Key Takeaways
- AI is already deployed across Energy, Minerals and Resources (EMR) capital programmes - from Chevron's generative AI exploration platform to Shell's document automation that cut processing costs by over 85% - but 42% of enterprises abandoned most of their AI initiatives in 2025, up from 17% the previous year.
- The failure pattern is consistent: RAND Corporation research attributes over 80% of AI project failures to governance problems - misaligned problem definition, inadequate data, absent business ownership - not technology shortcomings.
- Data preparation consumes up to 80% of AI project timelines according to the PMI-CPMAI framework, yet 63% of organisations either lack or are unsure whether they have the data management practices to support AI.
- The EU AI Act's high-risk regime takes full effect on 2 August 2026, with penalties up to 3% of global annual turnover, - and EMR operators using AI for pipeline integrity, predictive maintenance on safety-critical assets, or energy infrastructure management will need to demonstrate conformity.
- Organisations that embed AI governance at C-suite level are 2.6 times more likely to report material EBITDA uplift from AI than those that treat it as a technology initiative.
Artificial intelligence is no longer a future consideration for Energy, Minerals and Resources (EMR) capital programmes. It is already there - and in some cases, producing measurable results. The question facing Boards, Project Directors and PMOs is not whether AI will enter the capital programme. It has. The question is whether the governance architecture around it is adequate to ensure it delivers value rather than introducing new, poorly understood risks.
Where Is AI Working Well in EMR Capital Programmes?
There are genuine, verifiable AI deployments producing results across the EMR sector and they tend to cluster around five use cases.
Subsurface exploration and discovery. Chevron's ApEX platform, launched in August 2024, uses generative AI multi-agent search across exploration data, first deployed in deepwater Gulf of America operations. Rio Tinto has partnered with Fleet Space Technologies for AI-enhanced ambient noise tomography at its Rincon lithium project and with Ideon Technologies for muon tomography combined with AI to accelerate critical minerals discovery. The IEA's Global Critical Minerals Outlook 2025 quantifies the potential: AI-based geological exploration can reduce drilling costs by up to 60% and increase discovery success rates by as much as four times.
Engineering document intelligence. Woodside Energy's long-running deployment - originally built on IBM Watson, now extended with watsonx.orchestrate - ingested 33,000 technical documents from prior projects to support engineering decisions on current programmes. Shell's managed information service with Wipro on a North American chemical capital project processed approximately 600,000 documents, automated 70% of manual work, and cut document-processing cost by over 85%.
Capital project scheduling and risk assessment. McKinsey notes that capital project teams are increasingly using generative AI for schedule generation and rapid scenario assessment when unexpected issues arise. IPA (Independent Project Analysis) has built machine learning performance prediction tools drawing on its proprietary database of capital project benchmarks.
Predictive maintenance on capital assets. Woodside's "Maint Intel" tool at the North West Shelf project, developed with AWS, reduced model processing time from five days to under two hours.
Integrated project delivery platforms. Rio Tinto's Kemano T2 tunnel project - a 7.6 km hydropower tunnel for the Kitimat aluminium smelter - used Palantir Foundry with ML vision models and achieved a 150% improvement in excavation rate per day following COVID-related disruption.
These are real deployments with named operators and quantified outcomes. But they represent the minority. For the majority of organisations attempting AI, the picture is very different.
The Scale of AI Project Failure
S&P Global Market Intelligence's 2025 survey of over 1,000 enterprises found that 42% had abandoned most of their AI initiatives, up sharply from 17% the previous year. On average, organisations scrapped 46% of proofs-of-concept before reaching production. RAND Corporation's research puts the overall failure rate above 80% - roughly double the rate for conventional IT projects - with a third of AI projects abandoned before production, and nearly half of those that do ship delivering no measurable business value.
Gartner predicts that over 40% of agentic AI projects will be cancelled by the end of 2027, citing escalating costs, unclear business value and inadequate risk controls. They also warned of widespread "agent washing", where vendors rebrand existing automation products as AI agents without substantive capabilities.
The gap between what the best-performing Energy, Minerals and Resources operators are achieving and what the broader market is experiencing is not a technology gap. It is a governance gap, and it is widening.
Why Do AI Projects Fail Differently from Traditional Capital Projects?
AI projects are not failing for technology reasons. They^re failing for governance reasons, but in ways that existing stage-gate and PMO frameworks were not designed to catch.
Probabilistic outputs versus deterministic acceptance criteria. Stage-gate reviews are built to accept deliverables that meet specification. AI models produce probability distributions, not fixed deliverables. Without governance that defines acceptable confidence thresholds and human-validation triggers, gate reviews default to "the demonstration worked", which is not the same as "the system is ready for production."
