In today’s omni-channel retail world, customers are prioritizing experiences that are personalized to their every need. Publicis Groupe empowers brands to transform retail experiences with digital technologies, but data challenges and team silos stood in the way of being able to deliver on the personalization that their customers required. With Databricks, they have enabled their retail clients to convert prospects into customers quickly, and retain them for longer periods of time, resulting in increased campaign revenue by as much as 50%.
As the third-largest communications company in the world, Publicis Groupe deals with massive volumes of data from both internal and external sources. When they began large-scale implementations of their customer data platforms, which contained data on millions of customers and billions of transactions, the limitations became clear: Publicis had to create multiple HDInsight clusters in order to support the automated data pipelines, as well as the analytics environments. Even then, the queries that the analytics users were running would take an incredibly long time to complete because of these scalability issues.
Meanwhile, cross-team collaboration was hampered by tool limitations. Using Jupyter notebooks didn’t allow for the data scientists to easily share and reuse code, for example. Instead, the team had to manually send code back and forth to check and debug.
“Our data processing inefficiencies lead to inconsistency across analytics and data science efforts, impacting our ability to deliver the right message to the right customer at the right time,” said Krish Kuruppath, Senior Vice President at Publicis Groupe.
Once the team implemented Azure Databricks, the unification of data — from ETL to running ML models — changed everything. With native integration with Microsoft Azure and its entire tech stack, Databricks is easy to manage and takes just minutes to get started. It provides the Publicis team with a Lakehouse platform that simplifies infrastructure management, accelerates big data processing, and boosts data team productivity at scale. The data engineers at Publicis were able to scale both memory and write optimized clusters to process over 2.5 billion transactions with ease.
In terms of collaboration, Databricks’s interactive notebooks improved productivity by allowing data scientists and analysts to not only share notebooks, but also do ad-hoc experimentation, train and score models, and operationalize them. At the same time, Delta Lake helped to enable the delivery of consistent data across reads and write, which meant that upsert was easily handled, as was the consolidation of both real-time and batch data.
“Databricks has enabled us to unlock a 360 view of the customer, allowing us to develop models designed to increase customer lifetime value and retention rates for any of our retail customers,” said Sharad Varshney, Vice President of Marketing Machine Learning at Publicis Groupe US
Since implementing Databricks, Publicis has been able to offer its customers a way to significantly improve their revenue by moving to a more real-time activation model. In one of many successes, a Publicis customer saw an increase of 45% in year over year revenue through the use of highly personalized campaigns. In another case, a Publicis customer integrated real-time tracking of customers who had their coupons in order to suppress those who had used them and focus on those who had not. Within a span of three to four days, that particular customer saw a 50% increase in revenue resulting from this campaign.
“Databricks has been key to not only reducing churn of the existing customers of our clients, but increasing revenue driven by recommendations based on their lifetime value, propensity to buy things, and channel affinity,” added Krish.
In addition to being the engine behind their customers’ real-time personalization success, Publicis themselves saw a 5x increase in data processing: with Databricks, they went from 36 hours to 5 hours to process 2.5 billion transactions. They also became more cost-efficient when it came to their Spark clusters. After enabling auto-scale, Publicis was able to reduce the operational cost of running data engineering pipelines by 22% year over year. Furthermore, they lowered customer costs for running ML workloads from upwards of $5000 every 2-3 days to only $800 per month.
And as for cross-team productivity, the ability to actually collaborate has enabled the data science team and data analyst to spin up clusters on-demand, cutting down their development timeline by approximately 30%. With all the extra time and resources saved with Databricks, the team at Publicis is excited to continue strategizing how to best serve their customers well into the future.
Databricks has enabled us to create a single view of the customer and to generate actionable insights that are near-real time.”
– Krish Kuruppath, Senior Vice President at Publicis Groupe, Epsilon