At Cascade, we believe AI should work for you, without compromising your privacy, your data, or your trust. Here's how we've built AI features you can rely on.
Private by default
Your data is yours. Cascade's AI features are designed with a privacy-first architecture:
- No data sharing across accounts. Your organisation's data is never used to inform results for any other organisation.
- No retention of sensitive inputs. Queries processed by AI features are not stored beyond what's needed to serve your request.
- Role-based access is always respected. AI features only surface information that the requesting user already has permission to see, AI doesn't create new access paths.
We don't train on your data
Cascade does not use your data to train, fine-tune, or improve any AI model, ours or anyone else's. What happens in your workspace stays in your workspace.
This applies to everything: your goals, metrics, comments, documents, and any other content you bring into Cascade.
Low hallucination risk
AI systems can sometimes generate confident-sounding information that isn't accurate. We take this seriously and have designed Cascade's AI features to minimise this risk:
- Grounded in your data. Our AI features are anchored to real information from your Cascade environment rather than generating open-ended responses from scratch.
- Transparent sourcing. Where AI summarises or surfaces information, it references the underlying data (where possible) so you can verify it directly.
- Narrow scope by design. Rather than building a general-purpose chatbot, Cascade's AI is purpose-built for specific tasks - which significantly reduces the surface area for inaccurate outputs.
- Human in the loop. AI outputs are presented as inputs to your decision-making, not as decisions themselves. You stay in control.
AI responses are clearly labelled
Wherever AI generates or contributes to content in Cascade, it's marked as such, you'll never mistake an AI-generated response for a human one or a verified system record.
We also surface a reminder that AI can make mistakes. Even well-designed AI systems can produce outputs that are incomplete, outdated, or incorrect. Cascade's AI labels are a prompt to apply your own judgement before acting on any AI-generated content - especially for important decisions.
Using AI appropriately
Cascade's AI features are designed to assist not to replace human judgement. Here's guidance on where AI works well and where to take extra care:
AI works well for:
- Summarising large volumes of information quickly
- Drafting content as a starting point for human review
- Surfacing patterns or gaps you might otherwise miss
Apply extra care when:
- Using AI outputs to inform high-stakes decisions (e.g. performance reviews, resource allocation)
- The underlying data is sparse, incomplete, or rapidly changing
- The output will be shared externally without further review
When in doubt, treat AI-generated content as a first draft, verify it against source data before acting on it.
Human review before acting on AI outputs
For consequential decisions, Cascade recommends establishing a review step before relying on AI-generated content. This doesn't need to be formal it can be as simple as a team norm that AI summaries are checked against the underlying data before being used in a report or discussion.
Where possible, we have enforced this directly within features. For instance Update suggestions require the user to actively add them to the update, rather than providing a fully formed suggested update for the user. This forces users to critically decide which information should be included.
(note the add buttons to add suggestions from meetings and files to an update)
Your organisation's Cascade admin can work with team leads to define where these checkpoints make sense for your workflows.
Our ongoing commitment
Responsible AI isn't a feature it's a practice. We continue to monitor our AI systems for accuracy, fairness, and unintended behaviour, and we update our approach as the technology and best practices evolve.
If you have questions or concerns about how AI works in Cascade, contact our support team.
Reporting inaccurate or unexpected AI outputs
If you encounter an AI response that seems wrong, misleading, or inappropriate, please report it. Your feedback directly helps us identify issues and improve reliability.
To report an issue:
- Use the feedback option on any AI-generated response directly in the platform. This feedback feeds directly through to our product team.
- Contact your organisation's Cascade admin, who can escalate through Cascade's support process.
- Reach out to Cascade support directly if you don't have an internal admin contact.
We review reported outputs and will follow up where a systemic issue is identified.