The workflow before#
Robbins Brothers sells engagement rings and fine jewelry from 15 showrooms, and a ring sale is a high-value, high-touch purchase. Its reporting ran on DecisionCast, an on-premise warehouse near end of life, with financials in Oracle EBS and customer data in Salesforce. Daily reports were Excel workbooks that staff rebuilt by hand from extracts. The monthly close took more than 10 days. Sales reports took days to produce, and store managers couldn't compare their showroom with others or see inventory across locations. A new dashboard took weeks. Marketing, finance and store teams each worked from their own version of the numbers.
What we built#
We replaced the aging warehouse with a cloud data stack on Azure and Snowflake, and gave every team one shared set of numbers. Sales, inventory, customer, finance and headcount data now live in modeled marts with common definitions. Executives, finance and store managers work from Power BI dashboards that update on schedule instead of spreadsheets someone rebuilt that morning.
Pipelines in place of Excel extracts
Azure Data Factory pulls Oracle EBS, Salesforce and headcount data into Azure Data Lake Storage, and Airflow handles scheduling, dependencies and retries. The manual extract-and-rebuild routine is gone. When a source runs late, the pipeline waits and retries instead of a person chasing it.
One semantic layer in Snowflake
We modeled sales, inventory, customer, finance, headcount and product marts on a shared semantic layer. Every report uses the same definition of a sale or a margin, so finance and the stores stop arguing about whose number is right. That shared layer is what cut the close from 10+ days to 5.
Dashboards for stores and executives
Power BI gives executives a company view, store managers a showroom performance dashboard with inventory across locations, and business users self-service reporting. A sales report that took days now takes minutes, and a new dashboard takes days to build instead of weeks.
How we delivered it
We worked across three engagements: strategy, the core build, then expansion. The highest-value sources moved first, so finance and store teams were on live dashboards before the later sources arrived. Each phase retired a piece of DecisionCast, and the old system came down only after its replacement was in daily use.
Before and after#
| Measure | Before | After |
|---|---|---|
| Monthly financial close | 10+ days | 5 days |
| Sales report turnaround | Days | Minutes |
| Inventory view across showrooms | Weekly, manual | On demand |
| New dashboard build | Weeks | Days |
What's next#
Sales, inventory and customer data now live in one governed Snowflake model. That sets up the next layer: an analytics assistant that store and finance teams can ask questions in plain language, and demand forecasting for the roughly 40% of bridal business that is special order, where better forecasts mean less capital tied up in stock.
