Building Enterprise Data Lakes: Consolidating 45 Systems at MGM Resorts

MGM Resorts operates 30+ properties globally with 180K+ employees. Before our data lake initiative, guest data lived in 45 separate systems: PMS, casino management, F&B, spa, retail, loyalty none connected. A guest's lifetime value was impossible to calculate without a week of manual data extraction.

The Architecture

We chose AWS as the primary platform. Source systems fed into AWS Glue for ETL and cataloging, then into an S3 data lake with three zones: raw zone (exact copy of source, never modified the audit trail), curated zone (cleaned, standardized, deduplicated), and serving zone (pre-aggregated for common query patterns). Redshift handled analytics workloads, feeding into Tableau executive dashboards.

The Identity Resolution Problem

The same guest appeared in 12+ systems with different name spellings, email formats, and IDs. Building a Master Data Management (MDM) layer to unify guest identities was the most complex technical challenge. Our probabilistic matching on email + phone + name combinations achieved 94% match accuracy meaning 6% of records were still duplicated. Perfection was impossible; 94% was transformational.

Data Quality Reality

The 35% data discrepancy rate was conservative. Some source systems had date fields stored as text, currency stored without denomination, and customer IDs that overlapped across different source systems. Data quality remediation took 40% of the project timeline twice the original estimate.

The Governance Layer

Without governance, data lakes become data swamps. Every dataset has an owner, description, and quality score. PII data requires justification and approval. Lineage tracking enables tracing any output metric back to its source.

After 18 months, 85% of MGM's strategic decisions referenced data from the lake. That's the measure of success: not the technology, but adoption.