Databricks SQL opens up potentialities for nearly every part we wish to do. It’s an all-in-one platform with full information intelligence. It’s largely automated below the hood so that you don’t have to fret – you may simply construct.— Tamas Bacskai, Head of Knowledge, Fizz.hu
Fizz.hu is a fast-growing ecommerce market backed by OTP Group. Launched simply two years in the past as a part of OTP’s “past banking” technique, Fizz hosts greater than 500 retailers providing over 1.5 million energetic product provides throughout electronics, family items, and extra.
From the start, information was a precedence. However the firm began with a easy basis: Microsoft SQL Server and Energy BI, operating each day batch masses for reporting. As product catalogs expanded and new use circumstances emerged, that setup started to point out its limits.
Fizz wanted greater than a conventional information warehouse. It wanted an all-in-one platform that would assist SQL, Python, and future AI initiatives with out including operational complexity. The staff discovered that in Databricks SQL and determined emigrate to a lakehouse structure constructed to scale with the enterprise.
A realistic migration, delivered in three months
When Tamas Bacskai joined as Head of Knowledge, his mandate was clear: construct a data-oriented staff and outline a scalable path ahead. The present SQL Server surroundings functioned as a fundamental warehouse, however Python workloads ran on a separate digital machine, governance was restricted, and scaling meant rising infrastructure spend.
The staff evaluated three choices: proceed focusing solely on warehousing, break up superior workloads to a different growth staff, or undertake a lakehouse structure that would unify SQL and Python. The lakehouse mannequin “ticked all of the containers,” Bacskai stated — together with future enlargement into machine studying and AI.
Moderately than aiming for an ideal redesign, Fizz took an MVP-first method. With assist from an exterior associate, they migrated roughly 50 tables and a number of other saved procedures, recreating core views in Databricks SQL. The objective was easy: hold studies operating, however level them to a brand new engine.
“It was unorthodox,” Bacskai stated. “We didn’t need an ideal migration the place every part is rewritten. We needed to maneuver as quick as potential and refine and modernize after. It’s a lot simpler to do as soon as the information is in Databricks.”
In three months, the legacy SQL Server was switched off fully. Energy BI studies continued seamlessly, now powered by Databricks. “It was not unimaginable, solely formidable,” Bacskai stated, “however predictable and achievable.”
Quicker reporting and higher service ranges
The fast affect was on efficiency. Beforehand, each day ETL cycles may take three to 4 hours, and reporting was not reliably out there till 7:00 or 8:00 a.m. That created friction with enterprise customers who started their day earlier.
With Databricks SQL, Fizz decreased its end-to-end nightly processing window to roughly 90 minutes. Stories at the moment are persistently prepared by 4:30 a.m., even on weekends and holidays. Energy BI refresh cycles had been reduce by roughly 50%, and gigabyte-scale exports now full in minutes.
The positive factors weren’t the results of overprovisioned infrastructure. Fizz runs comparatively average workloads — about 10 TB complete throughout bronze and silver layers — however the brand new SQL engine and auto-optimization capabilities delivered measurable enhancements with out fixed tuning.
“It’s not that we simply threw extra money or greater clusters at it,” Bacskai clarified. “The SQL execution engine is solely quicker. It auto-optimizes and every part is there for us.”
Equally necessary, Databricks eradicated the necessity for separate environments to run Python. All jobs now run natively inside the platform, simplifying operations and making a cleaner basis for future machine studying initiatives.
Increasing capabilities with AI and self-service
From the outset, Fizz needed a platform that may not restrict its AI ambitions. Even throughout migration, the staff anticipated rising demand for machine studying, generative AI, and extra superior information governance.
At this time, Databricks can assist SQL, Python, and machine studying workloads in a single surroundings. The staff is exploring masking insurance policies and governance controls to strengthen GDPR and EU AI Act readiness. AI-powered SQL features will assist clear and standardize product names, decreasing reliance on advanced common expressions and accelerating information preparation.
Self-service analytics can be increasing by way of Databricks Genie. Enterprise customers can ask natural-language questions, in Hungarian, with out writing SQL. About 20 energetic customers depend on Genie right this moment, reclaiming roughly 20% of an analyst’s time beforehand spent answering advert hoc requests – liberating the staff up for extra value-add efforts.
“Our Genie set-up will not be full but,” Bacskai famous, “nevertheless it means we don’t should study SQL to ask a query. You may simply chat together with your information.”
For a rising ecommerce firm, the worth extends past pace. Databricks gives a unified, AI-ready basis that scales with new use circumstances from advertising information integration to mannequin serving endpoints with out requiring a bigger staff to handle it.
“Databricks SQL was significantly better than what we anticipated,” Bacskai stated. “It’s one thing we like to work with. It might probably do every part we wish, so we are able to simply construct and create what we wish.”
