AWS Business Intelligence Blog

Amazon Logistics scales Business Intelligence to over thirty thousand users using Amazon QuickSight

Amazon Logistics (AMZL) is the Amazon last-mile delivery network. The goal of AMZL is to provide customers with a seamless package delivery service across multiple geographies. AMZL plays a critical role in Amazon Transportation’s supply chain by using continuous improvement initiatives and creative thinking to ensure that millions of packages reach their destination as efficiently as possible.

AMZL Data Engineering (DE) is a centralized team of data providers and aggregators who support a wide range of AMZL stakeholders including Business Intelligence (BI), Science, and Program Community. As the center of data excellence, AMZL DE owns, maintains, and supports a reliable and scalable data ecosystem consisting of nine Amazon Redshift databases, two Amazon Relational Database Services (Amazon RDS) instances, and two Amazon QuickSight instances.

Amazon QuickSight reporting and BI infrastructure hosts more than 3,200 dashboards that are consumed by over 30,000 users across the company with average weekly active users of more than 17,500. Critical insights such as AMZL daily and weekly business reviews, operational metrics that help visualize corporate business outcomes and drivers, and operational insights for delivery station managers and AMZL leadership are all hosted in Amazon QuickSight, making it the one-stop shop for AMZL analytics and reporting.

In this blog post, we share the challenges AMZL faced with their previous BI solution, their migration to QuickSight, and the best practices that helped the AMZL team to reduce costs and improve performance to help stakeholders make data-driven decisions.

What challenges was AMZL facing?

AMZL’s user base includes a variety of personas including tech and non-tech users. The previous reporting solution required a comprehensive local setup before you could begin creating any reports or dashboards. With a growing user-base, admins needed to scale servers that needed manual intervention to provision infrastructure. With our previous BI solution, we had limited connectivity for expanding data stores in AMZL. While it could connect to Amazon Redshift, it was difficult to integrate with Amazon Simple Storage Service (Amazon S3) data in a data lake, which is one of the most widely used data sources in last-mile operations. Integrating with broader AWS storage service was either not available or required significant work to employ alternate methods. We started noticing performance bottlenecks when queries were run. All queries wrapped around cursors, which impacted the ability to optimally plan running the queries. To support and scale for data and user base growth each year, AMZL needed compute scalability with predictable and low cost.

Considerations for using QuickSight

AMZL considered QuickSight on the following dimensions:

  • Security: QuickSight provides a secure platform that enables us to distribute dashboards and insights to tens of thousands of users with availability across multiple AWS Regions and built-in redundancy.
  • Scalability: QuickSight can easily scale to handle large volumes of data through elastic scaling and accommodate Last Mile Delivery Technology’s growing reporting and analytical needs without needing to manually provision additional infrastructure as was required with the previous BI solution.
  • Highly performant dashboards: QuickSight SPICE (a superfast, parallel, in-memory, calculation engine) offers consistently fast query performance, decreases query costs by caching and pre-processing the data, and automatically scales to meet high concurrency during holiday peaks and Prime Days. Because Amazon Redshift and S3 are the primary data sources in last-mile operations and tech, having the benefits of SPICE improves metric service-level agreements (SLAs) and the customer experience.
  • Low total cost of ownership (TCO): QuickSight allows usage-based pricing that can scale based on our usage requirements. We don’t need to buy licenses upfront.
  • Out-of-the-box supported data sources: QuickSight seamlessly integrates with Amazon S3, Redshift, Athena, and other data services and simplifies data connectivity, which makes it easier for users to analyze stored data.
  • Usability: The AMZL operations community consists of users who aren’t fluent with emerging reporting and analytics technologies. QuickSight offers a web user interface that that helps users without prior BI experience start building dashboards in a few hours without needing to go through complex set up and permissioning.

AMZL also needed a solution that can scale with reasonable cost to meet our growing business needs in terms of data and users. Today, AMZL has more than 30,000 users accessing over 4,700 dashboards published by more than 820 publishers across North America, Europe, and Asia, and using 44 TB of SPICE capacity.

“Back in 2020 we started moving into QuickSight, and what seemed like a burden quickly became a fluid transition with a smooth migration. Our team of 15+ BIEs were happy with the features provided, making implementation fast and improving data readiness for our customers, currently owning over 150 dashboards. We are up to speed with new enhancements and have delighted our customers with Amazon Q in QuickSight product as well.”

– Cosmin Gheoca, Business Intelligence Manager, EU Amazon Logistics

AMZL solution architecture

Figure 1: AMZL solution architecture

The main components of the AMZL architecture are:

  • Amazon data sources: These are operational authoritative systems of truth for AMZL data from various internal and external systems.
  • Data staging: An intermediate storage layer where data from source systems is extracted, transformed, and curated using Amazon EMR and Redshift.
  • AMZL data lake: AMZL repository for all historical data for advanced analytics and AI and ML use cases. This data is stored using Amazon S3.
  • Analytical database: These are our highly curated data products and datasets that are required for analytics stored in Amazon Redshift.
  • Data marts: We have Region-specific analytical subsets of data in data marts. We have a North America data mart, a Europe data mart, and so on.
  • Transactional databases: AMZL has databases that are used by production systems and provide real time insights. The data is stored using AWS purpose built databases and some of the metadata events in Amazon Redshift.
  • Amazon QuickSight: Our new business intelligence and dashboard solution that helps us derive insights with high performance from all the our data layers including data lakes, data marts, and transactional systems.

