Confidence by Spotify

[Confidence by Spotify @SpotifyConf]

When building gets this fast, the bottleneck becomes judgment.
We're expanding Confidence to connect the signals behind good product decisions, from analytics to customer conversations, and keep you in control of every call.

https://t.co/5WDrEf9Hwb

[Confidence by Spotify @SpotifyConf]

Introducing Confidence Agent, now in beta.

An AI collaborator inside Confidence that works with your flags, experiments, metrics, and docs. Describe what you want and it configures the experiment and explains what the results actually say.

Free to try: https://t.co/YTvDEHqDrA

[Confidence by Spotify @SpotifyConf]

Bayesian or frequentist? The Confidence Team at @Spotify wrote up why we think that's the wrong question. The right ones are about whether your experiments are actually producing learning. That's what we built Confidence to do.

https://t.co/XxDxvBkCTJ

[Confidence by Spotify @SpotifyConf]

Can LLMs replace humans in A/B tests? We tested it on thousands of real experiments.

Raw LLM predictions recovered only 39% of the true effect. Calibration helps, but the assumptions break for novel treatments.
https://t.co/mxNrKNNmH2

[Confidence by Spotify @SpotifyConf]

@Spotify Portal is now available on the AWS Marketplace. Faster procurement, consolidated billing, and one less barrier between your team and the context layer your services have been missing.

If you're already on AWS, getting started just got simpler. https://t.co/00FwIkUGFd

[Confidence by Spotify @SpotifyConf]

AI writes the code. Experimentation infrastructure helps decide what ships.

As build velocity accelerates, the advantage shifts from generating code to validating it and learning from every release.

Orgs that pull ahead will both build AND learn faster.
https://t.co/rqwcQdUPzw

[Confidence by Spotify @SpotifyConf]

LLM evals are automated judges — scoring relevance, coherence, and quality at scale — but they're not a replacement for A/B tests. They're what you run before them.

Evals verify. Experiments validate. Use both.
https://t.co/IPBILpoHhA