AI model policy

No third-party model training

Customer data — conversations, knowledge content, bookings, attachments — is never used to train the external AI models we rely on. This is contractually guaranteed by our infrastructure providers and applies regardless of which provider we use. The models process your data to produce answers; they do not learn from it.

How Omniday improves the product

To monitor and improve answer quality, chat messages are exported to our own EU-hosted observability workspace only after personal identifiers have been scrubbed. Tenant knowledge content is not modified by this process. Where we analyse product usage across customers, we use de-identified, aggregated telemetry. Customer-specific tuning — for example, adjusting how the assistant answers for your organisation — happens only under your instructions, within the data processing agreement.

There is no raw cross-customer reuse of your data, and no arrangement where one customer’s content improves another customer’s assistant, unless your organisation explicitly agrees to a different arrangement in writing.

Where this is documented

Observability runs first-party, in our own monitoring workspace on the Microsoft Azure infrastructure listed in the sub-processor register — no separate observability provider is engaged. The scrubbing step is part of our technical and organisational measures, described in the published data processing agreement.