How Seven.One Entertainment Orchestrates Data Pipelines with Prefect

Meet Seven.One Entertainment

When ProSiebenSat.1 relaunched its streaming service Joyn as a joint venture in 2019, the data engineering team had a rare opportunity: build the data infrastructure from scratch. They chose Snowflake for data warehousing, dbt for modeling, and a clean break from the on-premise legacy systems that had constrained them.

ProSiebenSat.1 Media SE Group, to which Seven.One Entertainment belongs to, is a private broadcasting company operating linear TV channels across Germany, Austria and Switzerland alongside Joyn, a free ad-supported streaming service also available in the same countries. The content spans international film and television, news, and sports, with a growing emphasis on local programming. It's primarily a B2C business: viewers can watch without even registering, though registration or paid subscriptions unlock additional features and content. The core audience is everyday viewers tuning in across a wide range of devices like smart TVs, mobile phones, tablets or web. Also strategic B2B partnerships ensure widespread availability.

Conrad Fritzlar, who heads the data engineering reach team at Seven.One, is responsible for making sure data ends up in a usable state in the data warehouse, ready for reporting and downstream consumers. Those consumers include the customer relationship management team, the marketing team, content controlling, and the analytics groups that help the company make data-driven decisions. There's transformation and enrichment logic layered on top of the raw data, and the outputs feed reporting tools, CRM workflows, and audience insights that shape programming and advertising strategy.

Before Prefect

Before the 2019 relaunch, data infrastructure ran on legacy, on-premise software with its own orchestration tool. Conrad recalls it was "actually quite powerful, but not a good fit anymore." The broader problems were typical of on-premise systems nearing end-of-life: things ran slowly, reprocessing was painful, and monitoring left much to be desired. "We were in the lucky position that we could basically say goodbye to the old tech stack and build something from scratch," he says. Moving to the cloud was about gaining the flexibility and performance needed to handle the volume of data a major streaming platform generates.

After the relaunch, the team handled orchestration within Snowflake using Streams and Tasks, Snowflake's built-in mechanism for triggering workloads on a schedule. For simple use cases, it worked. As pipelines grew more complex, with more dependencies and a need for state handling and proper monitoring, Streams and Tasks became limiting quickly. They were confined to Snowflake, and the data team needed orchestration that could also handle data distribution to systems outside the data warehouse. "That's where Prefect came in and was super handy" Conrad says. "For data extraction, picking up data from the different source systems, but also especially the overall orchestration of all the flows from raw data to something meaningful."

The team evaluated Airflow alongside Prefect. "Usually you come across Prefect and Airflow if you want to look into some proper orchestration," Conrad says. "For us, Prefect looked more promising." The Python-native, code-centric approach was central to that. "You can easily make adjustments, just use the libraries. You can tailor it to whatever you need with a flow and tasks, so it's super flexible," he explains. "Prefect is kind of best in class, I would say." The annual costs were also quite reasonable, a practical consideration that mattered alongside the technical fit.

Orchestrating from source to insight

Today, Prefect sits at the center of Seven.One's data infrastructure with roughly 70 different flow definitions. The team has built extensive custom implementations for fetching data from various APIs and source systems. AWS figures prominently: Prefect workers run on AWS ECS clusters, flows fetch additional information from S3 buckets and authenticate through AWS Secrets Manager. Deployments are automated in a standardised CI/CD workflow. The required data arrives from APIs, SFTP servers, and other sources, sometimes undergoing small transformations within Prefect itself before being ingested into Snowflake via SQL statements.

What makes Prefect's role at Seven.One worth studying is how thoroughly it functions as end-to-end orchestration rather than just a scheduler. The most critical pipeline, which the team calls the "mother of flows", illustrates this well. It begins with a foundational stage where data is prepared and loaded into Snowflake's persistent layer. From there, dbt runs are triggered through Prefect, handling the transformations that build out data marts. "dbt has its internal DAG and lineage, but the whole triggering and the monitoring is happening inside Prefect," Conrad says. After dbt completes, data quality testing runs automatically to verify the data meets expectations. Then the pipeline fans out: reporting data flows to reporting tools, specific information models get orchestrated for CRM, marketing and analytics teams, and additional data exports and sharing jobs kick off from there. "A lot of that is happening in Prefect, where we're connecting the different systems via flows and tasks," Conrad says.

"Prefect is the one-stop shop for us for extracting data from the sources and doing all the necessary steps for orchestration that happen in between," Conrad says. Beyond the orchestration itself, the day-to-day experience of working with Prefect matters to the team. Conrad highlights the UI as a key part of that: "Prefect gives you a good UI where you can very easily re-trigger things, see what the status is, get insights from the logs. Debugging and monitoring become a lot easier."

Lightweight architecture, deep control

Conrad points to the overall design philosophy more than any single feature. "I kind of like the overall architecture of Prefect," he says. "It's relatively lightweight, but you have a lot of control over what you do. It's pretty easy to orchestrate things. You have Python code underlying, lots of built-in libraries, and a good user interface that makes it easy for everyday work."

That combination of lightweight and controllable matters for a team connecting a wide variety of systems. Because everything is Python code, data engineers can use familiar libraries and patterns to tailor flows to whatever they need.

What a streaming data team actually needs

Seven.One Entertainment's story is a team that methodically built a modern data stack, identified the gap where orchestration needed to live, and chose a tool that fit the way they already worked. The value shows up in the daily reality of running interconnected pipelines across extraction, transformation, quality testing, and distribution, with the visibility and control that was missing before.

For media and entertainment companies dealing with similar complexity, with data arriving from many different sources and multiple downstream teams depending on that data, Seven.One's approach is worth studying. Start with the data warehouse, add modeling and transformation, then bring in orchestration that can tie the full pipeline together from source systems through to the teams that actually use the data.