AI provider (Gemini / Gemma)

Novro’s AI features (for example the AI action-plan suggestions after a closed survey) can run against different model providers Each organization chooses which provider its AI requests are sent to:

Gemini (default): Google’s Gemini service. Nothing changes for existing organizations unless a provider is actively chosen.
Gemma (Vertex AI): Gemma 4 running in Novro’s own EU cloud environment — recommended when data residency is required.
Gemma (self-hosted): Gemma 4 running on Novro’s own infrastructure.

All AI features are off by default. No AI feature sends anything to an AI provider until it has been explicitly turned on for your organization (by Novro). This covers the Startsida AI assistant, AI agents, AI survey insights, AI survey translation, AI-composed dashboards, the AI pay-gap/DEI summary, the AI assistant for OKR and PIP, and the other AI suggestions. Once a feature is on, an Admin can still turn off individual AI areas and control which roles may use them.

Today no Gemma option is deployed in any Novro environment, so Gemini is what is used in practice. Gemini is a global service with no guaranteed processing within the EU.

What is sent depends on the feature. For survey-based analyses (such as the action-plan suggestions) no prompt is built and no model is called unless the anonymity threshold (minimum group size) is met for the exact survey and scope being analyzed, and then only aggregate category averages and eNPS are sent. Other AI features send more: AI survey insights send free-text comments (above the threshold), and the assistant and AI agents send the user’s question, knowledge document excerpts and tool results, which may include personal data the user is authorized to see. Each AI area can be switched off for the organization. See the subprocessor list in the privacy policy for details.

You pick the provider under Settings → System Parameters → AI provider (Admin only). The picker lists all three options, but any Gemma option that isn’t available to you is disabled and marked “not configured”, so you can never select a provider that won’t work. Gemini is always available and is the default. The choice is stored per organization and is also validated on the server (only the known values are accepted).

When the provider is Gemini, an “AI Model” picker also appears — choose Gemini 2.5 Flash-Lite for a cheaper model per call, or Gemini 2.5 Flash (default) for the more capable one. The model picker only applies to the Gemini provider (the Gemma options have a fixed model, not a per-organization choice) and is automatically hidden if you select a Gemma option in the provider picker above. It saves together with the provider choice in the same Save click.