Training Our Own AI Models

Beyond features like the AI installation wizard, PostHog AI, and MCP, PostHog is pursuing model training using internal data to build proactive, autonomous products that solve problems on behalf of users.

The core target is PostHog Code, with the goal of creating a virtuous cycle in which the product improves itself. Specific areas of application include the following:

• Session replay analysis: Training on the underlying data of replays to lower the cost of large-scale analysis and improve problem detection performance.
• Synthetic user testing: Automatically identifying confusion or flow errors that users might experience before a product is deployed.
• User behavior prediction: Suggesting optimal changes to improve conversion rates and reduce user frustration.

Regarding data security, all training data is anonymized, and data is not sold or shared with third-party model providers.

The operating approach adopts an opt-out by default model. EU Cloud users and certain contracted customers are excluded by default, while US Cloud users are included by default. Users can opt out of training at any time in their organization settings before June 29.