What 2.6 million Ordovia requests taught me about SaaS architecture
The main lesson was not about choosing a cheaper backend. It was about being able to explain why the system talks to the backend at all.
Ordovia reached roughly 2.6 million monthly Supabase requests before it had a user scale that made the number feel natural. The useful question became: which product behaviors were producing that traffic?
A backend bill can be the price of not being able to explain your own system's behavior.
The lessons
- Total request count is not enough; measure per user, session, route and feature.
- Feature-local fetching decisions accumulate into system-wide cost.
- Cloud-first does not mean cloud-every-second.
- Polling should be justified rather than default.
- Cache is a system policy and cost control, not just a speed trick.
- Mutation invalidation should target affected domains.
- Fix application behavior before changing infrastructure vendors.
- Request per active user per day is a useful health signal.
measure -> reduce waste -> fix architecture -> then compare infrastructureAn integrated product does not need to become a tangled system.
SaaS Systems: Architecture, Cost and Simplicity
Field notes and decision frameworks connecting backend cost, cloud choices and product complexity as one systems problem.