datainterview.org
Company deep dives

Spotify interviews

Practitioner research · human outcomes and AI proxies · 2026

What identifies the effect?

“Unlike random assignment in a user experiment, neither assumption is guaranteed by design.”

Sebastian Ankargren, Joel Persson and Mårten SchultzbergSenior Data Scientist; Research Scientist, Computational Economics; Senior Manager, Staff Data Scientist

Spotify Engineering · 2026-08-13

Name the assumptions

What evidence connects a simulated preference to a human booking?

Example

Verify output. Validate outcomes.

“Without offline-online signal calibration, our evals are opinions, not evidence.”

Matilda Ankargren and Mårten SchultzbergSenior Data Scientist; Senior Manager/Staff Data Scientist

Spotify Engineering · 2026-05-18

Validate the proxy

Would a higher offline score establish better user outcomes?

Example

“Human experiments therefore remain indispensable for true product innovation.”

Sebastian Ankargren, Joel Persson and Mårten SchultzbergSenior Data Scientist; Research Scientist, Computational Economics; Senior Manager, Staff Data Scientist

Spotify Engineering · 2026-08-13

Collect human outcomes

What human experiment would change the decision?

Example

Last reviewed