Responsible AI

AI operators can trust.
AI procurement can sign off on.

Every Nova AI model has an owner, a test suite, and an audit trail. Voice AI escalates ambiguous orders. Vision AI alerts managers — it doesn't act unsupervised. Reporting AI cites sources. Built for the diligence enterprise restaurant brands and their CIOs actually run.

Four principles

How Nova's AI is governed end-to-end.

🗄

Data governance

Guest data is partitioned per brand. PII is encrypted at rest and in transit. Customers control what trains shared models — opt out at any time.

🎯

Model evaluation

Every model has a test suite, a benchmark, and a regression gate. We measure before we ship. Performance and bias metrics tracked over time.

👤

Human-in-the-loop

Voice AI escalates ambiguous orders. Vision AI alerts managers — it doesn't take action unsupervised. Manager Copilot recommends, the manager decides.

📜

Audit & explainability

Reporting AI cites sources. Manager Copilot shows its work. Every AI decision is traceable to its underlying data. Full audit log of AI actions.

🚫

Bias testing

AI upsell, scheduling, and pricing models tested for unintended bias across guest demographics. Tested before deployment and continuously in production.

⚠

Defined boundaries

Clear scope for what each AI capability can and cannot do. Voice AI doesn't authorize refunds. Vision AI doesn't take disciplinary action. Manager Copilot can't modify financial records.

Customer data & AI

What trains Nova's AI — and what doesn't.

Restaurant data is the most valuable signal we have, but how it's used matters.

By default

Customer-specific models (your menu, your guest history, your prep times) are trained only on your data and inaccessible to other Nova customers.
Aggregate, anonymized signals may inform shared models (e.g. benchmarks across the industry), but never with PII or brand-identifying data.
You control what is shared — opt out of shared model contributions at any time without losing access to your dedicated models.

For enterprise customers

Dedicated model deployments available — your data never leaves your tenant for any AI inference.
Custom AI governance addenda available in MSA negotiations.
Access to model evaluation reports and bias testing results during procurement review.

Where Nova AI runs

Every AI surface in the Nova platform — and how it's governed.

Voice AI — Escalates ambiguous orders to a human in the loop. Full transcript and audit trail on every call. PCI-compliant payment capture.
Vision AI — Alerts managers; doesn't take action unsupervised. Camera feeds processed on-premise or in dedicated tenant cloud.
Reporting AI — Every answer cites the underlying data rows. Role-based access controls what the model can read.
AI Marketing — Generated campaigns require human approval before send. Attribution back to revenue is logged.
AI Upsell — Recommendation engine; cannot override prices, menu, or modifiers. Tunable aggressiveness per brand.
AI Labor Optimization — Recommends; manager approves. Schedule changes require human confirmation.
Manager Copilot — Reasons across data, recommends actions. Role-based permissions on what Copilot can act on. Full audit log.

AI operators can trust. AI procurement can sign off on.