Fizz accelerates ecommerce analytics with Databricks SQL

Databricks SQL opens up possibilities for almost everything we want to do. It’s an all-in-one platform with full data intelligence. It’s mostly automatic under the hood so you don’t have to worry – you can just build.— Tamas Bacskai, Head of Data, Fizz.hu

Fizz.hu is a fast-growing ecommerce marketplace backed by OTP Group. Launched just two years ago as part of OTP’s “beyond banking” strategy, Fizz hosts more than 500 merchants offering over 1.5 million active product offers across electronics, household goods, and more.

From the beginning, data was a priority. But the company started with a simple foundation: Microsoft SQL Server and Power BI, running daily batch loads for reporting. As product catalogs expanded and new use cases emerged, that setup began to show its limits.

Fizz needed more than a traditional data warehouse. It needed an all-in-one platform that could support SQL, Python, and future AI initiatives without adding operational complexity. The team found that in Databricks SQL and decided to migrate to a lakehouse architecture built to scale with the business.

A pragmatic migration, delivered in three months

When Tamas Bacskai joined as Head of Data, his mandate was clear: build a data-oriented team and define a scalable path forward. The existing SQL Server environment functioned as a basic warehouse, but Python workloads ran on a separate virtual machine, governance was limited, and scaling meant increasing infrastructure spend.

The team evaluated three options: continue focusing only on warehousing, split advanced workloads to another development team, or adopt a lakehouse architecture that could unify SQL and Python. The lakehouse model “ticked all the boxes,” Bacskai said — including future expansion into machine learning and AI.

Rather than aiming for a perfect redesign, Fizz took an MVP-first approach. With support from an external partner, they migrated approximately 50 tables and several stored procedures, recreating core views in Databricks SQL. The goal was simple: keep reports running, but point them to a new engine.

“It was unorthodox,” Bacskai said. “We didn’t want a perfect migration where everything is rewritten. We wanted to move as fast as possible and refine and modernize after. It’s much easier to do once the data is in Databricks.”

In three months, the legacy SQL Server was switched off completely. Power BI reports continued seamlessly, now powered by Databricks. “It was not impossible, only ambitious,” Bacskai said, “but predictable and achievable.”

Faster reporting and better service levels

The immediate impact was on performance. Previously, daily ETL cycles could take three to four hours, and reporting was not reliably available until 7:00 or 8:00 a.m. That created friction with business users who began their day earlier.

With Databricks SQL, Fizz reduced its end-to-end nightly processing window to roughly 90 minutes. Reports are now consistently ready by 4:30 a.m., even on weekends and holidays. Power BI refresh cycles were cut by roughly 50%, and gigabyte-scale exports now complete in minutes.

The gains were not the result of overprovisioned infrastructure. Fizz runs relatively moderate workloads — about 10 TB total across bronze and silver layers — but the new SQL engine and auto-optimization capabilities delivered measurable improvements without constant tuning.

“It’s not that we just threw more money or bigger clusters at it,” Bacskai clarified. “The SQL execution engine is simply faster. It auto-optimizes and everything is there for us.” 

Equally important, Databricks eliminated the need for separate environments to run Python. All jobs now run natively within the platform, simplifying operations and creating a cleaner foundation for future machine learning initiatives.

Expanding capabilities with AI and self-service

From the outset, Fizz wanted a platform that would not limit its AI ambitions. Even during migration, the team anticipated growing demand for machine learning, generative AI, and more advanced data governance.

Today, Databricks can support SQL, Python, and machine learning workloads in a single environment. The team is exploring masking policies and governance controls to strengthen GDPR and EU AI Act readiness. AI-powered SQL functions will help clean and standardize product names, reducing reliance on complex regular expressions and accelerating data preparation.

Self-service analytics is also expanding through Databricks Genie. Business users can ask natural-language questions, in Hungarian, without writing SQL. About 20 active users rely on Genie today, reclaiming roughly 20% of an analyst’s time previously spent answering ad hoc requests – freeing the team up for more value-add efforts.

“Our Genie set-up is not complete yet,” Bacskai noted, “but it means we don’t have to learn SQL to ask a question. You can just chat with your data.”

For a growing ecommerce company, the value extends beyond speed. Databricks provides a unified, AI-ready foundation that scales with new use cases from marketing data integration to model serving endpoints without requiring a larger team to manage it.

“Databricks SQL was much better than what we anticipated,” Bacskai said. “It’s something we love to work with. It can do everything we want, so we can just build and create what we want.”