Industrialising AI in asset-intensive industries: from ‘pilots’ to operational advantage

Executive teams across Energy, Utilities, Transport and the extractive industries have moved beyond the “should we do AI?” debate.  The strategic question now is more practical:

How do we industrialise AI safely and profitably, without creating another graveyard of pilots? 

The evidence is clear: AI can deliver meaningful operational and financial outcomes at scale, but only when it is treated as an operating discipline, not a technology experiment.  

This article looks at what’s working, what’s holding organisations back, and how Newton partners with organisations to unlock value quickly, safely, and in a way that scales. 

What the evidence says is working now 

1) Industrial value is real: when AI is anchored in operations 

One of the clearest demonstrations of scale comes from a southern US utility that deployed 400+ AI models across 67 generation units, reporting $60m in annual savings and a reduction of 1.6m tonnes of CO₂.  The headline matters less than the pattern: value came from embedding AI into operational routines, not from isolated dashboards.  

The Newton view: The organisations that genuinely benefit from AI, use it in a similar way to reliability engineering or continuous improvement, i.e. repeatable methods, clear governance, and sustained adoption. 

2) Predictive maintenance remains the “highest confidence” starting point 

Across industries, predictive maintenance shows some of the most consistent economic returns, particularly for rotating equipment (turbines, compressors, pumps, fans).  Our research over the past two years report benefits up to ~30% maintenance cost reduction and ~20% availability improvement when implemented well.  

What distinguishes success: 

Start with critical assets that have sufficient sensor coverage and history.

Integrate outputs into the CMMS/EAM workflow (alerts → work orders → feedback loop).

Use reliability engineers to validate early results and manage false positives.

3) GenAI’s early ‘killer app’ is knowledge retrieval – not autonomy 

In asset-intensive environments, productivity is routinely lost to searching for the right information: SOPs, manuals, standards, incident reports, P&IDs.  Retrieval-Augmented Generation (RAG) systems in industrial contexts have demonstrated strong retrieval performance for technical queries (e.g. 88% mean reciprocal rank) and for internal regulatory content (e.g. 91.62% recall).  

The Newton view: In 2026, the pragmatic path is to deploy GenAI where it reduces friction and improves consistency, while keeping humans firmly accountable for decisions, especially in safety-critical contexts.  

4) Capital delivery is being reshaped, if (and only if) data and governance are ready 

AI tools for schedule risk analysis are now trained on substantial historical datasets (e.g., 750,000 schedules representing >$2T in construction spend).  This creates a credible foundation for early detection of schedule risk patterns, but value is very dependent on data quality, process integration, and decision rights.  

The Newton view: Capital analytics can be a major lever, but organisations should validate ROI in context and avoid ‘vendor case study optimism’ without evidence from their own portfolio.  

The biggest constraint isn’t algorithms, it is adoption. 

The human factor is the determining factor in many cases: 

~85% of AI projects fail to deliver expected business value, often due to inadequate planning, training, and change management.

~80% of workers report receiving no GenAI training from their organisations.

Meanwhile, many asset-intensive organisations operate in brownfield reality: imperfect historic data, legacy OT constraints, and cybersecurity architectures not designed for modern connectivity.  

The Newton view: The difference between success and failure is usually not model selection. It is: 

• choosing the right initial portfolio 
• integrating into operational workflows 
• building capability and confidence in the workforce 
• governing risk without slowing delivery 

What ‘good’ looks like in 2026 and beyond: a portfolio approach and repeatable foundations 

The evidence supports an emerging blueprint: 

1. Start with a small portfolio of high confidence use cases
   Predictive maintenance and industrial knowledge retrieval are proven starters; capital delivery analytics can follow where data maturity supports it.  
2. Build reusable foundations once
   Data integration patterns from OT/IT sources, scalable MLOps, and governance that supports safe scaling.  
3. Default to human-in-the-loop for safety-critical decisions
   AI recommends, scores, retrieves, and explains; humans decide and remain accountable.  
4. Invest in training and change management as core programme work
   If most workers aren’t trained, adoption won’t scale regardless of model performance.  

Turning complexity into measurable operational impact 

Newton’s contribution is not more AI for the sake of AI. It is unlocking complexity for meaningful impact and helping clients industrialise improvements that already fit their reality, technically, operationally, and culturally. 

Tenet 1: Relentlessly focused on unlocking complexity for meaningful impact 

In our experience, many organisations that we work with already have access to the technological resources they need to solve the problems they face now.  The constraint is often not tools; it’s complexity: fragmented data, unclear workflows, ambiguous ownership, and competing priorities. 

That’s why we start by focusing on: 

• the operational problem and value pool (not the model) 
• where decisions happen (and who makes them) 
• how work gets done today (and what must change tomorrow) 
• what data exists and is “good enough” to start generating value now  

Crucially, most of our operational excellence work is scalable where there is an existing technology landscape.  We design improvements to be robust in brownfield environments rather than waiting for ideal conditions.  

What this looks like in practice:

• Selecting a small, high confidence use case portfolio (e.g., PdM and knowledge retrieval) 
• Designing the workflow integration (CMMS/EAM, operational routines, governance) 
• Using the client’s current platforms wherever possible, avoiding unnecessary platform churn  

Tenet 2: Guaranteed operational excellence results and client-owned outcomes 

Newton 100% guarantees our operational excellence results.  That guarantee shapes how we deliver AI-enabled improvement: 

• We focus on outcomes, not outputs (production availability, cost, risk, time, safety)
• We build platform-agnostic Minimum Viable Products (MVPs) or work directly on the partner/organisation’s platforms
• Owning the solutions so they can scale for impact in their environment, with their governance and their teams.

This matters because a major root cause of AI failure is ‘pilot purgatory’, i.e. working demonstrations that never become operational reality. The way out is to design for: 

• adoption (training, roles, behaviours) 
• scale (reusable data and model operations) 
• ownership (client capability and control) 
• governance (risk-managed acceleration, not slow committees)  

A pragmatic engagement pattern: 90 days to traction, 12 months to scale

First 90 days: create evidence of value and a foundation to scale:

• Rapid diagnostic: use cases, data readiness, adoption readiness 
• Deliver 1–2 MVPs (often PdM + knowledge retrieval) 
• Define the operating model: decision rights, governance, training approach 
• Establish “good enough” data plumbing and measurement  

First 12 months: scale a portfolio, not a prototype:

• Extend use cases across assets/sites 
• Embed MLOps and governance for repeatable scaling 
• Expand role-based capability building 
• Build internal momentum through visible results and frontline confidence  

Closing perspective: pragmatic optimism wins 

AI in asset-intensive industries is no longer speculative.  The advantage will go to organisations that treat AI as a disciplined operational capability, supported by foundations that enable scaling, and by a workforce that is trained, confident, and accountable. 

At Newton, our focus is simple: unlock complexity, deliver measurable, meaningful, impact, guarantee outcomes, and leave organisations and teams with solutions they can own and scale in their environment. Get in touch to find out more.