Short answer
An AI implementation roadmap is a dated delivery sequence that says which AI systems will be built or rolled out, in what order, and what has to be true before each one starts: data access, system changes, policy and trained people. Levy Street writes roadmaps for New Zealand organisations that gate each step on readiness evidence, so work begins only when its dependencies are met.
- A roadmap sequences delivery; a strategy decides what is worth doing.
- Each item has a start gate and an exit measure, not just a date.
- Foundations such as policy, accounts and training usually run first and in parallel.
- Running costs are budgeted per item, not only build costs.
What a roadmap adds to an AI strategy
A strategy decides which AI uses are worth doing. A roadmap decides when each one starts, what must be true first, and how you will know it has finished. Without that second step, organisations end up with a ranked list and no movement. Datacom’s 2026 State of AI Index, a vendor survey of NZ organisations with 100 or more staff, found 81% still in the exploratory or implementation stages, and only 20% with a dedicated AI budget and clear cost tracking.
If you do not yet have a ranked list, start with AI strategy and advisory or an AI readiness assessment.
Who needs one
- You have priorities but no sequence. The strategy is approved, and now every team wants its item first.
- Dependencies are tangled. One AI use needs a system upgrade, another needs a data clean-up, a third needs a policy decision, and nobody has mapped the order.
- Budget has to be phased. Finance needs to know what is spent when, including running costs after launch.
- Previous plans slipped. An earlier AI plan had dates but no gates, and work started before access or data were ready.
A sample roadmap
Illustrative example: a 120-person New Zealand wholesale distributor on a cloud ERP, with an approved strategy naming four priorities. Phases overlap deliberately.
| Phase | Weeks | Workstream | Gate to start | Exit measure |
|---|---|---|---|---|
| 0. Foundations | 0–6 | AI use policy, business AI accounts, training for 40 office staff | Leadership approval | Policy adopted; staff trained; personal accounts retired |
| 1. First production system | 4–14 | Supplier invoice extraction into the ERP | Access to the accounts payable inbox; 200 labelled invoices; privacy check done | Agreed field accuracy on a held-out set; accounts team using it daily |
| 2. Data track | 4–20 | Product master data clean-up for the top 2,000 lines | Owner named in purchasing | Duplicate and incomplete product records below the agreed level |
| 3. Second system | 14–26 | Customer delivery enquiry assistant, read-only at first | Phase 1 live for four weeks; courier tracking API access | Answer accuracy and escalation targets met in supervised use |
| 4. Scale and review | 26+ | Monitoring, quarterly re-score, next candidates | Two systems in production | Running costs within budget; next item approved |
Note what is missing from the first six months: a customer-facing chatbot. It scored well on value, but its source content is spread across five systems, so it waits for the data track.
Readiness gates
Before any build on the roadmap starts, these should be true. Items that fail a gate stay on the roadmap with a task to close it, rather than starting anyway.
- A named business owner who will use the result and answer for its metric
- A baseline of the metric, measured on real volume, not estimated
- Data access tested: a real export or API call returned the fields the system needs
- A privacy check done, and a privacy impact assessment where personal information is involved
- Affected staff told and, where roles may change, consulted
- An evaluation set of real examples with agreed correct answers
- Running costs budgeted for at least the first year
What you receive
- Dated sequence
- Items in order with start windows, durations and overlaps, detailed for six months and outlined for the next six.
- Gate and exit criteria
- For each item, what must be true to start and how finished is measured.
- Dependency map
- Data, system, vendor, policy and people dependencies, each with an owner.
- Phased budget
- Build and running costs per item, by quarter, in NZD.
- Capacity plan
- Who does the work, internal or external, and when key people are needed.
- Review rhythm
- A quarterly review format and the triggers for re-planning early.
How we build the roadmap
- Discover: start from the priorities. Take the strategy or priority list and check each item’s feasibility on real examples, so the sequence is not built on untested assumptions.
- Discover: map dependencies. Walk each item back to what it needs: which systems, which data, which approvals, which people, and which of those are already in hand.
- Prove: sequence and gate. Order the work so early items build the foundations later ones need, and attach a gate and exit measure to each.
- Prove: cost and capacity. Put build and running costs against each item, and check the people named actually have the time.
- Scale: run the review. We facilitate the first quarterly review, then hand the format to your roadmap owner.
What drives the cost
Roadmap work scales with the number of priorities, how much feasibility testing they need, the number of systems and vendors involved, and how many business units must agree the sequence. A roadmap for one division with an approved strategy is modest; one that must reconcile competing priorities across regions is larger. We scope and price it before starting. For the items themselves, see what custom AI projects cost and the AI ROI calculator. Our free AI roadmap template uses the same structure if you want to start alone.
Sequencing in practice
Products we work on are built in sequence too. 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. Blackpearl launched it in March 2025 and released a rebuilt Bebop 2.0 on 26 August 2026: ship a first version, learn from real use, then rebuild the core. The Property Management Agent for APL Property followed the same logic at a smaller scale, starting with routine administration and keeping decisions with staff.
Where roadmaps go wrong
- Dates without gates. Work starts because the calendar says so, then stalls waiting for access.
- Everything in parallel. Five pilots at once means none gets the attention of the people who matter. Overlap deliberately, not by default.
- Build budget only. Systems are funded to launch and then starved of monitoring and support. See managed AI operations.
- Betting on unannounced features. Microsoft’s April 2026 update on in-country Copilot processing does not include New Zealand. Plan on what vendors offer here now, not what they offer elsewhere.
- Never re-planned. A roadmap that has not changed in six months is not being used.
New Zealand timing to plan around
- Consultation. Where AI may affect roles, the Employment Relations Act 2000 requires employers to share information and give staff a chance to comment before deciding. Schedule it before the build, not after.
- Privacy. The Privacy Commissioner expects a privacy impact assessment before AI tools are used. Items that gather information about people from other sources also carry the IPP 3A notification duty in force since 1 May 2026, one of the 14 information privacy principles.
- Financial year. Many NZ organisations budget to a 31 March balance date, so we align phase budgets with your planning cycle.
Sources
- 2026 State of AI Index: AI uptake at 91%, leadership lags, Datacom, 2026
- State of AI Index 2026: research report and methodology, Datacom, 2026-08
- Microsoft offers in-country data processing to 15 countries for Microsoft 365 Copilot, Microsoft 365 Blog, 2025-11-04, updated 2026-04-03
- What’s New in Bebop 2.0, Bebop (Blackpearl Group), 2026-08-26
- Employment Relations Act 2000, New Zealand Legislation, version as at 10 July 2026
- Enhancing the Privacy Act (IPP 3A), Ministry of Justice, 2026-05-01
- Artificial intelligence and the Information Privacy Principles, Office of the Privacy Commissioner, 2023-09
Written by the Levy Street team in Wellington. Last updated 19 October 2026.