Snowflake consulting
Snowflake is straightforward to start with and easy to overspend on. Most of the engagements we’re called into are one of three things: a bill that grew faster than the data, a migration that stalled, or a model nobody trusts.
Cost control
Snowflake bills for compute time, so cost is a function of how long warehouses stay awake and how much they scan. The recurring causes, in rough order of how often we find them:
Warehouses sized for the worst query. One heavy job justifies an X-Large, so everything runs on an X-Large. Splitting workloads across right-sized warehouses is usually the single biggest saving available.
Auto-suspend left at the default. A warehouse idling for minutes after each query, dozens of times a day, across several warehouses. The arithmetic adds up quietly.
Full scans that should be pruned. Clustering keys that don’t match query patterns, or none at all on large tables. Micro-partition pruning is where Snowflake’s performance comes from; if your predicates don’t align with the data layout you pay to read everything.
Dashboards on live queries. BI tools polling raw tables on refresh, keeping a warehouse awake all day to answer questions whose answers barely change. Materialise them.
No cost attribution. Without resource monitors and per-team tagging, nobody owns the bill, so nobody reduces it.
Modelling and dbt
We build in dbt: staging, intermediate and mart layers, tests on the assumptions that matter, documentation generated from the models rather than maintained beside them.
The failure we see most is a warehouse full of transformations nobody can trace to a source. When two dashboards disagree on revenue, the argument is unresolvable, and trust in the whole platform goes with it. Lineage is not documentation overhead. It’s what makes numbers defensible.
Migrations
Coming from Redshift, BigQuery, Synapse or on-premise SQL Server. What we’ve learned to insist on: run both systems in parallel and reconcile outputs before cutting over. Migrations that skip the reconciliation window discover their discrepancies in front of the people who use the numbers.
Data type edge cases, timezone handling and semi-structured data are where the surprises concentrate.
How we engage
Senior data engineers embedded in your team on time and material, starting within 14 days. No fixed scope, no retainer, no minimum term. Rates by role and market are on the daily rates benchmark.
Go deeper
- Data warehouse consulting — architecture across platforms
- Data engineering services — pipelines, orchestration and quality
- Databricks consulting — the lakehouse alternative
- Hire data engineers — embed engineers directly
Want a read on your Snowflake bill?
Bring your warehouse configuration and a month of query history to a free 30-minute call. You’ll get specific, named savings, not a generic optimisation checklist.
staffai.eu · Senior AI and data engineers from Eastern Europe, on T&M