Pricing optimization for a national supermarket

📈 The Business Challenge

Promotions represent nearly 50% of unit sales and 40% of revenue in grocery retail, yet many chains rely on repeating the same promotions year after year — often applying identical prices across vastly different stores. This blunt approach ignores local demand elasticity, leading to margin loss, supply chain inefficiencies, and missed growth opportunities.

One of Canada’s largest grocery retailers, sought to modernize its promotional pricing strategy by integrating advanced AI and optimization models. The goal: create store-level price differentiation and optimize promotion plans to drive revenue and margin improvements while reducing waste across 1,000+ stores.

🧩 Our Approach

As technical lead on the project, I oversaw the design and development of a modular AI-driven solution capable of:

• Clustering stores based on historical sales, price sensitivity and demographics
• Recommending optimal price points per cluster for unadvertised promotions
• Scaling to generate full quarter promotional schedules while respecting business constraints

Our hybrid approach combined machine learning, mathematical optimization, and simulation:

• Demand prediction models forecasted store-level sales at various price points
• Mixed-Integer Non-Linear Programming (MINLP) formulations optimized both pricing and store clustering
• Custom simulators validated financial and operational impacts before deployment

🚀 What We Delivered

• A production-grade pipeline ingesting 24 billion+ transactional records
• Store-level price elasticity models across 1,000+ stores
• Clustering algorithms tuned to balance model accuracy with interpretability for merchant teams
• A scalable optimization engine proposing price points and promotion schedules
• User interface integration enabling merchants to review, adjust, and approve AI-driven plans
• Frameworks for A/B testing and iterative model improvement

🎯 The Impact

• $85k CAD / store / year in free cash flow impact
• 4.6% gross margin improvement on unadvertised promotions
• Reduced the “bullwhip effect” by improving upstream supply chain visibility
• Obtained buy-in from leadership to further shift towards AI-assisted promotion planning — improving decision quality and freeing merchant time for strategic tasks

👥 Team & Collaboration

• Cross-functional team of
  • data scientists
  • mathematicians and optimization experts
  • retail domain specialists
  • data engineers
• Joint work with retailer’s supply chain, merchandising, and data science leaders

🛠 Tech & Tools

Tech stack:

• Google Cloud BigQuery
• Google Cloud Vertex AI
• Google Cloud Composer (Apache Airflow)
• Google Cloud DataProc (Apache Spark)

Science stack:

• Mixed-Integer Programming
• Demand Forecasting
• Hierarchical Clustering

💡 Key Takeaways

This project demonstrated the power of blending AI, mathematical optimization, and human-in-the-loop decision-making in a high-stakes retail environment. By moving beyond static pricing models, we delivered a solution that not only improved financial outcomes but also unlocked new capabilities for merchant teams — laying the foundation for scalable, data-driven retail planning.