A B2B fintech client in the cross-border payments space came to us with a problem that's more common than people admit: their analytics infrastructure was working fine, but the bill was becoming impossible to justify. They were processing $50–70M in daily transactions, querying multiple terabytes of data, and paying $13,000/month for the privilege — with projections showing that number heading toward $35,000 as volumes grew 3x. The stack was entirely AWS managed services: DMS for CDC from PostgreSQL, Glue for ETL, and Athena for querying. It wasn't broken. It was just expensive, and getting more expensive in a way they had no control over.
The deeper issue with fully managed services isn't just cost — it's the absence of control. You can't tune compression. You can't change how storage is laid out. You can't make the kind of low-level decisions that actually move the needle when you're operating at terabyte scale. You're renting someone else's defaults, and those defaults are priced for convenience, not efficiency.
We helped them migrate to a stack built entirely on open-source software: ClickHouse as the analytical database, PeerDB for CDC replication from PostgreSQL, Prefect for orchestration, dbt for transformations, and Apache Superset for dashboards and reporting. Every component in the new stack is something the community has built, audited, and runs in production at scale worldwide. The client now owns their infrastructure, can tune it, and isn't locked into any single vendor's pricing or roadmap.
Monthly spend dropped from $13,000 to $4,000 — roughly $108K in annual savings. But the cost reduction is almost secondary to what they actually gained: operational transparency, the ability to move their stack anywhere, and the confidence that comes from running software that isn't a black box.
In this talk, I'll walk through how we approached the migration — the architectural decisions we made, the places where open-source tooling genuinely surprised us by outperforming the managed alternative, and the honest operational realities of owning your infrastructure instead of renting it. I'll also get into what it took to cut over a live analytics pipeline without disrupting a fintech client's reporting obligations, and what observability looks like when CloudWatch is no longer in the picture.
If you've been eyeing your AWS bill and wondering whether the convenience is still worth it, this talk is for you.
When managed services stop being worth the cost, and how to recognize that inflection point early
What a production-grade open-source analytics stack actually looks like end-to-end — warts and all
How to migrate a live data pipeline without downtime
The operational tradeoffs you're signing up for when you own your infrastructure instead of renting it