Migration timeline

The migration timeline from 2021 through 2024 was:

2021

  • QuickSight launched in AMZL organization
  • Twenty percent of highly used reports migrated into QuickSight from the legacy BI tool
  • Adoption of five thousand users into QuickSight

2022

  • Fifty percent of highly used reports migrated into QuickSight
  • Adoption of fifteen thousand users into QuickSight
  • Previous BI tool deprecated for AMZL delivery station employees

2023

  • All critical analytical dashboards migrated into QuickSight
  • Adoption by thirty thousand last mile users into QuickSight
  • Strategy developed for full deprecation of the previous BI tool

2024

  • Complete migration of the organization into QuickSight
  • Previous BI tool deprecated

Governance practices

Because of the rapid organizational growth of AMZL from a few hundred to tens of thousands of users, managing the volume of dashboards and datasets that were created every day became a significant challenge. This lack of controls on the creation and sharing of dashboards eventually led to the problem of too many dashboards. AMZL users often got lost in the ocean of dashboards that were sent their way and they started losing trust in the perceived sources of truth.

After moving to QuickSight, AMZL was able to implement robust data governance practices that involved closely tracking unauthorized dashboard access, implementing better dashboard publishing and sharing mechanisms, separating development and production environments, and so on. The detailed logging features provided by QuickSight captured metadata on dashboards, datasets, and data sources, including usage patterns. AMZL DE used these logs to create comprehensive datasets in Athena for in-depth analysis that aided in governance efforts. This approach facilitated the development of a robust dashboard governance plan, along with automated alerts to proactively address any potential risks to the plan’s effectiveness.

Improved dashboard publish and access processes: Prior to transitioning to QuickSight, there was a lack of control over the creation and publication of dashboards, as well as access permissions. This resulted in our users, especially the frontline station operators, encountering difficulties in identifying and using the appropriate dashboards for their daily tasks. With the implementation of QuickSight, we established stringent rules regarding dashboard publication and access for these operators. Using QuickSight permissions controls, we set up controlled access policies to ensure that stations were only presented with dashboards that were thoroughly curated for accuracy and need. Additionally, we implemented automated alerts to flag any deviations from these established processes, using QuickSight user activity logs for monitoring and enforcement.

Improved permissions management: QuickSight offers robust dashboard management capabilities, including the organization of dashboards into folders and the implementation of permissions at the folder level. This feature enabled us to establish distinct production and development folders within the environment, which helped streamline workflow management for dashboards. Additionally, the folder-level access control empowered corporate teams to create and publish dashboards within their respective team folders while ensuring adherence to strict access controls. This framework facilitated efficient collaboration across teams while maintaining granular control over data access and governance measures.

Admin utilities for improved developer experience: Building a data model to capture logs from Amazon S3 helped us establish a suite of admin utilities that improved experience of dashboard authors and users. The utilities:

  • Allow tracking usage of dashboards across different user demographics and geographies.
  • Help new users understand the top N dashboards in their organization that they needed to be aware of.
  • Drive cleanup activities on the servers. Unused dashboards are regularly purged to maintain cleanliness and efficiency of our platform.

Paul Codani, Delivery Station Manager at AMZL, emphasizes the pivotal role of Amazon QuickSight in transforming their business operations:

“The migration to Amazon QuickSight has been instrumental for our business. Previously, the lack of features hindered our daily workflows and impeded our ability to swiftly make data-driven decisions on behalf of our customers, products, and overall business. This migration was a collaborative effort, yielding rapid returns.”

Figure 3: Growth of user adoption from 2020 through the first quarter of 2024

Conclusion

Through this migration, AMZL scaled to ten times the original number of users in QuickSight with an 80 percent cost reduction. As AMZL continues its expansion into new territories and markets, driven by the relentless pursuit of innovation and customer-centricity at Amazon, the need for strong BI systems grows. QuickSight fits the bill perfectly. Its adaptability allows us to manage the increasing complexity and diversity of data as AMZL scales up. With QuickSight, we’re not just meeting current demands but also preparing for future challenges.

QuickSight’s insights aren’t just about looking back; they guide our forward strategies, helping AMZL stay ahead in a competitive market. With further integration of Generative AI via Amazon Q in QuickSight, we are confident in having a future proof BI solution that will provide AMZL with the intelligence needed to navigate growth effectively in the fast-paced world of logistics and e-commerce.

About the Authors

Shashank Bangera is a Sr Manager of Data Engineering at Amazon logistics. He heads the world wide data engineering team across North America, Europe and Japan, which plays a key role in providing robust data platforms and building data pipelines essential for Business Intelligence, Science, Operations and Tech. Shashank is all about diving into data, finding patterns, and turning them into actionable insights that fuel innovation and drive success. Outside of work, Shashank indulges in his passions for photography and music. You will either find him exploring the expansive landscapes of Pacific Northwest, or jamming on his electric drums along with his group of friends.

Maitri Shah is a Sr Technical Program Manager at Amazon QuickSight (AWS). She leads cross-functional programs focused on product adoption and drives strategic migrations across Amazon. Maitri is passionate about process improvement and finding innovative ways to solve customer problems. Outside of work, she likes to create intricate and meditative mixed media art form.

Pradeep Misra is a Principal Analytics Solutions Architect at AWS. He works across Amazon to architect and design modern distributed analytics and AI/ML platform solutions. He is passionate about solving customer challenges using data, analytics, and AI/ML. Outside of work, Pradeep likes exploring new places, trying new cuisines, and playing board games with his family. He also likes doing science experiments and watching anime with his daughters.