AI Strategy in New Zealand

Specify the deliverable before you engage

This is the one category on the site where the output is entirely within your control at the point of buying, and where failing to specify it reliably produces something you cannot act on.

Write into the engagement that you will receive:

• A ranked list of specific opportunities. Not "use AI in customer service" — "draft first responses to the 400 support emails a month currently handled by two people".
• A cost and an expected return against each, however rough.
• A named owner in your business for each.
• One project defined well enough to be quoted by someone else.
• A usage policy your staff can follow from Monday.
• An honest statement of what is not ready — data, systems or people.

Ask for a redacted example from another client before engaging. The gap between a genuinely useful strategy document and a maturity model with four quadrants is immediately visible, and it is the single best predictor of what you will receive.

The policy is the urgent part

Your staff are already using these tools. Some of them are pasting customer information into personal accounts, and nobody has told them not to because nobody has decided.

That is worth fixing in weeks rather than in a strategy cycle. A workable policy covers:

• Approved tools, paid for by the company, on business terms rather than personal accounts.
• What may go in — and specifically what may not: customer personal information, anything under a client confidentiality clause, health or financial records, unreleased commercial material.
• What needs approval before it is used in a customer deliverable.
• Accountability. The person who sends the output owns it. "The AI wrote it" is not a defence to a client.
• Disclosure — whether clients are told, which some contracts now require.

Written well, this is a two-page document rather than a manual. It removes most of the real risk in the next twelve months, and it costs a fraction of a strategy engagement.

Look for boring opportunities

The projects that pay are unglamorous, and the ones that make good slides usually are not.

What reliably works:

• Documents somebody retypes — invoices, forms, dockets.
• Enquiries somebody reads and routes.
• Long material somebody summarises before a meeting.
• Questions answered repeatedly from documentation.
• A first draft somebody writes from a template.

What usually does not, yet:

• Anything where a single mistake is expensive and invisible.
• Anything requiring data you do not have or never recorded properly.
• Anything replacing judgement your customers are paying you for.
• Anything whose value depends on a system integration nobody has costed.

A consultant who returns with three unglamorous opportunities and a clear first project has done the job. One who returns with a transformation programme has written a document.

Ask them to say no

The most valuable sentence in an AI strategy is often that the business should do one small thing this year and wait on the rest.

Ask each firm directly: will you tell us if the answer is to do nothing much yet? A consultant willing to say it has just reduced their own follow-on work, which is exactly the incentive you want to test for.

Also ask whether they would bid to build what they recommend. That does not disqualify them — often they are best placed — but you should know the arrangement before you read the recommendations.

Readiness is usually a data question

Most AI opportunities in an ordinary business are blocked by the same things, and they are not exotic:

• The outcome you would want to predict was never recorded consistently.
• The documents are scanned images with no searchable text.
• The knowledge is in people's heads and in three versions of a folder.
• The system the output would need to reach has no API.
• Nobody owns the process end to end, so no one can approve a change to it.

A good readiness assessment names which of these stands between you and each opportunity, and what it would cost to remove. Frequently the honest first project is not an AI project at all — it is tidying the document store or starting to record an outcome.

Pick one thing and do it

The best possible outcome of a strategy engagement is that one project starts within a month.

Nothing teaches a business what it can do with these tools like putting one into production and living with it — the accuracy, the exceptions, the running cost, the reaction from staff. Six weeks of that is worth more than another quarter of planning, and it makes the second decision far better informed.

Comparing five quotes

Read for:

• Who showed you a redacted example deliverable?
• Who committed to costs and returns against each opportunity?
• Who included a staff usage policy?
• Who was clear about whether they would bid to build?
• Who would be willing to recommend doing very little?

Describe what prompted the question, what your staff are already doing unofficially, and what decision the work is feeding. Consultants who do this well will ask what your people are already using before they ask about your strategy.