ETL or ELT, and does the difference actually matter?
It matters operationally more than philosophically. ELT, land the raw data, transform inside the warehouse with dbt, is the modern default: cheaper to build, easier to audit, and re-runnable when logic changes because the raw history is still there. Classic ETL still earns its place when data must be cleaned or anonymised before it may land anywhere. We choose per source, not per ideology.
Do we need real-time data?
Ask what decision changes in the minutes you would save. Fraud checks, stock levels and operational alerting genuinely need streaming; most reporting, finance and planning workloads are perfectly served by batches every fifteen minutes or every night at a fraction of the cost and complexity. A good architecture mixes both, and we will show you the arithmetic behind every choice.
Our dashboards disagree with each other. Can you fix that?
Yes, and the cause is almost always structural: three teams computing revenue three ways from three copies of the data. The fix is one governed source of truth, metrics defined once in the transformation layer rather than in every dashboard, and tests that pin those definitions down. It is unglamorous work with an outsized payoff: meetings stop arguing about whose number is right.
What does a warehouse cost to run, and how do you keep it sane?
The warehouses meter compute and storage, and left alone the compute line grows until finance notices. We keep it boring by design: partitioning and clustering so queries touch less data, incremental models instead of full rebuilds, workload isolation so an analyst's experiment cannot starve production, and per-team cost attribution so consumption has an owner. Most estates we inherit shrink their bill in the first quarter.
Our pipelines exist but break constantly. Rebuild or repair?
Repair first, usually. We stabilise what runs, add tests and alerting so failures are visible and attributable, then replace the worst pieces incrementally while the business keeps its data flowing. A big-bang rebuild is occasionally justified, when the foundation is genuinely unsound, but it is the expensive option and we will say honestly which side of the line your estate sits on.
