AI strategy and advisory.

When an AI strategy is the right purchase

Buy an AI strategy when you have more candidate AI uses than you can fund and leadership needs to choose between them on evidence. If you already know the one workflow you want to change, skip the strategy and go to AI consulting or a scoped proof of concept.

The organisations that get the most from strategy work are usually in one of these positions:

• The board has asked for an AI plan. Directors want to know what the organisation is doing about AI, what it costs and what could go wrong, and a list of tools is not an answer.
• Many pilots, nothing in production. Several teams have tried something on their own. None of it is shared, owned or measured, and nobody can say which experiment deserves real money.
• A budget round is coming. You need a defensible set of numbers for next year’s plan, including running costs, not just build costs.
• The work is changing anyway. A new ERP, a merger or a restructure is underway, and AI should be designed into the new process rather than bolted on later.

This is common. Datacom’s 2026 State of AI Index, a vendor survey of 207 senior leaders at NZ organisations with 100 or more staff, found 91% use some form of AI but only 15% are scaling it organisation-wide.

What the strategy contains

The output is a set of decisions with the evidence behind them, not a slide deck about the future of work.

Opportunity register

Every candidate use gathered from leadership and frontline interviews, written as a workflow with a trigger, inputs, a decision and an output.

Scored portfolio

Each candidate scored on the framework below, with the arithmetic visible so you can change a weight and watch the ranking move.

Build, buy or configure call

Whether Microsoft Copilot, ChatGPT, Claude or an industry product covers the need, or why a custom system is justified.

Sequenced plan

The first three to five items in order, with dependencies such as data access, system upgrades and policy made explicit.

Operating model

Who owns each system after launch, how outputs are reviewed, and who is called when something goes wrong.

Policy and risk baseline

A team AI use policy, privacy notes against the information privacy principles, and a risk register entry for each item.

Handoff pack

For the top item, a brief with acceptance criteria, a success measure and running costs in NZD, ready for a delivery team.

How we score candidate uses

We score each candidate from 1 to 5 on five criteria and weight them. The weights are agreed with decision makers before any scores are shown, so nobody can reverse-engineer a favourite project to the top.

Illustrative example: a 60-person Wellington engineering consultancy has three candidates. Proposal drafting scores 4, 4, 4, 4 and 5, which gives 1.2 + 1.0 + 0.8 + 0.6 + 0.5 = 4.1. Timesheet coding scores 3, 4, 3, 4 and 3 for 3.4. A public website chatbot scores 3, 3, 2, 2 and 2 for 2.55, mostly because its answers would reach clients unreviewed and its source content is scattered. Proposal drafting goes first; the chatbot waits until the knowledge it would answer from is in one place.

How the engagement runs

1. Discover: collect candidates. Interviews with leadership, then short working sessions with the teams who do the work. We collect real emails, forms and reports, not descriptions of them.
2. Discover: test the hard parts. For the top five to eight candidates we prototype the riskiest step on your own examples and record how often the output is right.
3. Prove: score and sequence. A decision workshop agrees weights, reviews scores and picks the order. Dependencies and owners are written down in the room.
4. Prove: hand off the first item. The top candidate becomes a brief for the narrowest production system that can move its metric, ready for implementation planning or a build.
5. Scale: re-score each quarter. Models, prices and your priorities change. After each delivery we update the scores and the plan, so the strategy stays a working document.

What drives the cost of strategy work

The main drivers are the number of teams and workflows in scope, how many candidates we test on real data, and how many stakeholders need to agree. A single division with a clear sponsor is a much smaller piece of work than a multi-site organisation with a board approval step. We do not publish rates; we fix the scope and price once we understand the organisation.

The strategy is the cheaper part. What matters more is what the recommended systems cost to build and run, which is usually dominated by integration and human review rather than model fees. Our guide to what custom AI projects cost shows the arithmetic, and the AI ROI calculator lets you test a candidate before it reaches the scoring table.

What we have built that shapes the advice

A strategy is only as good as its feasibility scores, and those come from building things:

• MOGGED Digital is Levy Street’s AI-native Wellington digital studio for websites, SEO and ads, which says it has served 100+ businesses, mostly in New Zealand. We run AI inside our own operations, so our advice on adoption and review comes from practice.
• 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 Property Management Agent for APL Property handles routine administration for a regional NZ property firm, with staff keeping the decisions.

Where AI strategies go wrong

• Written by people who will not build it. Items look good on paper and fail the first feasibility test. Ask who tested the shortlist, and on what data.
• Too many priorities. We recommend no more than three items in delivery at once for most mid-sized organisations.
• A vendor’s catalogue in disguise. If every recommendation is one supplier’s product, the strategy was a sales process.
• No owner after launch. Systems without a named business owner drift, then get switched off.
• Policy mistaken for strategy. An AI use policy is necessary, but it decides what is allowed, not what is worth doing.

Do not buy a strategy if you are small with one obvious workflow, or if the real gap is skills. In that case start with AI training for your team and one well-run pilot.

New Zealand context the strategy accounts for

• Policy. The current Government’s AI Strategy (MBIE, July 2025) takes a light-touch approach and relies on existing law, and MBIE’s voluntary Responsible AI Guidance for Businesses sets out the risks to consider. Both are dated as at October 2026; an election year can change policy.
• Privacy. The Privacy Act 2020 now has 14 information privacy principles, including IPP 3A on indirect collection, in force since 1 May 2026. The Privacy Commissioner expects a privacy impact assessment before an AI tool is used on personal information.
• Data location. No major AI vendor offers model processing inside New Zealand as at October 2026. Data can be stored here, and Amazon Bedrock’s AU inference profile keeps Claude processing within Australia and New Zealand. Each item in the plan records where its data goes.
• People. Where AI changes roles, employment law requires consultation in good faith. The sequence allows time for it rather than treating it as a launch-week task.

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