Embedded Files
Data Science project
for client SAUDIA Airlines
Saudia Airlines
Client: SAUDIA AIRLINES
Saudia, formerly known as Saudi Arabian Airlines, is the flag carrier of Saudi Arabia, based in Jeddah. The airline's main hubs are the King Abdulaziz International Airport in Jeddah and the King Khalid International Airport in Riyadh, the latter of which it plans to move out of by 2030.
Worked under the Digital Marketing Director and was responsible of model building in the D2C space (Direct to Consumer) in the Digital channels.
Worked with architecture teams, development teams, marketing and revenue management.
Worked on two separate projects: 1. Building Marketing Signals to manage the digital marketing budget allocation 2. create price predictions for flights with low price competitive advantage.
Presented results to high level stakeholders as the CDO & CTO, Head of HR, Head of Marketing, Infrastructure, Advisor to the General Director & former CTO.
Marketing Budget Allocation in the Digital Channels
Project description
The Commercial Brainfproject represents an advanced artificial intelligence and machine learning initiative designed to revolutionize revenue optimization and digital marketing strategies for Saudi Arabian Airlines Corporation (Saudia).
This comprehensive undertaking focuses on developing and implementing an intelligent decision-making system that automatically optimizes marketing investments, pricing strategies, and promotional offers across multiple digital channels.
The project centers around creating an integrated ecosystem comprising several interconnected components: a Digital Sales Brain that automates marketing decisions through reinforcement learning algorithms, an Offer Brain that dynamically calculates optimal pricing and selects appropriate voucher codes, and sophisticated integration capabilities with major platforms including Google SA360, Skyscanner, WeGo, SkyVizion, and Zytlyn predictive analytics.
The initiative employs a data-driven approach utilizing multiple data sources including load factor signals, competitiveness scores, demand forecasting, competitive pricing intelligence from Infare, and external market indicators from providers like PredictHQ, Semrush, and AccuWeather. The system is designed to operate in real-time production environments, continuously learning and adapting marketing strategies based on performance feedback measured through Net Cash Inflow (NCI) uplift and Return on Advertising Spend (ROAS) metrics.
Role & Responsibility
In this project, the primary role involves Senior Data Engineer/Machine Learning Engineer responsibilities, focusing on the design, development, and implementation of enterprise-scale AI systems for airline revenue optimization. The position requires extensive expertise in reinforcement learning algorithm development, particularly Q-learning implementations for dynamic decision-making in commercial environments.
The role encompasses full-stack data engineering responsibilities, including designing and implementing robust data pipelines for real-time integration with external platforms such as Google SA360, SkyVizion, Zytlyn, and metasearch engines. This involves developing secure API integrations, establishing data format specifications, implementing monitoring systems, and ensuring compliance with enterprise security requirements while maintaining high-frequency data updates across multiple systems.
Algorithm development and optimization forms a core component of the responsibilities, particularly in creating competitive pricing scoring mechanisms that differentiate between routes with and without low-cost carriers, generating route-specific intelligence signals. Additionally, the role involves developing automated pricing calculation systems that assess willingness-to-pay factors across demand trends, behavioral patterns, price competitiveness, and market events, with the capability to dynamically adjust pricing within specified parameters.
Accomplishment
The project achieved significant technical and commercial milestones through the successful implementation of multiple interconnected AI systems. The reinforcement learning framework was successfully deployed in a production environment, demonstrating the ability to continuously optimize marketing investment decisions while achieving measurable improvements in both NCI and ROAS metrics. The system effectively processes multiple data streams in real-time, making autonomous decisions that enhance revenue performance across selected airline routes.
The project successfully established comprehensive measurement and reporting frameworks, implementing sophisticated commercial impact calculation methodologies that track NCI uplift, conversion rates, sales performance, and comparative metrics between test and control groups. The intermittent testing methodology across selected routes provides statistically significant results that demonstrate the effectiveness of AI-driven marketing optimization strategies.
Explanation of libraries
The Commercial Brain project leverages a comprehensive technology stack specifically designed for enterprise-scale machine learning and data processing in airline revenue optimization scenarios. The implementation utilizes several key categories of libraries and frameworks optimized for real-time decision-making and large-scale data processing.
Reinforcement Learning and Machine Learning Frameworks form the core of the system's intelligence capabilities. The project likely employs TensorFlow or PyTorch for implementing Q-learning algorithms, providing the computational foundation for reward-based optimization where agents learn to maximize NCI uplift while minimizing ROAS penalties. Stable-Baselines3 or similar reinforcement learning libraries would be utilized for implementing production-ready RL agents capable of handling continuous action spaces for pricing and marketing budget optimization.
Data Processing and Pipeline Management relies heavily on Apache Spark and PySpark for distributed processing of large-scale airline operational data, enabling real-time processing of load factor signals, competitive pricing data, and demand forecasting across multiple routes simultaneously.
Cloud Computing and Infrastructure utilizes Microsoft Azure services as specified in the project dependencies, likely incorporating Azure Machine Learning for model training and deployment, Azure Data Factory for orchestrating complex data pipelines, and Azure Functions for serverless execution of pricing calculations and voucher selection algorithms. Docker containerization ensures consistent deployment across different environments.
API Integration and Data Exchange employs requests and asyncio libraries for managing multiple concurrent API connections to Google SA360, Skyscanner, WeGo, and other integrated platforms. pandas and numpy provide efficient data manipulation capabilities for processing competitive pricing scores, demand signals, and performance metrics, while SQLAlchemy manages database interactions with SkyVizion and internal data repositories.
Mathematical Optimization and Statistical Analysis leverages scikit-learn for preprocessing and feature engineering of demand signals and competitive intelligence data, scipy for statistical optimization of pricing algorithms, and specialized libraries for time-series forecasting to enhance demand prediction accuracy. The system likely incorporates optuna or similar hyperparameter optimization frameworks to continuously improve model performance based on real-world feedback from marketing campaigns.
Google Sites
Report abuse