AI for New Zealand recruitment.

Where AI helps a recruitment team

AI helps recruiters with the writing and admin that fills most of a consultant’s week: turning interview notes into client-ready summaries, keeping candidates informed, tidying a CRM full of duplicates, and researching which companies are hiring. It is a poor fit for deciding who progresses, because that decision carries legal risk and a model cannot see what a good recruiter sees in a conversation.

Candidates are paying attention to how their information is used. In the Privacy Commissioner’s 2026 annual privacy survey of 1,245 adults (fieldwork March to April 2026), 82% of New Zealanders said they want more say in how their information is collected and used. Recruitment databases are some of the largest stores of personal information most small businesses ever hold, which makes the privacy design as important as the time saved.

Recruitment workflow map

Worked guides cover interview note summarisation, drafting candidate communications and cleaning up recruitment CRM data.

Discrimination and privacy rules that apply to hiring tools

New Zealand has no recruitment-specific AI law. Three existing regimes do the work, and all three apply whether the tool is yours or a vendor’s.

• Human Rights Act 1993. Section 21 lists 13 prohibited grounds of discrimination, including sex, age, disability, ethnic or national origins and family status. Section 65 makes indirect discrimination unlawful: a neutral-looking practice that has the effect of treating people differently on a prohibited ground, subject to a good-reason defence. A ranking model trained on past hires is exactly that kind of practice. MBIE’s Responsible AI Guidance for Businesses illustrates the risk with “BigBuild”, a hypothetical construction firm trialling AI CV screening.
• Privacy Act 2020 and IPP 3A. Since 1 May 2026, an agency that collects personal information from someone other than the individual, such as a referee, a LinkedIn profile or another agency, must take reasonable steps to make the person aware of the collection, its purpose, the recipients and their access and correction rights, as soon as reasonably practicable (OPC guidance on IPP 3A). It does not apply to information collected before that date. Accuracy (IPP 8) and limits on reuse (IPP 10) matter too: the Privacy Commissioner treats inaccurate or made-up information about a person as personal information.
• Biometric Processing Privacy Code 2025. Automated processing of faces, voices and other physical or behavioural features is covered by the code, which has the force of law and limits intrusive uses such as predicting emotions or inferring ethnicity or sex. Video interview analytics sit squarely inside it.

This is general information. Get employment and privacy advice before introducing any screening or scoring tool; our AI governance advisory covers the privacy impact assessment and outcome testing.

The recruitment software stack

Most NZ agencies run an applicant tracking system or recruitment CRM such as JobAdder or Bullhorn, advertise on Seek, Trade Me Jobs and LinkedIn, interview over Microsoft Teams or Zoom, and use HubSpot or a similar CRM for client business development. AI earns its place when it writes back into those systems with the recruiter’s approval, rather than living in a separate chat window where notes get copied and lost. Our notes on HubSpot, Microsoft Teams and connecting AI to your CRM cover the permission settings that matter for candidate data.

A first project: interview note summarisation

Summaries are a good start because consultants already write them, the recruiter reviews every one before it is saved, and nobody is rejected by the system.

1. Agree the template. For each role family: criteria, evidence quoted from the candidate, concerns, the candidate’s own questions, and next steps.
2. Tell candidates. Update your privacy notice, and ask permission before any interview is recorded or transcribed.
3. Test on past interviews. Run 30 recent interviews and have consultants mark every factual error, missing piece of evidence and invented claim.
4. Pilot with edits. Consultants edit each summary before it reaches the CRM or the client, and log their changes.
5. Measure and decide. Track edit time per summary, invented claims (the target is zero) and time from interview to client feedback.

Illustrative example: an eight-consultant agency runs about 12 interviews per consultant a week, and each write-up takes 20 minutes, so 8 × 12 × 20 = 1,920 minutes, or 32 hours a week. If reviewing an AI draft takes 7 minutes, the total falls to about 11 hours, freeing roughly 21 hours a week for candidates and clients. Our AI workflow automation service builds this into the CRM rather than a separate tool.

Business development and Bebop

Agency growth depends on finding employers about to hire, which is a sales prospecting problem. Bebop is an AI sales agent for small and medium businesses, owned and sold by Blackpearl Group (NZX/ASX: BPG); Levy Street engineers work with Blackpearl Group’s AI team on Bebop. The lesson carries over: research agents are useful when they show their sources, and a consultant still decides who to call.

Where AI fails in recruitment

Workflows we map in this sector

• Interview note summarisationInterview transcripts turned into scorecard notes with quotes and timestamps, checked by the interviewer.
• Candidate communication draftingInvites, status updates and post-interview feedback drafted from real notes, approved by a recruiter before sending.
• Recruitment CRM data cleanupDuplicates, stale records, messy skills tags and risky notes found by AI, with every change approved by a person.

Written by the Levy Street team in Wellington. Last updated 19 October 2026.