End-to-End Machine Learning Framework for Conversion and Pricing Optimization

This repository provides a reusable framework for solving pricing optimization problems where the goal is to balance the twin objectives of winning deals (conversion probability) by offering discounts and maximizing profitability. It demonstrates a complete workflow including:

• Feature engineering at both item and quote levels (numeric, categorical, and temporal features)
• Predictive modeling to estimate quote conversion likelihood.
• Optimization strategies (brute force, Bayesian optimization, and multi-objective genetic algorithms) to recommend markups that maximize expected profit while maintaining strong win rates.

Business Problem

Sales team generates quotes that often contain multiple products (quote items). Each item has a base cost, and sales representatives have the discretion to apply a markup to determine the final selling price. This flexibility is a double-edged sword:

● High Markups: Increase profit margin per item but decrease the likelihood of the customer accepting the quote (lower conversion rate).

● Low Markups: Increase the chance of winning the deal but erode overall profitability.

Dataset Schema

Column Name	Data Type	Description
quote_id	String	Unique identifier for a sales quote
item_id	String	Unique identifier for an item within a quote
supplier_unit_price	Float	The base supplier unit price (cost)
quantity	Integer	Quantity of this item
product_category	String	Category of the product (e.g., Compute, Storage, Networking)
quote_type	String	Type of quote (e.g., normal, QTO)
region	String	Geographic region of the customer
markup	Float	Markup percentage applied (e.g., 0.2 for 20%)
quote_converted	Integer	Target variable: 1 if quote accepted, 0 if lost
quote_publish_date	Timestamp	When the quote was published

Repository Structure

• quote_features.py
  Python module for feature engineering and processing quote data.

• quote_feature_builder.pkl
  Serialized Python object (pickle) for the quote feature builder.

• calibrated_model.pkl
  Pickled calibrated machine learning model.

• data1_imputed.pkl
  Input dataset with imputed values.

• ConversionModel.ipynb
  Jupyter notebook for developing and analyzing the conversion model.

• RecommendationModel.ipynb
  Jupyter notebook for building and evaluating the markup recommendation model.

• environment.yml
  Conda environment file listing dependencies required to run the notebooks and scripts.

Getting Started

1. Clone the repository:

   git clone <repo-url>
   cd <repo-directory>

2. Set up the environment:

   conda env create -f environment.yml
   conda activate <environment-name>

3. Run the notebooks:
   Open ConversionModel.ipynb or RecommendationModel.ipynb in JupyterLab or Jupyter Notebook.

Usage

• Use quote_features.py to process and engineer features from quote data.
• Load the provided .pkl files for pre-trained models and datasets.
• Refer to the notebooks for model training, evaluation, and analysis workflows.

Requirements

• Python (version specified in environment.yml)
• Conda (for environment management)
• Jupyter Notebook/Lab

License

MIT license.

Contact

For questions or contributions, please open an issue or pull request.