Swire Coca-Cola, USA data science team identifies millions in savings with Snowflake, accelerating route optimization logistics
Swire Coca-Cola, USA is the local bottler for Coca-Cola and other beverage brands in 13 states across the American West. Every day, the company’s 7,200 associates deliver refreshments to 31 million consumers between the Mexican and Canadian borders and from the Pacific Ocean to the Great Plains.
Just as many customers have done along their data journeys, Swire originally had Snowflake as its single source of truth and a separately managed Spark platform for their AI/ML needs. In this blog, we’ll explain why the Snowpark feature addresses the associated challenges and allows Swire to begin consolidating AI/ML efforts onto the Snowflake Data Cloud.
Decreasing Total Cost of Ownership
Unlike Snowflake’s simple administration, with its separation of compute and storage, the managed Spark platform introduced multiple billing processes that were difficult to manage. Data was moved out of Snowflake for spark workloads, feature engineering, and model training. Results were sent back to Snowflake for inference as the presentation layer. The data movement incurred ingress and egress charges or what many call “double taxation”. Coupled with additional cloud provider costs from the managed Spark platform, Swire’s previous architecture made it extremely difficult to accurately track costs.
In addition, managing complex infrastructure diverted data teams’ valuable time and resources from building models that could drive business impact. These bottlenecks lengthened their time-to-value. For example, Spark clusters required manual maintenance to mitigate wasteful spending and spinning up clusters took 10–15 minutes. Perhaps most critically, the managed Spark platform outside of the secure Snowflake ecosystem presented data governance challenges, raising concerns about data integrity and security.
These initial challenges underscored the need for a more streamlined and efficient approach to data management, prompting Swire to explore avenues for improvement in its data infrastructure and governance processes. Now with Snowpark, Snowpark ML, and many other AI/ML features in Snowflake GA, Swire’s data science team was able to complete all their current tasks within Snowflake without the added cost of another product.
Snowpark unlocks AI/ML efforts within the Snowflake ecosystem
Recognizing the challenges posed by their initial data management infrastructure, Swire strategically migrated existing use cases to Snowflake to stress test and provide a like for like comparison. Snowflake emerged as the ideal one-stop-shop for their AI/ML needs, offering a singular platform that significantly reduced complexity, enhanced ease of use, and provided a robust framework for improved data governance. The allure of Snowflake extended to its impressive time-to-value proposition, eliminating the previously required cluster-management and facilitating faster access to critical data with zero movement. The ease of use of managing Snowflake warehouses, along with their near instant auto-resume, further streamlined operations, eradicating the need for manual intervention in cluster management and effectively curtailing wasteful spending.
For heavier workloads, Snowpark-optimized warehouses provided more than enough memory and compute, allowing for complex and memory intensive AI/ML projects to be completed.
The seamless transition was facilitated by the inherent advantages of Snowpark and Snowpark ML, which exhibited a syntactic similarity to Spark and SparkML, ensuring an easy learning curve for the data teams. Swire’s data scientists have had a positive experience incorporating Snowpark for their network optimization. They’ve shared that being on the same platform greatly streamlines their processes, making tasks faster and more efficient.
Accelerated model deployment by weeks
One of the most significant achievements was the optimization of planned logistics routes, resulting in significant cost savings. This encompassed reductions in costs related to fuel, driver expenses, and overall cost to serve. The impact on time-to-market was equally remarkable, with models developed on Snowflake showcasing a notable acceleration by not having to face the challenges noted above. This enhanced efficiency and expedited the deployment of critical models, saving multiple weeks, which translated into tangible business advantages.
In summary, Snowflake not only addressed the initial challenges faced by Swire, but delivered tangible, quantifiable benefits, positioning the organization for a more agile, cost-effective, and data-optimized future. This new future will support the company’s vision of unleashing its potential by performing as a leading technology company that delivers refreshment, anytime anywhere. With Snowflake’s innovation in AI/ML features such as Snowpark ML, we’re excited to extend Snowflake’s capabilities as Swire’s data science platform.
