PostHog will train AI models with your data (opted-in by default)

PostHog has announced that it will begin training AI models on user data, with opt-in by default for US cloud users, to enhance its product offerings. This move aims to improve features like session replay analysis, user testing automation, and predictive insights, making PostHog’s tools more proactive and powerful.

PostHog revealed plans to use customer data to train AI models, starting June 29, 2024. Users on the US cloud instance will be opted in by default, while EU users are opted out unless they manually opt in. All data used will be anonymized and only existing data within the customer’s PostHog instance will be employed. The company emphasizes that it will not sell or share data with third-party providers, and all training will be conducted internally.

PostHog’s initiative aims to make existing features smarter—such as detecting issues in session replays—and to develop new capabilities like synthetic user testing and predictive feature improvements. The company states that these AI-driven features are experimental and will require iteration to optimize.

Why It Matters

This development is significant because it marks a shift toward integrating user data more directly into AI-driven product enhancements. It raises questions about data privacy, user consent, and transparency, especially given the default opt-in status for US users. The move could influence industry standards around data use for AI training in SaaS products, emphasizing transparency and user control.

Background

Over the past year, PostHog has integrated AI-powered features such as an AI installation wizard and the PostHog AI tool, which have been well received. The company’s broader vision is to build proactive, self-driving products that surface insights and automate tasks. This announcement aligns with industry trends toward using customer data to train AI models, but it also marks a notable shift in PostHog’s approach to data use and transparency.

“Our goal is to improve PostHog as a product for our customers, not to expose or sell your data, or monetize your data.”

— PostHog CEO James

“Training won’t start until June 29, giving users ample time to opt in or out.”

— PostHog spokesperson

What Remains Unclear

It is still unclear how effective the trained models will be in practice, what specific improvements they will enable, and how users will respond to the opt-in default. The company emphasizes experimentation, so outcomes remain uncertain. Additionally, the extent of user data anonymization and privacy safeguards, while stated, could face scrutiny or challenge.

What’s Next

PostHog will begin model training on June 29, 2024. The company plans to notify users via email and in-app messages, and will monitor feedback and effectiveness of the new features. Future updates may include further transparency measures or adjustments based on user preferences and legal considerations.

Key Questions

Will my data be sold or shared with third parties?

No. PostHog states that all training will be conducted internally, and they will not sell or send your data to third-party model providers.

Can I opt out of having my data used for AI training?

Yes. Users on the EU cloud instance are opted out by default, and all users can opt out at any time via their organization settings, provided their legal agreements do not prohibit this.

What data will be used for training?

Only data that already exists in your PostHog instance will be used, and it will be anonymized before training.

When does training start, and how will I be notified?

Training is scheduled to start on June 29, 2024. PostHog will notify all users via email, in-app notifications, and public communication channels.

Will this improve PostHog’s features immediately?

The company describes these efforts as experimental; improvements will depend on ongoing iteration and testing of the models.

Source: Hacker News