The workflow before
Colliers runs in more than 70 countries, and its forecasting, reporting and valuation data sat in a legacy enterprise warehouse, regional systems and third-party feeds. Teams consolidated numbers in Excel, and reports took 2 to 5 days. A first-pass property valuation took 3 to 5 days, which capped screening at about 50 deals a week. A full cloud migration was too costly and too risky for systems that still worked.
What we built
We built a data lake that sits beside the legacy systems instead of replacing them. Azure Data Factory pulls from the enterprise warehouse, regional systems and market data into Azure Data Lake in bronze, silver and gold layers, and Snowflake serves valuation, investment, property and financial data marts to Power BI.
On that foundation we built an automated valuation model: an ensemble of gradient boosting, random forest and XGBoost models trained on property characteristics, comparables and alternative data, with a confidence score and the main drivers shown for every estimate.
Azure Purview catalogs the data, tracks lineage and classifies sensitive fields, and models retrain automatically with backtesting by property type and region.
Before and after
| Measure | Before | After |
|---|---|---|
| First-pass valuation | 3 to 5 days | Under 1 hour |
| Valuation error (MAPE) | 12 to 15% | 8 to 10% |
| Deals screened per week | About 50 | About 200 |
| Standard report generation | 2 to 5 days | Under 1 hour |
What's next
The next step is Level 3, Workflow Agents: an agent that runs the deal screening workflow end to end, pulling comparables, running the valuation model and drafting the screening memo, with an analyst reviewing exceptions.
