Big consultancy or specialist AI firm.

Levy Street is a specialist AI studio, so we have a side in this question. This page still sets out when a large firm is the better choice, using only what each firm’s New Zealand pages say, as listed on 20 October 2026.

How big consultancies and specialist AI firms differ

A big consultancy sells breadth: strategy, risk, change management and technology from one firm, at scale. A specialist AI firm sells depth in one thing: designing, building and running AI systems with a small team.

What New Zealand’s large firms say they offer

• Deloitte New Zealand lists AI strategy, AI insights, data modernisation and intelligent automation. Its Tower case study describes an AI contact centre built with AWS and an ongoing service alongside managed support.
• PwC New Zealand names a Chief AI Officer and an AI Transformation Leader, and frames AI work as redesigning processes, roles and skills.
• KPMG New Zealand runs KPMG Trusted AI, built on ten principles from fairness to sustainability, with Microsoft Copilot readiness as a related service.
• EY New Zealand’s AI page leads with global insights, EY.ai and case studies.
• Accenture serves New Zealand visitors its global AI and data page; its New Zealand company page cites 814,000 employees worldwide.
• Datacom says its AI consultants work on governance frameworks, proofs of concept and AI tools across Australia and New Zealand, and that AI agents write up to 70% of code in its legacy modernisation work.
• Spark offers data and AI services through Qrious, covering strategic advisory, project consulting and data operations.

How government panels change the answer

For agencies, procurement often decides before capability does. The All-of-Government consultancy services contract, led by MBIE, has run since 6 November 2015 with no set end date. A keyword search of its provider list on 20 October 2026 returned Deloitte, PwC Consulting, KPMG, Ernst & Young Limited, Accenture NZ Limited and Datacom Systems Limited among 927 providers. A search for “artificial intelligence” returned five.

Joined agencies can choose a provider directly without advertising on GETS. Providers sit in three tiers: Tier 1 firms are medium to large with at least $500,000 of professional indemnity cover, and Tier 3 firms are small to medium, with no insurance minimum and no admin fee. The last refresh closed on 30 June 2021, and the next tender is expected in the first part of 2027. None of the panel’s 11 categories is AI.

The government Marketplace is the other route. It is open to any business meeting a channel’s entry criteria, and its consultancy channel includes ICT professional services and data and information services. Firms outside the big six use it too: Catalyst, an open-source company with an AI practice, says it is a tier 2 Marketplace supplier for managed services and consulting. Ask every shortlisted firm, Levy Street included, which panel or catalogue it can be bought through.

When your auditor cannot build it

If your auditor also sells consulting, as the Big Four do, check independence rules before asking it to build. New Zealand’s PES 1 treats operating, maintaining or monitoring an audit client’s IT systems as a management responsibility the firm cannot take on. For public interest entities, an audit firm must not provide IT systems services that might create a self-review threat, such as systems that affect accounting records or financial reporting controls. An AI agent that codes supplier invoices into your ledger is the kind of system in scope. Ask your auditor early; this is general information, not advice.

How to test both on the same brief

Pitches favour whoever presents best. Pay one large firm and one specialist for a small, identical piece of work instead.

1. Write one brief. One workflow, its volume and the number to move. Illustrative example: triaging 500 public emails a week to the right team, measured by time to first response.
2. Ask for the same deliverable. A test on 50 of your own emails with personal details removed, a monthly running cost in NZD, and the names of the people who would build it.
3. Score evidence, not slides. Compare test results, stated assumptions, whether privacy and data location came up unprompted, and who did the work.

What each costs over three years

Neither model is cheaper by definition. AoG rates are commercially sensitive, and Levy Street does not publish rates, so use your own quotes.

Break-even rule for build quotes: the shorter quote is cheaper unless its day rate is higher by more than the ratio of the two day counts. Illustrative example: if quote A needs 60 days and quote B needs 40 for the same scope, B is cheaper unless its day rate is more than 1.5 times A’s (60 ÷ 40).

Running costs compound. Illustrative example: a support retainer equal to 2% of the build price per month adds 72% of the build price over 36 months. Ask for running costs in NZD per month before you compare build quotes.

Which should you choose?

Choose a big consultancy if

• AI is one part of a larger transformation with restructuring, process redesign and several business units.
• You are an agency buying through the AoG panel and the firm covers the right category.
• You need risk, governance and change management delivered alongside the build by one firm.
• You need many people in several cities at once.

Choose a specialist AI firm if

• You need one working system soon, measured against an operating number.
• You want to meet the engineers who will build it before you sign.
• You want models from several labs tested side by side on your data.
• Your auditor is a Big Four firm that cannot build or run it.

Many organisations use both: a large firm for strategy, risk and change, a specialist for the build. Our AI consulting and custom AI development pages describe what Levy Street does.

Where both go wrong

• The team changes after the pitch. Ask for named people and the share of days at each of the AoG’s five job levels, or the firm’s equivalent.
• Strategy without a system. A roadmap is a means. Set a date for the first system in production.
• Key-person risk. A specialist with two engineers needs documented cover and an exit plan. Our AI procurement questions cover support and exit terms.
• Platform lock-in. Proprietary accelerators and alliance platforms can be hard to leave. Get the code and prompts in accounts you own.
• No owner inside. Neither works without an internal owner; see AI agency vs internal hire.

For a shortlist of New Zealand firms, see our guide to AI consultants in New Zealand, or book an AI assessment.

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