Virgin Australia's 11 Oracle AI agents: lessons for NZ finance teams

By DataForge team17 November 20265 min read

Virgin Australia has mapped 11 Oracle AI agent uses in finance and procurement, with invoice scanning live. What NZ ERP owners can copy.

What Virgin Australia has done

Virgin Australia has worked out 11 detailed use cases for Oracle AI agents in its finance and procurement teams, iTnews reported on 16 November 2026. Only one of the 11 is in production. Document IO is an invoice scanning agent, and Sean Percy, the airline's finance and procurement product manager, told iTnews it has had a high success rate with invoice ingestion. Two question-and-answer agents are in user acceptance testing and two procurement agents are being assessed. The general ledger, accounts payable and expenses agents are still candidates. Percy said accounts payable is currently a very manual process at the airline.

The agent programme sits on an ERP platform Virgin Australia rebuilt first. According to iTnews, the airline moved from Oracle E-Business Suite to Oracle Fusion Cloud in 2023. It now runs Fusion Finance, Fusion Procurement, EPM account reconciliation and Oracle Integration Cloud. After go-live the airline had a very large backlog of incidents and enhancements. Virgin Australia brought in Tata Consultancy Services for Oracle application management, and the TCS team cleared that backlog before the airline started extending its processes and footprint. Oracle Human Capital Management is now being implemented.

Percy's most useful comment was about vendors. Every supplier Virgin Australia uses is adding AI, so the airline needs a defined position on which features it'll use, where they'll run and what makes sense to adopt. He called that position "probably the biggest thing for us at the moment."

Where Oracle's agent push stands

Oracle has been building agents into Fusion for more than a year. At AI World in October 2025 it released Payables, Ledger, Planning and Payments agents, and Computer Weekly reported that the pre-built agents integrate with Fusion ERP and EPM at no extra cost to customers. On 26 October 2026 Oracle announced Fusion Companion and Fusion Agentic Home. Both products let users define a goal, hand the work to agents and track progress. Oracle also released new agentic applications that run on its Fusion Claw runtime. Computer Weekly counted 25 Claw-powered applications at launch across ERP, HCM, supply chain and CX, covering work such as ledger, payroll and procurement. Oracle calls its model "governed delegation", meaning agents do the routine work and people keep the decisions that need judgement.

Oracle has spent a year adding agents, and Virgin Australia, a large Fusion customer, has put one agent into production after building a ranked list. We read that gap as a sensible pace. Vendors announce on a conference schedule, while customers have to absorb new agents into controls, audit trails and staff roles that already work.

What NZ businesses can take from the airline's approach

Few NZ mid-sized firms run Oracle Fusion. Most run Business Central, NetSuite, MYOB Acumatica, Xero with add-ons, or an older on-premise system. Virgin Australia's sequencing still applies to those firms, because their vendors are putting agents into the same finance processes: invoice capture, AP matching, expense claims and ledger queries.

Virgin Australia stabilised its ERP before adding agents. An invoice agent posting into a system with unresolved configuration issues and messy supplier records will produce the same errors faster, and staff will spend the saved time correcting them. A firm with an open incident backlog on its ERP, duplicated supplier masters or manual workarounds around the GST coding should clear those problems before switching on any agent.

Invoice ingestion is a sound first agent. The input is a document, the output is a draft transaction that a person approves, and the accuracy can be measured against what the AP team achieves now. A ledger agent that drafts adjusting journals carries more audit exposure, because errors flow straight into reported numbers. Virgin Australia's order follows that risk gradient, with invoice scanning live and ledger work still in the queue.

The airline's list of 11 use cases is also worth copying. Each use case needs a process owner, a defined data source, a measure of success and an approval rule. Without those four items an "agent pilot" usually becomes a demo that nobody owns once the vendor's consultant leaves. Percy said there's a large backlog of further ideas and the airline is starting small, which suggests the list gets ranked and worked through one agent at a time.

Some details remain unpublished. Virgin Australia hasn't released Document IO's accuracy rate, invoice volumes, cost or time savings, and there's no stated date for the other agents to go live. The story shows how one airline sequences the work. Anyone looking for proof of financial returns will have to wait for figures the airline hasn't published.

The vendor AI policy most firms lack

Percy's point about vendor strategy applies to most NZ organisations we talk to. A typical mid-sized firm has an ERP, a payroll system, an expense tool, a CRM and a document platform, and each supplier now ships AI features on its own release cycle with its own pricing and data terms. Few firms have written down which of those features they'll allow, who approves new ones, or where the data goes. A short policy per system, reviewed at each major release, stops agents from arriving by default through an update. Our AI consultancy page covers how we approach AI governance, and our ERP page covers platform health checks.

Practical next steps

  • Pull your ERP incident and enhancement backlog and count the items older than six months. If the number is high, schedule clean-up work before any agent pilot.
  • List every finance and procurement process where your ERP vendor has announced or released an agent, and note the release date, the licence cost and whether the feature is on by default.
  • Pick one document-driven process, usually supplier invoice capture, and measure current accuracy and handling time so any agent has a baseline to beat.
  • Write each candidate agent as a use case with an owner, a data source, a success measure and the approval step a person keeps.
  • Draft a one-page AI position for each major system and review the positions whenever a vendor publishes a major release.
  • Keep ledger-posting and payment agents behind human approval until the lower-risk agents have run cleanly for at least a full month-end cycle.

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