Offshore or New Zealand AI developers.

How do offshore and New Zealand AI developers compare?

Levy Street is a New Zealand AI studio, so we have a side in this question; this page still says where offshore wins. “Offshore” means any development firm or freelancer outside New Zealand, from Sydney to Bengaluru to Kraków. “New Zealand developer” means one working here, under New Zealand law.

What does an offshore build really cost?

Compare cost per working result, not day rates. We know of no reliable public survey of AI development rates by country, so put your own quotes, excluding GST, into this formula for offshore cost as a share of the local quote:

Effective cost = r × (1 + c) + m, where r is the offshore day rate divided by the New Zealand rate, c is the extra offshore effort (written specification, handoffs, rework, waiting across time zones) and m is your own oversight time as a share of the local quote. Offshore is cheaper below 1.

The same day rate produces a saving or a loss; the brief and the oversight decide which. Over three years, add support, model migrations and incidents: an outage reported at 9am in Wellington lands at 1.30am in India during New Zealand summer.

On GST: a registered business claims back GST on local invoices, and imported services usually arrive without it. If less than 95% of your use of them is for taxable supplies, Inland Revenue’s reverse charge means you pay the 15% yourself.

How much do time zones matter?

Shared hours decide how fast questions get answered. These figures compare 9am–5pm in New Zealand with 9am–5pm locally, from the IANA time zone database; New Zealand daylight time runs from the last Sunday in September to the first Sunday in April.

Little overlap suits build work with a written specification and a daily handoff note. It is expensive for discovery and live incidents.

What happens to personal information when the developer is overseas?

Developer location and data location are separate decisions. A New Zealand developer can send your prompts to a model in the United States; an offshore developer can work inside your Auckland cloud account on synthetic data. Our guide to data residency for AI systems covers where models run.

• On your behalf. Under section 11 of the Privacy Act 2020, information a developer holds only for you is treated as held by you, even if the developer is overseas. In the Privacy Commissioner’s words, the third party is “you”, and you stay responsible.
• For its own purposes. If the developer reuses the data, say to train its own models, the transfer can be a disclosure. IPP 12 then needs a ground: the recipient is subject to the Act, has comparable privacy laws, or signs a protective agreement (the Commissioner publishes model contract clauses), or the person authorises it after being expressly told protection may not be comparable.
• IPP 3A. Since 1 May 2026, if your system collects personal information from someone other than the person, such as enriching records from a data supplier, you must take reasonable steps to tell them unless an exception applies. Teams new to New Zealand law may not know the rule exists, so put it in the brief.
• Foreign law. MBIE’s responsible AI guidance warns that other countries’ laws may require providers to on-share your data.

The practical rule: give offshore developers synthetic or de-identified data. MBIE’s guidance suggests sharing data profiles or sample synthetic data with suppliers where privacy is a concern; see testing AI with synthetic data. Customers notice too: in the Privacy Commissioner’s 2026 survey, 11% rated their trust in overseas-based businesses to handle personal information as high or very high, against 27% for New Zealand-based businesses.

Who owns the code, and where should it live?

Ownership comes from the contract; control comes from the accounts. Section 21(3) of the Copyright Act 1994 makes whoever commissions and pays for a computer program its first owner, but only subject to any agreement to the contrary, and section 114 requires assignments in writing and signed. An overseas developer’s standard terms, and their local law, may say otherwise. Do not rely on defaults.

• A written, signed assignment of code, prompts, evaluation sets and documents, plus a licence for tools the developer already owned
• Governing law and a dispute process you could realistically use
• Code repository, cloud, model API and billing accounts in your organisation’s name, with the developer as a removable member
• No production personal information without a privacy impact assessment and written limits on use
• Offboarding: revoke access, rotate keys, confirm copies are deleted

Our AI procurement questions give contract wording to ask for.

How do you control quality across borders?

Hold the evaluation set yourself: a fixed collection of real or realistic cases with the expected output, scored the same way every time. MBIE’s guidance tells businesses to set success metrics and thresholds and monitor performance over time. Across time zones it turns “is this good?” into a number nobody argues about on a late call.

1. Set pass marks before the build. For example: 95% of invoices coded correctly, every low-confidence case sent to a person.
2. Keep the set in your repository. The developer runs it but cannot change expected answers without your sign-off.
3. Run it on every change. Prompt edits, model switches and library upgrades all count.
4. Pay against results. Tie milestone payments to the evaluation report, not a demo video.

Our AI vendor evaluation checklist shortlists suppliers on evidence, and the agency versus internal hire comparison lists the handover items to require.

Which should you choose?

Choose an offshore developer if

• The work is well specified: an integration, a data pipeline, an interface, a defined module.
• Someone in your business can review code and run the evaluation set.
• The data is non-personal, synthetic or de-identified.
• You know the team, or can start with a small paid module.

Choose a New Zealand developer if

• The job starts with discovery: mapping how your staff really work.
• The system touches personal, health or financial information, or a regulated process.
• Incidents must be handled in your business hours.
• Nobody in your business can direct and check offshore work.

Or combine them

One workable hybrid: a New Zealand lead owns discovery, architecture, the evaluation set and acceptance, and an offshore team builds defined pieces against it. It works only if the lead can reject work. Ask any provider, Levy Street included, whether it will lead work someone else builds. Our custom AI development service agrees acceptance criteria and an evaluation set before code is written. To talk through your case, book an assessment.

Where offshore AI projects go wrong

• Specification by chat thread. Requirements drift across late-night messages; keep one versioned brief.
• The developer owns the accounts. When the relationship ends, so does your access.
• Production data as test data. Real customer records copied to laptops in another country.
• Demos instead of measurements. A polished video and no evaluation results.
• The cheapest bid on a discovery job. Offshore teams do well on known problems; discovering your workflow over video at a lower rate usually costs more.

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