Skip to content
  • Pricing
  • Contact us
  • Login
Start free
January 7, 2026· Updated May 12, 2026/Yu Zhao, Mårten Schultzberg

Why We Use Separate Tech Stacks for Personalization and Experimentation

Why Spotify runs personalization and experimentation on separate tech stacks: ML systems need their own infrastructure, and A/B tests evaluate them.

Diagram showing separate tech stacks for personalization and experimentation

Start free with Confidence. No credit card needed.

Start free

#TL;DR

At Spotify, we build personalization systems using our ML stack and evaluate them through our experimentation stack. Each tech stack does what it's good at.

Personalization systems have strict infrastructure requirements: access to diverse model types (neural networks, boosting, bandits), rich feature sets, low-latency inference, and real-time data collection. These don't fit naturally inside an experimentation tool. And even if you use a contextual bandit, you still need to evaluate that bandit as a system through A/B tests on different bandit versions. When A/B tests and multi-armed bandits live in the same tool, you get confusing dependencies between instances of the same tool.

Keeping a clean separation of concerns helps us scale with less friction for product teams.

Read the full post on Spotify Engineering: Why We Use Separate Tech Stacks for Personalization and Experimentation

Start free with Confidence. No credit card needed.

Start free

Read next

  • The Feature Flag Toolbox: Cloud, Edge, and Local

    Feature flagging with Confidence: compare cloud, local, and edge flag resolution, the latency and resilience tradeoffs, and how to pick the right setup.

  • A/B Test Bandwidth: The Currency of Innovation

    Learn about how experiment bandwidth is the currency of innovation, and how you can improve yours.

  • Experiment like Spotify: With Confidence

    Confidence brings Spotify experimentation to your team: a warehouse-native platform built from 10+ years of testing at scale, designed to grow with you.

PreviousWhen Proxy Metrics Break: How Optimizing for Proxies Can BackfireNextTwo Questions Every Experiment Should Answer
Spotify

Product

  • Feature flags
  • Confidence Loop
  • Product Analytics
  • Session Recordings
  • Pricing
  • Comparisons

Solutions

  • Startups
  • Scale-ups
  • Enterprise
  • Engineers
  • Data scientists
  • Product managers
  • Data warehouse teams
  • Security

Resources

  • Docs
  • Blog
  • Bootcamp
  • Glossary
  • Why not Bayes?
  • RFP guides
  • Playlist (opens in a new tab)

Company

  • About
  • Contact us
  • Status (opens in a new tab)

Legal

  • Terms of Service
  • Data Protection Agreement
  • Privacy Policy
  • Security

© 2026 Spotify