Purpose
Win Rate Optimization is an accelerator designed to assessment the win rate for deals, for a price for specific item and customer and all additional parameters like product availability, delivery time… Based on this win rate assessment, the model can recommend a price that would increase expected revenue and or profit.
Outputs
The main outputs of the win rate model are an assessment of the win rate for a specific set of parameters and recommended prices.
-
Win Rate Prediction: For each quote item, a predicted win rate at the current and optimized prices.
-
Feature Attribution: Identification and quantification of key factors affecting win rate
-
Optimal Price Recommendations: Price points that maximize expected revenue/profit
Approach
The win rate assessment relies on a boosted tree machine learning model to predict the win rate and define the impact of each parameter of the model. Those impacts can be reviewed in details and provides valuable insights about the sales and customer behaviors to identify and quantity key factors affecting win rate, like how much does discount % impact chances of winning.
Key steps in the approach:
-
Data Preparation: Collect quote data and store them in a data source (either DS or DM). This should be available up front. It's vital to map wins (e.g., 'Ordered', 'Delivered') and losses ('Expired', 'Project Lost'), and have clear loss reasons recorded for each quote item and exclude duplicate cases like variants and option in Quotes, to not bias the win rate assessment.
-
Feature Selection: Identify variables that may influence win rates. These include:
-
Price, Discount %, Margin %. Those will be computed directly by the model by providing the proper list price, revenue and profit.
-
Customer segment, classification, type…
-
Product category, Delivery time…
-
-
Model Training & Testing:
-
Data is split into training and test periods (commonly, the last 20% is used for testing).
-
Train the machine learning model to predict the win rate probability for a deal per quote item, using selected features.
-
Assess model performance by examining feature importance and the impact of parameters on win rate.
-
-
Optimization Logic: Using the predictive model, simulate win rates across different price points to estimate expected revenue and profit:
-
Expected revenue = price × win rate
-
Expected profit = (price - cost) × win rate
-
Choose the price that maximizes revenue and/or profit expectancy, according to business strategy (revenue, profit, or a mix).
-
Limitations
-
Data Quality & Price Changes: The effectiveness of the approach relies on the quality, completeness, and volume of historical quote and transaction data, especially proper win/loss labeling and loss reasons. Sparse data, especially for certain discount levels, can undermine model robustness. All lack of price changes and different price points will undermine the ability of the model to learn from the customer behavior and recommend proper optimized price.
-
Integration in Quotes is additional configuration not part of this accelerator, which scope is limited to provide recommendations through a “model evaluation” that should be used for integration.
-
No predefined extension point – There is no out-of-the-box extension point defined for now. If you intend to add specific features, custom code should be written. (But then the accelerator becomes specific and it cannot be updated without extra effort to port those modifications.) Do not hesitate to report specific requirements and possible extension points to Pricefx.
-
Data requirements – See Data Requirements (Optimization - Win Rate)