AI for New Zealand professional services.

Where AI pays off in a consultancy or advisory firm

AI pays off in professional services firms where senior people spend unbillable hours on first drafts: proposals, background research, meeting write-ups and status reports. Those tasks draw on material the firm already has, such as past proposals, reports and CVs, which is exactly what retrieval-based AI works from. The expertise clients pay for, and the professional sign-off, stays with people.

The sector is large in New Zealand and heavily concentrated in the main centres. Professional, scientific and technical services employ about 24,600 people in Wellington City (14.7% of employees) and 93,200 across the Auckland region, according to Stats NZ business demography data for February 2025 (provisional; shares computed by us). It is also the Wellington region’s largest industry by GDP on our ranking of Stats NZ figures, at about $7.2 billion in the year to March 2024 (Stats NZ regional GDP). These firms also dominate the evidence on AI use: 43% of the 166 responses to the AI Forum’s August 2025 productivity survey came from the sector, and 91% of businesses reported efficiency improvements. It is a self-selected industry-body survey, so expect it to lean towards enthusiasts.

Professional services workflow map

There are worked guides for proposal preparation, research briefing preparation and turning meeting actions into project tasks.

Ownership, claims and confidentiality rules to settle first

• Copyright in deliverables. Under section 5(2)(a) of the Copyright Act 1994, the author of a computer-generated work is the person who made the arrangements necessary to create it. No court has applied that to generative AI prompts, and Cabinet has asked for a report back by 31 March 2027 on a possible copyright framework for generative AI. Put ownership of deliverables in your engagement terms.
• Claims in proposals. The Fair Trading Act makes unsubstantiated representations an offence (s 12A), and the Commerce Commission says claims must be substantiated when they are made. An AI-drafted proposal that invents a past project or outcome is a legal problem as well as an embarrassment.
• Client confidentiality and privacy. Client documents often contain personal information and commercially sensitive material, and many contracts limit where it can be processed. The Privacy Act holds your firm responsible for what its providers do with personal information on your behalf.
• Professional codes. Bodies such as Engineering New Zealand set competence and confidentiality expectations that apply to AI-assisted work as to any other. Check your own body’s current position.

The tools most firms already pay for

Most NZ firms already run Microsoft 365, Teams for meetings, a project tool such as Asana, Monday or Jira, a CRM such as HubSpot, and a practice or job management system. Microsoft 365 Copilot is often the cheapest place to start, and Microsoft says it only surfaces organisational data a user already has permission to view. That is a strength and a warning: years of over-shared SharePoint folders become searchable overnight. Our notes on Microsoft 365, Asana and the Copilot versus ChatGPT comparison cover the set-up choices.

A first project: proposal preparation from approved material

Inputs

The RFP, your last 30 proposals (won and lost), approved CVs, and project summaries cleared for reuse.

Systems

A SharePoint proposal library with retrieval and citations, Word, and a model chosen by testing.

Steps

Build the compliance matrix, draft an outline against the evaluation criteria, then draft each section citing the library documents it used.

Human review

The bid lead checks every claim of experience against an approved project sheet; a partner edits approach and price.

Outputs

A compliance matrix and a cited first draft.

Measures

Hours per proposal, compliance gaps found late, share of experience claims traceable to an approved source, and win rate over six months.

Risks

Invented experience, and client names reused without permission.

Illustrative example: a 40-person engineering consultancy submits 10 proposals a month at about 16 hours each, 160 hours of mostly senior time. If a cited first draft and matrix cut that to 10 hours, the firm frees 60 hours a month. At an assumed $150 an hour of senior time, that is about $9,000 a month (NZD) of capacity before running costs. An AI consulting assessment tests this on your own library first.

Research agents, and how we test them

GTM-Bench is a 72-task benchmark for AI agents doing go-to-market research, published by Blackpearl Group in June 2026 and co-authored by Levy Street team members. Its lesson for consultancies is simple: judge a research assistant on real tasks with checkable answers, not on how fluent the brief sounds. The same team works on Bebop, Blackpearl Group’s AI sales agent for small and medium businesses; Levy Street engineers work with Blackpearl Group’s AI team on Bebop.

Where AI fails consultants

Workflows we map in this sector

• AI proposal preparationDraft tender and client proposals from your own project library, with every claim checked before it goes out.
• AI meeting actions into project systemsTurn meeting transcripts into owned, dated tasks in your project tool, with the project lead approving each one.
• AI research briefing preparationShort, cited briefings on a client, sector or question, with every statement checked against its source.

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