Data dependency at a scale traditional projects do not face. PMI's CPMAI framework identifies that data preparation alone can consume up to 80% of total AI project execution time. Yet Gartner's survey of 248 data management leaders found that 63% of organisations either lack or are uncertain whether they have the data management practices required to support AI. This is a governance gap, not a technology gap.
Model drift over the asset lifecycle. EY flags model drift - AI decisions losing precision as new operating data arrives - as a persistent failure mode in oil and gas, particularly when cloud system maintenance and continuous monitoring are under-resourced after go-live. Traditional capital project governance ends at handover. AI governance cannot.
Absent business ownership and misaligned incentives. RAND's research identifies leadership shortcomings - inflated expectations, poor problem framing, premature abandonment - as the leading root cause of AI failure. NTT DATA's June 2025 survey of 2,300 senior decision-makers found that 72% of organisations still lack a formal generative AI usage policy, despite 99% of C-suite executives planning further AI investment.
What Does Good AI Governance Look Like for a Capital Programme?
The organisations that are extracting real value from AI share a common feature. McKinsey's State of AI 2025 report found that high-performing organisations - those reporting material EBITDA impact - are three times more likely to have senior leaders demonstrating ownership and commitment to AI governance, and 2.6 times more likely to achieve material returns when that governance operates at C-suite level.
PMI's research paints an equally stark divide. Organisations using AI in over 50% of their projects reported 91% success on quality management versus 40% for organisations with under 15% adoption. On cost management, the gap was 85% versus 42%. But the critical variable was not adoption volume, it was whether governance structures had adapted to accommodate AI's unique requirements.
For EMR capital programmes, five governance adjustments matter most:
- Add an AI-readiness gate before FEL-2. This should require evidence of data availability and quality, a model-validation plan, drift-monitoring commitments, human-validation thresholds, and - critically - an EU AI Act high-risk classification assessment for any system that may be deployed in or sold into the EU.
- Assign named executive ownership for every AI capability. McKinsey's data shows this is the single largest correlate with positive EBIT outcomes. AI without an accountable business owner is AI that nobody governs after the pilot team moves on.
- Treat data as a capital-readiness asset. Programmes that succeed earmark 50-70% of AI timeline and budget for data readiness. Programmes that fail treat data preparation as a task to be completed quickly so the "real work" of model building can begin.
- Extend PMO assurance beyond project handover. Model drift, bias and accuracy monitoring on safety-critical applications - predictive maintenance on rotating equipment, pipeline integrity, emissions monitoring - require continuing governance that traditional handover processes do not provide.
- Build AI project management capability. PMI's research indicates that only around 20% of project managers report substantive experience with AI tools, while nearly half have no experience at all. PMI's CPMAI certification exists specifically because traditional project management frameworks do not address the data, ethics, drift and operationalisation phases that AI projects require.
The Regulatory Clock Is Running
There is one more reason AI governance cannot wait. The EU AI Act entered into force on 1 August 2024. The full high-risk regime - including obligations for AI systems used in energy infrastructure management, pipeline integrity monitoring and other safety-critical EMR applications - takes effect on 2 August 2026, with penalties up to EUR 15 million or 3% of global annual turnover. For any EMR operator with European operations, assets or supply chains, this is not a distant compliance exercise. It is a deadline that sits roughly ten weeks from publication of this article.
The European Commission also opened consultations in August 2025 on a Strategic Roadmap for Digitalisation and AI in the Energy Sector, signalling that sector-specific requirements are likely to follow. And the IEA's Energy and AI special report confirmed that the energy sector is not yet taking full advantage of AI's potential, citing insufficient digital skills and data availability as the primary barriers - exactly the governance gaps this article describes.
The Bottom Line for EMR Boards and Project Directors
AI is entering capital programmes whether organisations are ready or not. The vendors are already embedded. The pilot projects are already running. In many cases, project teams are using AI tools - schedule optimisers, document AI, estimating assistants - without formal governance oversight. That is the real risk: not that AI will fail spectacularly, but that it will fail quietly, producing outputs that look plausible but erode the reliability of the decisions being made around them.
The organisations that will extract genuine value are the ones treating AI governance with the same rigour they apply to capital project assurance, cost control and safety management. As a governance discipline, one that requires board-level ownership, structured assurance and independent verification.
PDAS provides independent governance and assurance for EMR capital programmes, including review of AI integration readiness, data governance and compliance positioning. If your organisation is deploying AI into capital project delivery and wants independent assurance that governance is keeping pace with technology, we can help.
