NZ mid-market AI operations: the wedge is not chat, it is exception handling
A practical alpha memo for founders and operators looking for AI opportunities that can ship before a grand platform rewrite.
Thesis: The highest-value near-term AI work in New Zealand mid-market companies is buried in exception queues: documents that do not match, orders that need judgement, jobs stuck between systems, and approvals waiting for context.
Signals to look for
- Manual triage queues with high variance
- Shared inboxes acting as workflow engines
- ERP/CRM fields corrected after the fact
- Managers approving work based on context scattered across emails and PDFs
Forward-deployed moves
- Build audit-first extraction and recommendation tools
- Keep humans in the loop at the decision boundary
- Measure time-to-resolution, rework, and exception ageing
- Productise repeated exception patterns into workflow primitives
Risks and traps
- Over-automating high-liability decisions
- Ignoring data ownership across customer systems
- Treating LLM accuracy as the only metric