From data to decisions: building an AI-enabled maintenance capability at one of the world's most complex facilities

THE PROJECT

To develop AI-based digital tools to optimise planning, scheduling, and execution of maintenance activities onboard an offshore platform.

WHO WE’RE WORKING WITH

Secora is working alongside the client’s Engineering, Maintenance and Scheduling teams.

The operation is a deepwater facility in the Browse Basin off the northwest coast of Australia.

THE CHALLENGE

The client’s operation faces growing maintenance pressure. A large corrective maintenance backlog, the integration of a gas field, and the inherent complexity of a deepwater asset were combining to stretch the available workforce envelope. The challenge was not simply to do more — it was to do the right things, on the right assets, with the labour capacity already on hand.

WHAT SECORA DID

The program is being delivered in three phases:

Phase 1 established the analytical foundation, applying evidence-based lenses to the client’s Preventive Maintenance baseline through their Asset Information Factory (AIF) data environment. The analysis identified approximately 11,947 annualised hours of recoverable PM capacity — tasks consistently completed in less time than allocated, intervals that exceeded defect-justified frequencies, and operations delivering near-zero detection value. The finding established a credible, evidence-based pathway to reduce maintenance Rate of Effort while maintaining asset integrity and regulatory compliance.

Phase 2 extended this work in two directions simultaneously. On the analytical side, Secora assessed the corrective maintenance backlog across three structured focus areas — constrained and out-of-stock materials, overdue work orders carrying escalating risk, and work order duplication identified through ML-based similarity analysis — and designed a governance framework to protect any realised savings from gradual re-inflation.

On the technical side, Phase 2 delivered a functional AI prototype demonstrating that a Large Language Model can be given structured, controlled access to the client’s AIF data environment and can autonomously formulate database queries, retrieve maintenance information, and surface insights without manual analyst intervention. The prototype implements three operational pipelines — Z6 notification validation, AIF SQL query, and work order long-text search — and passed technical scrutiny from the client’s global IT and Digital reviewers. The program's emerging architecture has been recognised as broadly compatible with the client’s enterprise AI target architecture, positioning the local operation as a potential reference implementation for how AI is deployed responsibly across the client’s global asset base.

Phase 3, now underway, is transitioning the program from analysis and design into controlled pilot implementation — including governed SAP changes, a dynamic criticality capability, SCE re-validation, and AI pilot deployment within the client’s cloud environment.

THE RESULT

The overall program targets a reduction of 7,000+ hours of maintenance Rate of Effort and a return of $2.5M, alongside the structural capability uplift required to sustain evidence-based maintenance governance on an ongoing basis.

Critically, the program is not a one-time analytical exercise — it is building the infrastructure for maintenance to be managed as a dynamic, evidence-governed system rather than a fixed cost baseline.