One definition, or none at all.
Snowflake is where the numbers get settled, if the tables feeding it agree. We build the layer between the systems of record and the board deck, and the contracts that keep it honest as the business changes underneath it.
Everyone has a number. Nobody has the number.
Sales reports one pipeline figure, finance another, the product team a third. All three are defensible inside their own tool. The warehouse is where that gets resolved — or where it gets formalised as three permanent disagreements.
One definition, one number.
SnowflakeWhat we build.
Ingestion you do not babysit
CRM, billing, product and support loaded on a schedule with alerting, schema drift handled, and a failure that pages someone rather than an empty dashboard on Monday.
A modelled layer with definitions
Customer, active, booking, recognised revenue and churn defined once, tested, and documented where the person reading the number can find them.
Reporting the board can open
Dashboards built on the modelled layer rather than on raw tables, so a metric changes in one place and everywhere at once.
Reverse ETL to the operators
Modelled fields pushed back into the CRM, so the segment or health score a rep sees is the one the board saw.
Cost and performance
Warehouse sizing, clustering and query patterns kept in hand, because a warehouse that quietly triples in cost is a warehouse somebody turns off.
Map, define, build, hand over.
Map
Sources, owners and the reported numbers that cannot currently be traced back to a record.
Define
The metrics that matter written as testable definitions, each with an owner and a place a reader can find them.
Build
Ingestion, modelled layer, tests and dashboards, with alerting on the definition and not only on the pipeline.
Hand over
Documentation, the tests, and a period running it beside your team until a full board cycle has been through it.
Definitions drift faster than pipelines break.
Ingestion failures are loud. A definition quietly changing — a new stage, a repackaged plan, a field reused for something else — is silent, and it is what makes last year chart stop matching this year. Tests on the definition are the cheapest thing in the project.
