Spotify cuts quarterly forecast time in half with Flyte

Spotify serves nearly 200 million subscribers across 180 markets. Behind the scenes, finance, automation, modeling, and data teams must collaborate each quarter to produce regulatory-compliant financial reports, including two-year P&L projections and model-based revenue forecasts for emerging markets.

According to Dylan Wilder, Engineering Manager on Spotify’s Vivaldi finance team, end-to-end forecasting previously took 3–4 weeks every quarter. The process involved:

• 8 teams across multiple business units
  
• ~25 contributors
  
• 15+ forecasting models
  
• Inputs from diverse areas like premium subscribers, streaming activity, ad inventory, music ad revenue, and royalties
  

Each team owned models for different business verticals (e.g., podcasts) or horizontal functions (e.g., royalties). The lack of a unified orchestration layer led to manual handoffs, configuration errors, and slow turnaround.

Spotify needed a cohesive platform to coordinate heterogeneous pipelines and reduce operational friction across its forecasting ecosystem.

After evaluating leading orchestrators, Spotify selected Flyte as the main runtime engine for its forecasting platform.

Using Flyte, Spotify built One Model, a unified system that links workflows from Automation and Tooling, Data and Insights, Business Modeling and Forecasting, and Workflow Scheduling into a single forecast engine capable of real-time data crunching.

Flyte improved engineering and data science collaboration by enabling direct model refinement and transparent workflow execution.

By coordinating workflows built with many different technologies, Flyte created a consistent development and deployment experience that dramatically reduced manual overhead and operational complexity.

With Flyte as the orchestration backbone, Spotify accelerated its quarterly forecasting workflow from four weeks to less than two, thanks to automated coordination, reduced handoffs, and higher model consistency.

Flyte also enabled a dramatic increase in forecasting flexibility:

• Previously: 1 major forecast per quarter, requiring alignment across all teams
  
• Now: Thousands of model runs for scenario analysis and analytics
  

“The number of forecasts run went from being limited to once per quarter… to thousands for analytics purposes.”

Manual workflow errors caused by handoffs were eliminated, and the business can now evaluate more cases, more often, giving leadership faster, clearer insights.