Auto & General's AI agent pilot: what NZ contact centres can copy

By DataForge team23 November 20265 min read

Auto & General is piloting Genesys AI agents on one low-risk, high-volume service workflow, after years spent consolidating platforms and data.

What Auto & General announced

Auto & General, the Australian insurer behind Budget Direct and the white-label policies sold under the Coles and Qantas brands, has started deploying agentic virtual agents in its customer service operation. iTnews reported on 20 November 2026 that the insurer is piloting Genesys' Agentic Virtual Agent (AVA) product in a single part of its service business. Kelly Lyons, head of unified communications for sales and service, described the plan at the Genesys Xperience APAC summit, which ran in Sydney on 11 and 12 November.

The first use case was picked on purpose. Lyons said the pilot targets work with low risk, high volume and well-understood workflows, and she hopes the pilot becomes a blueprint the insurer can reuse elsewhere in customer service. Auto & General hasn't published which service area the pilot covers, how many interactions the agent handles, what share it resolves without a person, or when a wider rollout might happen.

The groundwork came first

The groundwork behind the pilot matters more to NZ businesses than the agent itself. According to the iTnews account, Auto & General used to run voice, digital channels, workforce planning, quality evaluation and gamification on five separate platforms, with five vendors, five contracts and a set of integrations holding them together. The insurer replaced all five with one Genesys platform and moved support across in stages. All support transcripts now stream into the company's data lake for analysis.

Before trying an autonomous agent, the insurer put AI next to its human staff. Genesys' Cloud Agent Copilot writes conversation summaries automatically. A pilot group in pet health insurance saw after-call work drop enough to justify expanding the tool, and the larger service business now generates about 133,000 automatic summaries a month. Auto & General reports real full-time-equivalent savings, better customer satisfaction and net promoter scores, and more consistent notes. The size of those savings hasn't been published, so other organisations can't benchmark against the result.

The order is worth noticing. Auto & General consolidated its platforms, captured its conversation data, and gave staff an assistive tool with measurable output. Only after those steps did the insurer let software act on customers' behalf, and even then only on one narrow workflow.

Why the vendor's direction matters

Genesys is pushing hard on agentic contact centres. At its global Xperience event in September, the company launched an AI Control Plane that lets organisations set where AI agents may act, what data the agents can reach, and when a human has to step in. Genesys also added spec-driven development to AVA, which turns process documentation and call transcripts into virtual agent configurations. A routing product called Navigator, which decides whether an AI agent, an automated workflow or a person handles each contact, has a general availability window of November 2026 to January 2027.

The spec-driven approach explains why Auto & General's transcript archive is valuable. An AI agent built from documented processes and real conversations depends on the quality of both. An organisation whose procedures live in people's heads, and whose call recordings sit unanalysed in a legacy system, has little to feed an agent and no baseline to measure the agent against.

At the APAC event, Genesys described Australia and New Zealand as operating almost entirely in the cloud for contact centre technology. NZ organisations on cloud contact centre platforms will see agentic features arrive in vendor roadmaps whether or not they asked for them, which makes the readiness question more pressing.

What this means for NZ businesses

Most NZ contact centres are far smaller than Auto & General's, but the pattern scales down. A 30-seat service team at an insurer, utility, council or distributor faces the same questions about where an AI agent can safely act and how anyone would know whether the agent is working.

The fragmented platform problem is common here. Many mid-sized NZ organisations run telephony from one provider, email and chat from another, a CRM or ERP holding customer records, and quality checks in spreadsheets. An AI agent placed over that mix would have no single record of what happened in a conversation, and an incident review would mean stitching logs together by hand.

Assistive AI is the cheaper place to learn. Call summaries and suggested replies keep a person in control while producing hard numbers on handling time and after-call work. Those numbers become the baseline for deciding whether an autonomous agent is worth the risk. Auto & General's pet insurance pilot did exactly that job before the insurer expanded anything.

Choosing the first agent use case is a risk decision more than a technology decision. Password resets, policy document requests, delivery status checks and appointment changes tend to be high volume, rules-based and easy to reverse. Complaints, claims decisions, hardship conversations and anything touching vulnerable customers belong at the end of the queue. For NZ businesses, the Commerce Commission's recent AI guidance adds a reason for caution, since misleading statements from an AI agent still fall under consumer law.

We help organisations through this sequencing as part of our AI consultancy work, and the platform and data steps usually take longer than the AI pilot.

Practical next steps

  • List every system your service team touches in a typical customer interaction, and note which ones keep a searchable record of what was said.
  • Check whether your contact centre platform can export or stream transcripts to a store you control, and confirm where that data is hosted.
  • Run an assistive AI pilot, such as automatic call summaries, with one team for six to eight weeks, measuring after-call work and note quality before and after.
  • Write down your three or four highest-volume contact reasons and score each one for risk, reversibility and how well the process is documented.
  • Pick one low-risk workflow for any AI agent pilot, and define in writing when the agent must hand over to a person.
  • Ask your contact centre vendor what governance controls ship with its agentic features, what they cost, and whether they're available in your region and plan.
  • Document the processes the agent will follow before building it, using process mapping and automation methods that your team can maintain after the pilot ends.

Auto & General hasn't yet shown results from the AVA pilot, so the evidence is still incomplete. What the insurer has shown is a sequence: consolidate, capture data, assist staff, then automate a narrow task. NZ organisations can follow that sequence at their own scale without waiting for the outcome.

Want to know what this means for your organisation? Our team in Auckland can show you your top three opportunities in a free 30 minute assessment.

Get my free assessment How we help
Get new articles by RSS