Data warehouse

A data warehouse solves that problem by serving as a central collection point. Data from every relevant source is imported regularly, converted into one uniform model and stored permanently. On that consolidated basis, analysis becomes possible that individual tools could never deliver.

The decisive difference from an ordinary database lies in the purpose. An operational database serves day-to-day running and has to process many small writes quickly. A data warehouse is optimised for the opposite: reading and aggregating vast amounts of data over long periods, as reporting and analysis typically require.

The data's route into the warehouse follows a process traditionally called ETL: extract, transform, load. Data is first extracted from the source systems, then transformed, that is, cleaned and unified, and finally loaded into the warehouse. In modern cloud architectures the ELT variant has often prevailed, in which the raw data is loaded first and transformed only in the warehouse.

Specialised tools are often used to connect the sources. Services such as Fivetran or Airbyte handle automatic transfer from hundreds of systems, while tools such as dbt organise the transformation of the data within the warehouse. That creates a repeatable, documented data pipeline.

Clean modelling matters. Data from different sources has to be structured so metrics are unambiguously defined and the same terms mean the same thing everywhere. Without that shared data model, contradictory figures arise and undermine trust in the whole reporting.

The market is shaped by a few large cloud platforms differing in architecture, pricing model and ecosystem. Google BigQuery meshes closely with the Google marketing stack and is particularly popular for analysing raw Google Analytics 4 data. Snowflake counts as platform-neutral and flexibly scalable. Amazon Redshift is deeply anchored in the AWS ecosystem, and Microsoft Fabric bundles several data services into one platform.

The overview below classifies the four solutions by the most important decision criteria. For GDPR compliance, choosing an EU region matters in particular.

As soon as personal data flows together in a data warehouse, the GDPR's strict requirements apply. The choice of storage location is decisive: if an EU region is used, the data stays within the European legal area, which avoids transfers to third countries. All the platforms named offer EU regions, but the configuration has to be set deliberately.

Data minimisation also plays a central role. Not all raw data has to be stored unchanged. Pseudonymisation, aggregation and defined deletion periods reduce the risk and meet the principle of storage limitation. A well-considered permissions concept also ensures only authorised people access sensitive data.

At Elisabit we design data warehouse setups to be technically solid and privacy-compliant at the same time. Data from first-party sources can thus be brought together sensibly without first-party data becoming a legal risk.

For smaller companies with manageable data volumes, a full data warehouse can at first be oversized. Thanks to serverless cloud models and usage-based billing, though, the barriers to entry have fallen: you largely pay only for the data and queries actually processed, which makes starting out economical for mid-sized companies too.