One NZ's Nokia AI deal: a staged model for autonomous operations

What One NZ announced

On 11 November 2026, Nokia announced that One New Zealand will deploy two more Nokia products, Nokia Assurance Center and Autonomous Networks Data Suite. Nokia will also provide R&D co-creation and consulting services. According to Telecoms.com, Assurance Center combines network, service, subscriber and security data and uses it to trigger automated operations processes. Data Suite catalogues network data in a standard format so AI-driven automation tools can use it.

The deal extends work One NZ started in 2025, when it deployed Nokia's Unified Inventory and Orchestration Center. Telecoms.com describes Unified Inventory as a real-time record of the network. It says Orchestration Center automates the design, delivery and lifecycle management of services across equipment from several vendors. The press release text says One NZ will use open interfaces and on-premises deployment options to keep control of its data, models and operating environment. Converge Digest reports that neither company disclosed the contract value or the implementation schedule.

What the autonomy levels mean

The stated goal is to move One NZ towards TM Forum Level 3 and Level 4 autonomous network capabilities. TM Forum is the telco industry body, and its scale runs from Level 0 to Level 5. TelecomTV's explainer describes Level 0 as manual operations and Level 5 as a fully autonomous network. Level 3 is a "conditional" autonomous network. Level 4 is a "highly autonomous" network that makes decisions from predictive analysis and runs closed-loop management of services, where the system detects a problem, acts on it and checks the result without waiting for a person.

Converge Digest points out that Level 3 and Level 4 are targets One NZ is working towards. Installing AI-enabled products doesn't by itself put an operator at a given level. The announcement contains no assessment results, no target date and no measures such as the number of use cases in production. That gap is common in vendor announcements, and buyers should expect it when they read their own suppliers' press releases.

Why the order of the work matters

The useful part of the One NZ story is the order of the work. The operator built an accurate inventory and an orchestration layer in 2025. In 2026 it's adding a data catalogue and AI-driven assurance on top of those systems. Nokia's head of autonomous networks, Shkumbin Hamiti, told Telecoms.com that the aim is for networks to sense changing conditions, make trusted decisions and act at machine speed. A machine can only make a trusted decision when the data behind the decision is current and complete. One NZ's 2025 investment in inventory provides that data.

Most mid-sized NZ organisations try AI in IT operations the other way round. They buy a service desk or monitoring tool with AI features, switch on automated triage or remediation, and then find their configuration management database (CMDB), the register of IT assets and how they connect, is a year out of date. An AI agent working from a stale CMDB will route incidents to the wrong team, or it'll restart a service that other services depend on, and nobody will know until something breaks. One NZ's sequence is slower to show results, but each later step rests on records the business already trusts.

Nokia's "glass-box" approach also deserves attention. Converge Digest describes it as transparency into network decisions and into how automated actions are governed. The same test works for any AI tool allowed to change production systems. An operations manager should be able to see which data the tool used, why it chose an action and who approved the level of autonomy it had.

What it means for NZ businesses

The direct effect for One NZ business customers is still unclear. Nokia says the tools should help One NZ anticipate network conditions and act before problems reach customers. However, One NZ hasn't published service targets or a rollout timeline tied to the deal. Firms that depend on One NZ connectivity can reasonably ask their account manager how the change will affect fault detection and notification. They shouldn't expect contractual improvements yet.

The indirect lesson matters more. TM Forum's levels give operations leaders a ready-made way to describe how much freedom an AI system has. A finance team, a service desk or a warehouse can rate each process from 0 to 5. A process at Level 2 might let AI suggest the category for an incoming ticket while a person confirms it. A process at Level 3 might let AI restart a known service within agreed conditions and log the action for review. Writing those levels down for each process gives the board, the IT lead and the vendor the same reference point, which most AI rollouts lack.

One NZ's insistence on open interfaces and on-premises options matters too. If an AI tool keeps your operational data and learned models inside the vendor's platform, changing vendors later means rebuilding the history the tool depended on. One NZ has written data control into its arrangement with Nokia, and smaller buyers can ask for the same terms at their own scale.

Practical next steps

• Check your records before you buy AI operations tooling. Take a sample of 20 assets or services and confirm the CMDB or asset register matches reality. If more than a few entries are wrong, fix the records and the processes that update them first. Our ITIL 4 service management work starts at this point.
• Give each process an autonomy level. Use a 0 to 5 scale based on TM Forum's and record which actions AI may take without a person, which need approval and which are off limits. Review the levels after an incident or a change of tool.
• Ask vendors for glass-box evidence. Ask to see an example decision log showing the data used, the action taken and how a person reverses it. A vendor that can't show one in a demo is unlikely to provide one in production.
• Settle data and model ownership in the contract. Confirm where models run, whether you can export your operational data and learned configuration, and what happens to both if you leave.
• Set your own milestones. One NZ's announcement didn't publish any, so don't rely on vendor claims. Track measures you already collect, such as mean time to restore, incidents resolved without escalation and changes rolled back, and compare them with your figures from before the tool went live.

Setting autonomy levels, governance and data terms before the first AI agent goes live is the work our AI consultancy team does with NZ and Australian organisations.