Skip to main content
CUSTOMER STORY

Enabling business growth with happier consumers

ABFRL’s single view of customers on Databricks helps drive growth

20x

faster ML serving for markdown marketing models

More

value on less infrastructure spend

Faster

BI reporting

CLOUD: Azure
aditya databricks lockup

“ABFRL is an industry leading fashion retailer in India, with several leading brands and store formats, focused across diverse markets and customer segments. A key element of our digital transformation journey is migrating our data environment to cloud, for which we have selected Databricks as the preferred partner, enabling advanced analytics and AI use cases for future business growth.”

— Praveen Shrikhande, Chief Digital and Information Officer, Aditya Birla Fashion and Retail Ltd.

​​Retail success depends on accurately anticipating the right product mix, sending it to the right location in the right quantities and selling it at the right price. For Aditya Birla Fashion and Retail Ltd. (ABFRL), this means anticipating the needs of millions of customers visiting 4,000 retail locations and delivering a high-quality consumer experience. Looking to provide the scale and performance necessary to deliver insights to regional decision-makers in minutes — not days or weeks — ABFRL realized they needed a data platform that could easily ingest and layer together structured and unstructured data from a myriad of sources. This would allow them to build a single source of truth for all business users and data scientists to better forecast trends, demand and pricing. The platform needed to do all this efficiently so data teams could be nimble and maintain budget discipline even as they explored exciting technologies like generative AI and LLMs. ABFRL selected Databricks as the data, analytics and AI platform that allows them to do more with less infrastructure. Data and analytics have become a driver for increasing sales and profitability. And that never goes out of style.    

Powering a retail revolution through modernization

As a retail leader in one of the largest global economies, it’s critical for ABFRL to ensure customer satisfaction through superior experiences across the customer journey. This starts with understanding customers better —  in a way that gives ABFRL the insights to improve customer service and personalize shopping experiences. One area of focus was to mitigate the impact of brand detractors by serving them targeted call center communications designed to address their concerns within a 48-hour SLA. For instance, by identifying segments of detractors, ABFRL can leverage CRM and sales data to pinpoint opportunities to bring detractors back to the brand. Given that this is a highly complex task requiring significant amounts of data and advanced machine learning techniques, ABFRL knew it was time to migrate to the cloud. But as they surveyed the landscape of cloud-based solutions, they saw many data silo and inflexibility issues. “We have to frequently replicate data from the data warehouse for analytics or AI. What we needed was a unified, single source of truth that could support business users, data scientists, machine learning or whatever new use cases came up in the future that we can’t anticipate today,” said Srinivas Gopinath, Head of Data and Analytics at ABFRL. 

Gopinath added, “The need of our business is timely insights. Tier one data is supposed to be available for business leaders to analyze before 8:00 AM, but took us much longer. Reporting queries would take minutes — it needed to be in seconds.”

Of bigger concern for the team was the inability to support data science use cases. Machine learning data had to be manually copied from data system to data system, leading to even more data silos (with no clear data lineage). This led to mismatches between what data scientists were looking at when building predictive models and what the business was seeing in BI and analytics. As the volume of data and use cases kept increasing, the leadership at ABFRL realized they had to take a whole new approach to data management, governance and collaboration — one that was beyond what their current data stack could provide. They needed something that not only met today’s challenges, but would be agile and adaptable enough for future, unforeseen AI uses. 

Future-proofing retail to deliver profitable insights

As an industry leader, ABFRL is undertaking the development of a modern data platform ahead of its time. To facilitate the acceleration of innovation without limits, ABFRL chose the Databricks Data Intelligence Platform as the foundation for their data, analytics and AI efforts. Given its large legacy ecosystem and its ability to unify 20 to 30 different data sources, Databricks was the best choice to further establish ABFRL as an industry pioneer by integrating engineering, data warehousing and ML under the centralized lakehouse architecture.

“Databricks has helped us to build a modern lakehouse that is future-proof,” said Sandeep Gutal, Lead Data Engineer at ABFRL. “We now have improved data analytics and ML capabilities, improved collaboration between teams, and have the required security and governance controls in place to scale.”

To minimize disruption of their existing data warehouse, which supports the entire BI and analytics ecosystem, they were able to securely replicate the data model and the analytical ecosystem through Delta tables, Databricks SQL and Unity Catalog. With the Delta Live Tables framework, they were able to simplify ETL processes by minimizing the amount of code to be converted and migrated.

“ABFRL was able to successfully leverage lakehouse framework and implement some of the latest features that were made available, thus making it one of the most modern Databricks Platform implementations,” said Gopinath.

Transforming retail strategies that drive market share with AI

Core business functions like determining markdowns, making in-store recommendations, segmenting detractors and performing market basket analysis are all happening on the Databricks Platform. Through the platform, ABFRL has reduced model development times from three to four months previously to half that time, allowing them to scale ML across various use cases, from demand forecasting to store-level SKU assortment and allocation. Another target use case is active merchandising, which involves reviewing how products are performing in the market. What used to be done at the end of a three-month season — using manual spreadsheets that were laborious and error-prone — can now be done on a weekly, biweekly or monthly basis using the markdown recommendation ML model built in the Databricks Platform. Teams at the store can now quickly make in-season discounts, leading to higher revenues on lesser discounts. From an operational standpoint, the ABFRL’s data team has taken full advantage of the platform’s automation — delivering these new use cases with fewer resources.

Additionally, ABFRL has been able to implement further efficiencies in their supply chain through better collaboration with internal and external partners. Insights that used to take six months are now available to review in two to three weeks with the help of Databricks — empowering ABFRL to more accurately adjust production based on actual consumer purchasing trends and deliver popular styles to stores while they’re still in season. 

The last use case ABFRL focused on was improving detractor retention. The team segmented consumer data via k-means clustering by layering internal and external data sources within the Databricks Platform to identify worrisome consumer behavior and develop more impactful retention campaigns. ABFRL was then able to automate personalized outreach with customized offerings to match the segmentation.

“The Databricks Data Intelligence Platform has enabled the business to react more quickly to consumer changes through the delivery of more granular data insights across different areas of the business,” said Gopinath. “For instance, we can now personalize communications to target brand detractors and customized offerings to improve retention rates.”

The adoption of Databricks is now far-reaching within the organization. For example, they are also using LLMs to enrich product descriptions in their product catalog. Not only are they able to create thousands of product descriptions in just a few hours compared to days, but they also found the AI-generated product descriptions to be of better quality compared to the content created by individual contributors. Perhaps the most forward-thinking use is happening around sales attribution by product features. The team has done a deep dive on every single SKU to understand how different product features and attributes impact the rate of sale by location and demographics, which was enabled at scale only by using the Databricks Data Intelligence Platform.

What most excites the team is the ability to explore cutting-edge AI in ways that weren’t possible before. The team has already implemented natural language processing for sentiment analysis and topic classification to better understand what trends their consumers are talking about and then pass those insights along to the call center team. “With Databricks, we can enhance our understanding of customer preferences, streamline content categorization and understand consumer expectation, which will help in optimizing decision-making processes,” explained Gopinath. “With ML-powered topic and sentiment classification, our teams can now focus more on strategic initiatives and the development of game-changing business hypotheses.” This tailored way of looking at voice of the customer (VOC) data in a more structured format will also allow their team to come up with personalized marketing strategies, leading to higher customer retention and impact analysis. “This ability to quickly sift through massive amounts of structured and unstructured data to surface insights that generate sales has transformed the role of the team,” concluded Gopinath. “The platform provides the unified tool for data engineers and ML engineers to collaborate on a single business outcome in a highly productive fashion.”