Balance in
Motion.

We establish methods to regularly test your AI models for bias and unfair outcomes. This includes defining what "unacceptable bias" means for your context (e.g. demographic bias in lending decisions) and setting up periodic audits. You'll get templates and tools for running these tests before and after deployment.

Black-box algorithms won't cut it when an explanation is needed. We provide guidelines for using explainable AI techniques (like SHAP values or decision trees for interpretable models) so that for any automated decision – especially high-stakes ones – your staff can generate a human-readable rationale. This helps with both internal understanding and meeting external transparency requirements.

Aligning with the NZ Privacy Act, we'll document how long you retain data used in AI systems and when to delete or anonymize AI. We set rules for sourcing data ethically and ensuring consent, and include a schedule for regular data hygiene checks. These policies safeguard privacy and give regulators confidence that you're handling personal information responsibly.

In the event something does go wrong – say an AI error causes an incident or a user lodges a complaint – you'll be prepared. Our framework includes an incident response plan specific to AI issues: clear steps for investigation, notification (if needed), remediation, and communication. Having this ready can be the difference between a contained issue and a public relations fiasco.

We help you formalize an AI Governance Committee or assign roles (like an AI Ethics Officer) appropriate for your organisation. Rather than layer on heavy bureaucracy, we define a lean governance structure that fits your operations. For example, a monthly AI review meeting or an AI ethics checklist for project approvals might be all that's needed to enforce the new standards.