AI-Driven Accuracy: Enhancing Pos Malaysia’s Delivery ETA with Minfy & AWS

Project Overview:

‍Pos Malaysia Berhad, the national postal and parcel service provider in Malaysia, has a history dating back to the early 1800s. With a vast network of over 1,000 touchpoints, including post offices, automated machines, and mobile outlets, Pos Malaysia plays a crucial role in the country's logistics and supply chain sector.

As e-commerce demand surged, the ability to provide accurate Estimated Time of Arrival (ETA) for parcel deliveries became a critical need. However, their existing system, built on a legacy SQL database and Power BI, struggled to deliver reliable predictions due to inconsistent data, scalability limitations, and a lack of monitoring. These inefficiencies led to inaccurate delivery estimates, negatively impacting customer experience and operational effectiveness.

To address these challenges, Pos Malaysia partnered with Minfy, an AWS Partner, to modernize its ETA prediction system. The goal was to develop a scalable and accurate real-time ETA prediction system leveraging AWS cloud services, enabling improved delivery accuracy, enhanced customer engagement, and increased operational efficiency.

‍

Challenges

‍Pos Malaysia's existing ETA prediction system suffered from several issues that affected its performance and reliability:

• Inconsistent Data Sources: Data resided in multiple disparate sources, including databases, emails, and ad-hoc files, leading to inconsistencies and making it difficult to generate accurate insights.
• Scalability Issues: The system struggled to handle large volumes of data, resulting in prediction failures during peak demand periods.
• Lack of Monitoring: Without robust monitoring capabilities, the system failed to proactively detect issues affecting ETA accuracy, leading to delays in troubleshooting.
• Limited Insights & Predictive Capabilities: The reliance on static reports from Power BI limited real-time insights and data-driven decision-making.
• Impact on Customer Experience: Inaccurate ETAs led to customer dissatisfaction, as Pos Malaysia couldn't provide reliable delivery updates.

These shortcomings hindered Pos Malaysia’s ability to streamline logistics operations, engage proactively with customers, and optimize its service efficiency.

‍

Solution

‍Minfy designed and implemented a modernized ETA prediction system on AWS using a hybrid data processing architecture to enhance real-time prediction accuracy and operational efficiency. The key components of the solution included:

‍

Data Ingestion & Processing

• Implemented a hybrid data processing approach combining both batch (EMR) and streaming (Kinesis and Lambda) pipelines to handle real-time and historical data requirements.
• Raw data was collected from multiple sources, including IoT-enabled delivery tracking, warehouse systems, and customer service logs, and ingested into an AWS S3-based data lake.
• AWS Glue ETL was used to process structured and unstructured data, ensuring seamless data transformation and integration.

‍

Data Storage & Governance

• Data was stored in a centralized AWS S3 data lake for scalable storage and efficient processing.
• Amazon Redshift was used for analytical processing, enabling complex queries and trend analysis on parcel movements and delivery times.
• AWS Data Lake Service ensured data quality, security, and compliance throughout the entire lifecycle.

‍

Advanced Machine Learning for ETA Prediction

• Amazon SageMaker was leveraged to build, train, and deploy AI-powered ETA prediction models, which continuously improved through real-time learning and model retraining.
• The ML models were optimized for inference, reducing latency and computational costs while enhancing prediction accuracy by 37%.
• The system dynamically adjusted delivery ETAs based on live traffic, weather conditions, and operational factors.

‍

Data Analytics & Visualization

• Amazon QuickSight dashboards provided business users and logistics teams with real-time visualization of KPIs such as ETA accuracy, delivery success rates, and bottleneck trends.
• Custom alerts and reports enabled proactive decision-making by identifying potential delays in real time.

‍

Real-Time Monitoring & Issue Resolution

• Amazon CloudWatch was integrated for continuous monitoring of data ingestion pipelines, ML model performance, and system health.
• Automated anomaly detection was implemented to flag irregular patterns, enabling early intervention in case of performance degradation.

‍

Outcome

• Improved ETA Accuracy: The ML-enhanced prediction system delivered a 37% improvement in ETA accuracy, significantly boosting customer satisfaction.
• Operational Efficiency: The scalable architecture improved data handling during spikes, resulting in a 70% improvement in operational efficiency.
• Proactive Customer Engagement: Pos Malaysia was able to provide customers with accurate and timely delivery estimates, enhancing overall experience.
• Actionable Insights: The solution enabled data-driven decision-making through intuitive dashboards and enhanced analytical capabilities.