Pabis Retail

Optimizing scalability and performance

Realizing that it needed much greater query performance than a traditional row-based database could offer, Pabis Retail searched for an advanced columnar storage solution. Impressed by a customer reference for Vertica, the company selected the Vertica Analytics Platform as its new data warehouse.

Paulo Rios, IT manager at Pabis Retail, commented, “Of the data warehouse solutions we reviewed, Vertica stood out for its exceptional performance and parallel processing capabilities, which would enable upwards of 1,000 users to run complex queries concurrently. Also, we noted the Vertica platform would require minimal maintenance, saving us valuable time.”

Deploying Vertica in Eon Mode running on Amazon Web Services is ideal with Pabis Retail’s high transaction volumes. It provides the highest levels of scalability to tackle the most demanding client workloads. The architecture combines a central data repository with separate storage nodes, making it quicker and easier to add extra processing resources when demand grows, as Rios confirmed, “Eon Mode has been a gamechanger. Several major clients run massive queries at the same time each month, and the elastic scalability of Vertica in Eon Mode ensures we have the resources to meet their requirements.”

Introducing an AI-driven effective price simulator

As the team familiarized themselves with the AI-driven opportunities presented by Vertica, more use cases emerged to support Pabis Retail clients. Many clients want to understand how price changes can affect sales and overall revenue. To receive detailed insight, they would need to create interactive what-if dashboards where price changes are plotted against demand with a fine level of detailed elasticity analysis. However, to analyze this effectively for each product in each store, a two-year history of daily sales and price data needs to be searched. This is represented by millions of local data models that must be aggregated by week, filtered for extreme values, and pre-processed.

With the data volumes involved, even Vertica struggled to meet the requirements associated with creating a price simulator based on elastic demand. The data was batch-processed and published into Vertica, but the team found the execution times to be prohibitively long.

Putting machine learning to work with VerticaPy

VerticaPy is the perfect blend of the scalability of Vertica and the flexibility of Python, bringing a unique and indispensable set of data science tools. It contains a Python library that is used for machine learning and data science tasks in conjunction with the Vertica database. It provides a Pythonic interface to interact with and leverage the power of Vertica for analytical and predictive modelling tasks.

Leveraging VerticaPy, the team could effortlessly build models to perform the complex calculations. These are optimized for efficiency and performance using many of the InDatabase scalable machine learning algorithms. Python scripts call SQL queries, and the results are directly published in Vertica.

Carenne Ludeňa, director data science for Pabis Retail, commented, “We started using VerticaPy to solve our massive elasticity model generation for retail pricing, a process that previously took hours to complete. Using VerticaPy, powered by machine learning and AI-driven, we now receive results in almost real-time without any delay. We are very excited about the results and the huge possibilities Vertica provides.”