Price Optimization

Approach

M&A's strategy was divided into three interconnected phases, building progressively from historical analysis to real time testing and portfolio level recommendations.

Phase 1: Price Elasticity Model

• Objective: Build a machine learning model to understand how price changes historically impacted units sold and GP.

• Steps:

  • Mine historical sales data.

  • Develop and fine tune a model to predict units at various price points.

  • Output: CSV tables, PowerPoint curves (e.g., demand elasticity for top SKUs, and inferred margins.)

• Value: Enabled initial insights, such as optimal pricing for inelastic vs. elastic demand segments. Limitations due to sparse data were addressed in subsequent phases.

Phase 2: Signal Testing

• Objective: Implement A/B/n testing to generate more data points and refine the Phase 1 model.

• Steps:

  • Create testing infrastructure with guardrails defined by our client’s leadership.

  • Run experiments incorporating signals like time of year and competitor prices.

  • Build a feedback loop: Analyze results statistically and feed back into the ML model for iterative improvements.

• Value: Created a continuous optimization framework, revealing deeper relationships (e.g., seasonal impacts on GP).

• Output: Enhanced data files for our client’s pricing tools, with examples showing multiple price tests, yielding GP outcomes "X" to "C".

This phase ensured tests matched user journeys, minimizing bias and maximizing statistical validity.

Phase 3: Portfolio Wide Price Optimization

• Objective: Model inter product effects and recommend prices to maximize total GP.

• Steps:

  • Develop a substitution model to measure cannibalization across SKUs.

  • Create a portfolio profit model assessing multi-SKU price changes.

  • Deliver recommendations, including final A/B tests comparing original vs. optimized prices.

• Value: Holistic view preventing isolated optimizations from harming overall performance.

• Output: Detailed CSV/PowerPoint reports explaining relationships.

The overall approach emphasized iteration: modeling, experimentation, and analysis to refine prices dynamically.