When we reach for each tool
Situation
Why it works
Situation
Why it works
Legacy platforms we migrate to Snowflake
Teradata
Oracle
Cloudera / Hadoop
Amazon Redshift
Cassandra
Architecture patterns that combine Snowflake, ClickHouse and enterprise platforms
Snowflake + SAP
ClickHouse + Kafka
Snowflake + ServiceNow
Datadog + ClickHouse
Independent in what we recommend. Selective in what we master
AWS
Role:
The client's cloud environment, where we build.
Best for:
— When you are already there and the architecture supports it — Workload placement aligns with existing infrastructure
Watch out for:
Cross-region data transfer and services left running after the work that required them has ended
Azure
Role:
The client's cloud environment, often alongside the Microsoft ecosystem.
Best for:
— The organization is embedded in Microsoft and has no reason to change — Coexistence with Microsoft stack is a priority
Watch out for:
Resources scaled up manually and never scaled back down
GCP
Role:
The client's cloud environment, often alongside BigQuery.
Best for:
— Analytics already live there and moving them would not create a measurable benefit — BigQuery-native workloads that perform well at current cost
Watch out for:
Queries that scan full tables where partitioning or clustering would reduce work
Scaleway
Role:
European cloud hosting for cost-efficient compute and localized data storage.
Best for:
— European workload placement and predictable billing matter more than a broad catalog of specialist managed services — Data residency requirements in the EU
Watch out for:
Architectural friction where workloads depend on specialized, non-standard cloud primitives
Apache Kafka
Role:
Event bus and streaming backbone between source systems and analytical engines.
Best for:
— Event streams and real-time messaging — Telemetry and sensor data ingestion — Workloads that need low-latency data movement
Watch out for:
Infrastructure complexity and event retention that outlives its operational value
dbt
Role:
Transformation layer and controlled home for shared metric definitions.
Best for:
— Conflicting numbers across departments — Tests and documentation needed alongside transformations — Shared metric definitions in one place
Watch out for:
Full model rebuilds where incremental processing would be sufficient
Apache Airflow
Role:
Orchestration, scheduling, and dependency management for complex data pipelines.
Best for:
— Complex multi-system dependency graphs — Workflows that native schedulers cannot manage clearly — Cross-platform pipeline coordination
Watch out for:
Heavy scheduler environments maintained for simple periodic work
dlt
Role:
Lightweight data loading in Python without connector subscriptions.
Best for:
— Custom API ingestion where maintained code is cheaper than enterprise connectors — Rapid prototyping of data pipelines
Watch out for:
Broken pipelines after unmonitored upstream API changes
Datadog
Role:
Infrastructure and pipeline observability, alerting, and data-health monitoring.
Best for:
— Pipeline failures need detection before stakeholders notice — Infrastructure monitoring for data platforms — Early warning systems for data quality
Watch out for:
High-cardinality metrics and unfiltered logs that increase SaaS costs
Power BI
Role:
Presentation layer that is usually retained while the underlying data platform improves.
Best for:
— The client already uses it; we improve the models and pipelines beneath it so reports remain responsive — Report rationalization and consistent numbers
Watch out for:
Business logic duplicated in reports and licenses growing with recipients
Tableau
Role:
Presentation layer, often in organizations with an established analytics culture.
Best for:
— The client retains Tableau and needs an agreed data model beneath it — Shared metric definitions across the organization
Watch out for:
Pushing business logic into extensions and calculated fields, which becomes difficult to maintain or audit
Looker
Role:
Presentation layer with its own semantic model.
Best for:
— Metric definitions should sit close to reporting and have one clear owner — Shared metric definitions and consistent numbers
Watch out for:
Unoptimized nested queries generated repeatedly for minor report interactions
Sigma
Role:
Presentation layer for teams that work primarily in spreadsheets.
Best for:
— Report users would otherwise export analysis to Excel — Reducing manual work and ensuring consistent numbers
Watch out for:
Large numbers of uncached warehouse queries triggered by routine filters and column changes
