$11.4M+ annual revenue boost through personalized recommendations - Impetus

$11.4M+ annual revenue boost through personalized recommendations

How F200 Airlines transformed travel choices with precision machine learning

01

Business needs

One of the major American airlines wanted to enhance customer engagement, drive revenue growth, and offer real-time recommendations for destinations and ancillary product sales. They wanted a tailored solution that would leverage customers’ travel history and in addition needed to:

  • Elevate customer engagement with personalized offers through web and app interfaces 
  • Provide real-time, personalized destination recommendations 
  • Optimize the check-in process with relevant ancillary products 
  • Ensure the solution remains pertinent to evolving travel trends

The airlines wanted data-backed insights for improved conversion rates and lasting customer relationships.

02

Solution

The Impetus team leveraged AWS to unlock several ML-driven use cases and address the business needs of the airlines. The solution involved two phases – 

  • Feature engineering, which involves extracting features from the raw data to make it suitable for analysis and recommendations by machine learning models. 
  • Model training and inference to generate relevant recommendations and actionable insights. 

A high-level architecture diagram of the solution is given below:

Architecture diagram

Feature engineering

The Impetus team utilized PySpark on Amazon EKS to extract, transform, and load data from diverse sources into the AWS SageMaker Feature Store, rendering raw data suitable for machine learning models. Amazon EKS and ArgoCD were leveraged for seamless pipeline orchestration of complex workflows.

To achieve low latency, an in-memory database employing Amazon ElastiCache for Redis functions as an online feature store for real-time inference, optimizing MLOps architecture. Feature store management was further enhanced through compaction, ensuring optimal storage utilization.

Processed 2.6 million requests per day each under 60 milliseconds latency

Model training and inference

ML models, powered by a versatile SDK, efficiently generated recommendations. The API/GRPC solution ensured streamlined predictions, actively deployed across AWS regions for high availability and disaster recovery. Utilizing Redis for optimized query times, it achieved a rate of 2.6 million requests per day, each with latency under 60 milliseconds.

The solution was integrated with Datadog for active monitoring. The system captures logs, enabling swift issue response. The Spark-Kafka data pipeline added a layer of meticulous recording for thorough analysis and continuous enhancement of model predictions.

Elevated user engagement: 1.6 million clicks post enhancement of the home page carousel

03

Impact

The AWS-based ML solution facilitated seamless onboarding of use cases while ensuring data accuracy and quality for downstream applications. It played a pivotal role in enhancing the passenger experience, resulting in remarkable achievements:

  • The home page carousel, post-enhancements, witnessed a substantial increase in user engagement. A surge of 1.6 million clicks within personalized impressions, highlighting elevated customer interaction. 
  • Personalized destination recommendations inspired customers to explore new destinations, contributing to an annualized incremental revenue of $11.4M+. 
  • Content personalization, tailored to past preferences, incentivized customers, and empowered them to make informed choices. 

Choose a lab aligned to your Data & AI journey

Address your desired use case across critical analytic dimensions

  • Explore architecture options with experts

  • Ensure strategic alignment of business and technology

  • Architect an ideal solution for a pressing problem


  • Validate new or refactor existing architecture

  • Develop a prototype with expert guidance

  • Establish a roadmap to production


Learn more about how our work can support your enterprise