If there’s one thing I’ve learnt from working with organisations of all sizes, from startups to massive enterprises like bet365, it is that there’s no such thing as a perfect architecture. Every company has its own unique needs and priorities. So, if someone who knows nothing about your business starts dishing out rigid, one-size-fits-all advice, my recommendation is to run for the hills!
With that said, I’d like to share six high-level lessons that I’ve seen work well in practice.
-
Go Live Quickly:
When you are in startup mode, there’s nothing more important than going live. Until you have users interacting with a live service, you’re not truly learning or growing. Any barriers that stop you from getting customers on board whether technical, organisational, or bureaucratic are hindrances you want to avoid or remove. Rule number one: ship early and often.
-
Initially Observability Beats Performance:
Don’t worry too much about perfect performance or scale in the early stages, focus on observability.
Proper metrics, logging, and tracing achieve two major things:
- Product insights: Understand exactly how customers are using (or misusing!) your software.
- User experience insights: Spot usability or performance issues as they happen, so you can address them quickly.
Armed with these insights, you will be able to make data-driven decisions that inform both product direction and technical improvements. This approach has served me well when building my own startups and when transforming large-scale systems with millions of users. -
Strong Interfaces Are Everything:
When designing or evaluating a system, pay attention to interfaces between components. Why? Because well-defined interfaces make it easy to swap out implementations as you learn more from your observability data and as your product evolves.
For instance, in several startups I’ve worked with, we used a serverless approach to iterate quickly and keep operational costs down. However, beyond a certain threshold, serverless can become more expensive than other architectues like Kubernetes. By having clear interfaces between the glue code (e.g. AWS Lambda) and the actual business logic, we were able to migrate easily to Kubernetes when the time was right. We could then shift from a pay-per-invocation model to a more predictable monthly spend.
Remember, no architecture is static. Design for change from day one.
-
Automate Once You Have Traction:
Once your user base starts growing and your product gains traction, automation will save you a world of trouble. It is often easier to define your entire infrastructure as code, rather than manually ticking boxes in some console.
This approach:
- Reduces the likelihood of human error.
- Makes it much simpler to roll out new versions or roll back if something goes wrong.
-
Security Is Non-Negotiable:
Trust is everything in business. The fastest way to lose it is through a security breach or poor reliability.
With that in mind:
- Use managed services whenever possible such as API gateways or identity providers (e.g. AWS Cognito). Security professionals have built and vetted these systems, making them a safer option than rolling your own.
- Focus your efforts where you add the most value. Creating your own authentication flow rarely delivers more value than simply using a secure, off-the-shelf solution.
- Invest in a proper penetration test as soon as it’s practical to do so. Identifying security gaps early will save you a lot of pain down the road.
-
If All Else Fails: Simplicity Wins:
When the design or implementation starts feeling overly complex, step back and ask yourself: What’s the simplest solution that could possibly work? Nine times out of ten, that simple solution is also the best starting point.
Final Thoughts
In summary, my advice is to “move fast with pragmatism” using balanced, data-driven decisions. Focus on shipping early, collecting data, iterating based on that data, and automating where it counts. Pay special attention to security and never be afraid to pivot as your understanding of the problem evolves.
No architecture remains static for long, so build for change and growth from day one. Good luck, and remember: there may be no perfect architecture, but yours will constantly evolve as your business grows!
Need Fractional CTO Support?
If you’d like expert guidance in crafting a roadmap that balances bold innovation with real-world feedback, I offer fractional CTO services at bytesizedcto. My goal is to help organisations of all sizes build tech and product strategies that align with their business goals and grow sustainably.
Contact me if you’d like to discuss your vision.
← Back to Blog Index