Skip to content

SYSTEM Cited by 1 source

Procore Analytics

Procore Analytics is Procore's customer-facing data product. It gives construction owners, general contractors, and specialty contractors programmatic access to their connected Procore dataset inside their own tools: Power BI, Tableau, Snowflake, BigQuery, their own Databricks workspace, or on-prem SQL Server. (Source: sources/2026-10-29-databricks-how-procore-is-powering-construction-analytics-with-opensharing)

Generation 1 — API extraction into SQL Server

  • Data was extracted from Procore's public APIs and loaded into an Azure SQL Server database that customers queried.
  • Refresh: one run per day; paying customers got at most three per day.
  • At tens of thousands of customers, the deep recursive API calls were expensive, brittle, and prone to timeouts and limits. Refresh frequency was the biggest product limit.

Generation 2 — multi-region medallion + Delta Sharing

  • Regional medallion. Bronze (raw app DB + integration events), silver (cleaning, validation, data-quality checks), and gold (aggregated, modeled) run in multiple regions near the source data.
  • Central sharing region. Gold tables from each region go to one primary analytics region. OpenSharing / Delta Sharing fans them out from there.
  • Scale. Hundreds of curated tables per customer, partitioned per customer; hundreds of thousands of distinct shares (concepts/tenant-isolation).
  • Two recipient paths. An open-protocol token generated in the Procore app (BI tools, other warehouses through the open clients), or Databricks-to-Databricks sharing keyed by the customer's workspace ID.
  • Entitlement reconciliation. A scheduled job detects new shares and permission changes, reads user permissions from the Procore app, and reconciles share grants in Unity Catalog. The result is per-user, per-project access with no manual provisioning (concepts/fine-grained-authorization).
  • Granularity and incrementality. Shares at schema, view, table, or partition level. Delta Change Data Feed lets recipients pull only the changes since their last sync (concepts/change-data-capture).
  • One semantic layer. The same gold layer serves in-app reporting and the extract, so names and values match across products. Two parallel pipelines became one.
  • Change contract. 4 breaking-change releases/year (6-week notice) and 12 non-breaking releases/year (3-week notice) (patterns/data-contract, concepts/schema-evolution).
  • Org shape. One shared pipeline, then data-domain teams (for example, finance) that own their data end to end (concepts/data-mesh).

Outcomes (self-reported)

  • Catalog grew to hundreds of tables in less than half the time Gen 1 took.
  • The daily refresh ceiling is gone; many customers refresh several times a day.
  • New tool features become extractable almost as soon as they ship.

Not disclosed

How gold moves between regions, how often the entitlement job runs, how it bounds revocation lag, and any latency, volume, or cost numbers.

Seen in

Last updated · 854 distilled / 2,448 read