitecture and quality

System architecture: the decision that costs the most when made by accident

Architecture is chosen at the start of a project, when you know the least. If the choice is wrong, it shows a year or two later, when the company grows and the system can no longer keep up. Here is how I approach this decision so you do not pay for it later.

The most expensive mistake I have seen

A startup built its product as one large system, a monolith. At the start it was a quick and cheap solution. After about two years the system stopped scaling and every new feature took longer and longer. The company needed to grow faster than its technology allowed.

The rebuild cost several hundred thousand złoty. On top of that came the time in which the team, instead of developing the product, was moving over what already worked. Most of that cost could have been avoided with a few decisions at the beginning.

Questions I ask before choosing an architecture

  • What should the system look like in two to three years? Not a detailed plan, just the scale: how many users, how many locations, how many integrations.
  • Which parts will change most often? These should be developed and deployed independently of the rest.
  • What must the system integrate with? Partner systems, field devices, payments. Every integration is a place where something can break independently of us.
  • What must always work, and what can wait a moment? This decides where high availability is needed and where a simpler solution is enough.
  • What can be taken off the shelf? Authentication, notifications or payments rarely need to be written from scratch.

An architecture that grows with the company

These answers usually lead to splitting the system into independent services with clearly defined responsibilities. Each can be developed, deployed and scaled separately, and a new feature does not require touching the whole system. I build most projects this way, from urban traffic management to car wash networks controlled from a single platform.

That does not mean the first version has to be elaborate. It may have only a few services. What matters is that the boundaries between them are drawn from the start. Then the system grows by adding new elements rather than by being rebuilt.

What a good architectural decision should deliver

  • a document with diagrams that more people than its author can understand,
  • the reasoning behind key choices, so that a year later it is clear why something was done one way and not another,
  • a list of risks and a plan for what to do if they materialise,
  • a delivery plan split into stages, with a working system after each one.

Let’s talk about your project

Tell us briefly what you want to build or fix.