$40M Strategic Advantage Through ML Forecasting

The Situation

In 2021, Gilead's APAC/LATAM business units were struggling with forecasting accuracy. Manual processes, siloed data sources, and inconsistent methodologies meant:

• Forecasts were off by 12-15% on average
• Senior leadership lacked confidence in budget allocations, affecting QBRs, semi-annual financial reviews, and annual planning cycles
• Cross-departmental projects were nearly impossible to budget when different teams used different forecasting assumptions
• Resource allocation decisions were made on unreliable projections

I was tasked with improving forecasting reliability to support strategic decision-making across regions.

Strategic Approach

1. Root Cause Analysis

• Conducted stakeholder interviews across 6 business units
• Identified pain points: some teams used Excel, others used SQL, some used specific budgetary tools - no consistency
• Teams were using different data sources for the same metrics
• No version control or standardized assumptions

2. ML Analytics Tool Development

• Partnered with data science team to build predictive model incorporating:
  • Historical sales data (5+ years)
  • Market intelligence signals (pricing, competitor launches, regulatory changes)
  • Logistics and supply chain variables
• Designed intuitive interface for non-technical users
• Built scenario planning capabilities (base case and worst case)

3. Change Management & Adoption

• Ran pilot with 2 business units before full rollout
• Created training materials and ran 10+ workshops
• Established feedback loops to continuously improve model accuracy

Quantified Impact

• Improved forecasting accuracy by 10% (from 12-15% error to 2-5% error)
• $40M impact through better resource allocation and inventory management
• Faster decision-making: Forecast generation time reduced from 2 weeks to 3 days
• Gave leadership confidence in budgetary and financial planning - enabling better roadmapping and strategic prioritization for future projects

Key Learnings

1. Technology is only 30% of the solution - The other 70% is change management and adoption
2. Build for users, not data scientists - The best model is useless if business units won't use it
3. Incremental accuracy compounds - 10% improvement sounds small, but at $400M+ revenue scale, it's transformational