Data Platforms · 01 of 04 · Snowflake

Elastic compute. Decoupled storage. Predictable bill.

Snowflake's killer idea was simple: separate compute from storage so each scales on its own clock. We build warehouses that exploit that idea ruthlessly — multi-cluster for concurrency, sized for the workload, governed at the column.

Plan a Snowflake sprint ↗What we build
120+Warehouses tuned
−62%Avg. credit-cost reduction
SelectServices Partner
What we build on Snowflake

A warehouse that scales. A bill that doesn't.

We treat Snowflake credits like any other production budget: instrumented, alarmed, owned by a person.

Account & Warehouse Design

Org → account → role hierarchy designed for clarity. Per-workload warehouses sized to query mix, not vibes.

Org · Account · WH sizing

ELT with Snowpipe & Streams

Continuous Snowpipe ingestion, change streams for incrementality, dynamic tables where they earn it. Replayable, idempotent, observable.

Snowpipe · Streams · Tasks · Dynamic Tables

Snowpark & ML

Python in the warehouse for feature engineering and model inference. Cortex for the LLM use cases that live next to your tables.

Snowpark · Cortex · ML

Governance & PII

Column-level masking, row-access policies, classification, native tag-based propagation. Audit-grade access without bottlenecking analysts.

Masking · Row access · Tags · Horizon

Iceberg & Open Sharing

Iceberg tables for the workloads that need open formats. Native Snowflake sharing for partners. Both, designed without duplicating the lake.

Iceberg · Secure Sharing

Cost Guardrails

Resource monitors, query timeouts, warehouse auto-suspend, scheduled review of the top-10 cost queries. The bill stops being a quarterly surprise.

Resource monitors · Query timeouts · Tagging
Reference Account Layout

Roles, warehouses, schemas — designed once.

A clean account is the cheapest performance win in the warehouse. We start there.

ETY · SNOWFLAKE REFERENCE · ACCOUNT: acme.eu-west-1ROLESACCOUNTADMINSECURITYADMINSYSADMINENG_WRITERBI_READERDS_WRITERPII_VIEWERWAREHOUSESWH_BIXS · mc 1-4auto-suspend 60sWH_ELTL · sc 1on scheduleWH_DSXL · on-demandSnowparkWH_ADHOCXS · 5min capmonitor: $50/dDATABASESRAW_DBsnowpipe landSTG_DBdbt stagingMART_DBdbt martsSHARE_DBpartner reachML_DBfeature store

A boring account is a cheap account.

Most Snowflake bills are paying for sprawl — overlapping warehouses, the same dashboard hitting four different schemas, ad-hoc queries from the largest cluster the user can grab. We collapse the sprawl on day one.

  • 1
    Functional roles, not user-named

    Roles map to job-to-be-done. People come and go; the role hierarchy stays.

  • 2
    One warehouse per workload class

    BI, ELT, Data Science, ad-hoc — each on the warehouse sized for it, monitored independently.

  • 3
    Resource monitors on every warehouse

    Hard caps and notifications. Cost surprises become cost conversations.

  • 4
    Dynamic tables where they fit

    Declarative pipelines without orchestrating tasks by hand. Refresh policy a deliberate decision.

The Snowflake surface area

Features we've shipped, not just slides.

Snowflake ships a lot. We've put each of these in production and have an opinion.

Compute

Standard WHMulti-clusterSnowpark ContainersCortex

Ingestion

SnowpipeSnowpipe StreamingStreams & TasksDynamic Tables

Tables & Storage

Native tablesIceberg tablesHybrid tablesExternal tables

Programming

Snowpark PythonSnowpark ScalaSnowpark JavaJS UDFs

Governance

Horizon CatalogMaskingRow accessObject tagging

Sharing & Apps

Secure SharingMarketplaceNative AppsStreamlit

Transformation

dbtSQLMeshCoalesce

FinOps

Resource MonitorsQuery AccelerationCost Insights
Recent Snowflake work

The bill goes down. The dashboards get faster.

Three quick takes.

B2B SaaS · 2,400 customer tenants

Slashed Snowflake credits by 62% in a quarter.

Right-sized warehouses, multi-cluster only where concurrency demanded, resource monitors on every workload, query timeout discipline.

−62%Credits
+0.3sP95 query (faster)
WH designMonitorsdbt
Adtech · streaming events

Snowpipe Streaming · 1.2B events/day.

Streaming ingestion with sub-minute latency, dynamic tables for incrementality, dashboards always within 90 seconds of reality.

1.2BEvents / day
<90sFreshness
Snowpipe StreamingDynamic Tables
Healthcare data network

Native Sharing for 38 partner hospitals.

Secure Sharing for de-identified analytical sets, row-access policies per partner, full audit. Zero data movement, full control.

38Partners live
0Data copies
Secure SharingRow accessTags
You might also like
Databricks

The lakehouse — when ML and ETL deserve to live next door.

→
dbt

SQL transformations as software. The discipline layer on top of Snowflake.

→

Snowflake without the surprise.

30 minutes. Bring last month's credit usage; we'll point to the top three places the bill is leaking — for free.

Book a discovery call ↗All Data Platforms