84.51° strengthens customer loyalty for Kroger shoppers with Databricks Notebooks
Households served
Decrease in setup and run times
Increase in daily workflows
84.51° provides actionable insights to The Kroger Co. to improve the shopper experience
A wholly owned subsidiary of The Kroger Co., 84.51° is a retail data science, insights and media company that creates more personalized and valuable experiences for shoppers across the path to purchase and beyond. 84.51° utilizes first-party retail data from over 62 million U.S. households sourced through the Kroger Plus loyalty program to build long-term customer engagement and customer satisfaction.
Every day, 84.51°’s Price and Promotions team answers questions such as what products The Kroger Co. should advertise or display in stores. They do this at scale across categories like coffee, produce and meat throughout stores and regions. As the size of the company grew along with the complexity of both transaction data and analyses, they knew they had an opportunity to improve standardization across their teams.
Three teams with fragmented processes came together
Historically, 84.51° used many different solutions to answer requests for clients. They used on-prem solutions, legacy data warehouses, Hadoop and HDFS systems, and homegrown tools to collect data and conduct reporting. These systems required maintenance and became time-consuming and inefficient to support, integrate and upgrade, especially as 84.51° began to explore cloud functionality. This fragmented approach often meant that reports, curated data and the corresponding tools were maintained separately by individual data scientists, which took time away from data science work that contributed to the business. “Our engineering teams were supporting legacy technologies while trying to also learn new cloud tools and capabilities. The business needed to move and scale quickly in the cloud, so much of the infrastructure provisioning fell on data scientists, which prevented them from doing actual data science,” says Chris Kolik, Science and Data Platform Lead at 84.51°.
84.51° gathered a cross-functional team to rethink everything from the ground up. The goal was to apply learnings from across teams in a consistent way so that each team could benefit from the best practices. This would also make it faster for data scientists to get up to speed on new projects, making it easier for 84.51° to cross-pollinate data scientists across projects. The cross-functional team created templates, processes and automation in Databricks to simplify sharing, onboarding and code reuse, and shared the templates across functions and teams.
Unified templates benefited each of the teams
The Standardized Reporting team applied these templates to foster greater collaboration and reduce code maintenance. The Databricks integration with GitHub repositories allowed the team to easily understand what analyses people across the business had already completed, to see what teammates had changed, and to share notebooks with stakeholders. Consistent processes through these templates meant that data scientists could spend more time on data science and less time supporting legacy systems and data pipelines. One data scientist was eager to adopt Databricks when she learned that she could easily connect to Azure DevOps pipelines, reducing the manual setup that had previously been required. “The data scientists could easily start out in a notebook and then could simply share it back and forth for feedback. They can now go from an idea to something in production so much quicker than they could in the past,” says Michael Carrico, Data Science Director at 84.51°.
The Insights and Delivery team applied the development principles established by the Standardized Reporting team to deliver insights faster and improve code reuse and discoverability. For example, one data scientist got a question about which cereal should be promoted. The resulting report from the data was very helpful, so The Kroger Co. wanted to scale that analysis across more products and geographies. Rather than changing hard-coded product lists, she structured her code so she could rerun the same analysis with new parameters with Databricks Workflows and get the answer in minutes. Databricks Repos with GitHub integration tracked all the changes she made so that she or any other team member could view the progression from the original request.
The Promotion Optimization team adapted these templates and practices to their sophisticated codebase and shared their own learnings through the templates to help other teams. They could add complex business logic on a subject and other teams could immediately utilize it. They also leveraged the scale of Databricks to enhance their analyses. Now data scientists can present 2x the number of scenarios to business stakeholders and run them 5x faster. Through reusing these templates, the team aligned more closely with other teams, reduced technical debt and enabled faster onboarding for new data scientists across the teams.
“Using and sharing Databricks Notebooks allowed multiple teams to reuse the same code and understand what other analysis had been done.”
— Erica Gentile, Lead Data Scientist, 84.51°
Data scientists were happier and more effective with Databricks
The team at 84.51° has become more efficient at setting up and running projects with Databricks. Getting a job or workflow stood up now takes minimal effort. Some of their larger jobs took six to eight hours to set up and run. Now they take one hour, and they aren’t competing for resources so they don’t have to queue up. This lets data scientists iterate faster and improve their effectiveness. The standardized templates mean that 84.51° now has less overhead associated with moving data scientists between projects, making it easier to develop them by exposing them to different parts of the business. They also have less technical debt to maintain on established projects.
The standardized templates also improved sharing and collaboration. One of the templates that the Pricing and Promotions team created has been forked and replicated more than a dozen times organically through GitHub. The team can see how much the new templates are in demand from the internal interest to leverage the work they did.
Data scientists also got new visibility into how their recommendations were used downstream by stakeholders. Databricks Workflows allowed the data science team to scale and gave them visibility into what analyses have been done from their work. Their Pricing dashboard gets more than 400 views per month by 50 distinct users, and a Promotions report has been used over 2,800 times in the last six months by 500 users. They can easily see and quantify the impact their team is having on the business. Now everyone has similar tools for working together and measuring their success.
“When we first started, we had a double-digit number of jobs running each month. Now we have thousands of workflows running each day. The ease of use of the Databricks tools for data scientists just lets them go use them.”
— Chris Kolik, Science and Data Platform Lead, 84.51°
