Key AI Themes from Aker Systems GRID Roundtable 2026

Aker Systems recently engaged with industry experts at the GRID 2026 event in London.

We asked our panel, assembled from across the Energy supply chain, to help us understand:

- What have your own experiences been with AI deployment to date ?

- Where do you see the most immediate and impactful opportunities for data and AI ?

- What are the primary technical and organisational issues holding back broader AI deployment ?

- What specific industry initiatives, supply chain collaborations, or technology solutions are needed ?

Our thanks to the panel and for creating a highly engaging and interesting debate.  We can now present aseries of themes and summary from the discussion – based on our topic of: Are operational leaders capturing real value from AI, or does fragmented data limit its potential ?
‍

1. AI is helping to solve operational problems, but should not be treated as an end in itself

A consistent theme was that organisations should begin with a defined business or engineering problem and then determine whether AI is the appropriate solution. Our panel member operating in energy and utilities consulting described this as a deliberate principle: discuss client problems in industry language rather than attempting to sell “agentic AI.” An industry expert on AI problem-solving similarly cautioned that machine learning, generative AI, statistical analysis and agentic systems solve different problems and should not be grouped together indiscriminately.

Benefits: Investment remains linked to measurable operational outcomes.

Limitations: AI enthusiasm can lead organisations to deploy sophisticated models where conventional statistics, automation or better data management may be simpler, more transparent and sufficient.

Action: We recommend prioritising and aligning right-sized, AI-ready infrastructure and capability based on budgets, operational need and assessment of the problem ahead of the solution.

2. Human judgement and assurance remain essential

Participants consistently positioned AI as decision support rather than an autonomous decision-maker.

·        In complex organisations, users create different narratives from the same AI-generated output.

·        At a fundamental level, users of AI observed simple transcription errors in generated outputs – meaning draft AI outputs can only be treated as draft and need human input.

·        One of our Transmission experts emphasised validation of machine-learning outputs and expects network planners to determine the appropriate intervention.

·        Senior professional judgement must be used as the controlling layer in AI-augmented consulting.

·        Several attendees advocated for maintaining user intent and technical critical thinking.

·        Understanding that sometimes AI is not the answer - use deterministic and repeatable methods where they can adequately solve the problem.

Benefits: AI can process more information and surface patterns more quickly while specialists provide contextual interpretation.

Limitations: Quality assurance, explainability and accountability become increasingly important when outputs affect safety, regulatory compliance, costs or live network operations.

Action: Identify the critical roles that will act as decision makers for AI adoption and operational usage, then empower those roles to be accountable for data and AI in production.

In the context of the Energy industry supply chain, we see initiatives such as larger Transmission and Distribution organisation AI Control Plane rollout being critical to a wider industry standard on safe, ethical, orchestrated wider adoption of AI technologies.

3. Data quality and interoperability are the principal constraints

The strongest shared conclusion was that the main obstacle is usually not the AI model but the underlying data. Across our panel the same themes emerged:

·        Data integration challenges dominate organisations that grow by acquisition, leading to data challenges for AI models across numerous enterprise systems

·        Inconsistency intransmission-data standards between providers andoperators leading to challenges with inputs to AI processes for sensor dataaccuracy and network models

·        Essential to distinguish between large quantities of low-cost data and smaller quantities of reliable, repeatable data that compound in value over time.

·        Emphasis on data availability, sharing, quality and analytical capability

·        Focus on highlighted metadata, provenance, naming standards, security and long-term stewardship, drawing on experience from defence.

·        Across the utility sector, GIS, SCADA, asset-management, imagery and vegetation datasets often remain in separate organisational and technical silos.

Benefits: Better-integrated data improves decisions even before AI is introduced and creates a reusable foundation for multiple future applications.

Limitations: AI cannot reliably compensate for missing provenance, inconsistent formats or inaccessible data. Valuable models may remain trapped in pilots if systems cannot exchange trustworthy information.

Action: Aker Systems regularly encounters and addresses all of the above points in its work with Energy and wider Central Government customer AI programmes. We strongly support the promotion of data health, accessibility and interoperability in all of our customer engagements.

4. Regulation and investment incentives shape adoption

For regulated energy companies, technical feasibility does not automatically create an investable proposition. Participants noted that:

·        Expenditure must usually be connected to recognised regulatory outcomes such as reduced capital costs, improved flexibility, reliability or faster infrastructure delivery.

·        Broader aims - such as an autonomously managed grid - may be too diffuse for an individual network operator to fund.

·        Smaller applications are more likely to proceed where they have a self-contained business case. Participants suggested that industry-level organisations, shared standards or government-backed mandates may be needed to address common data and interoperability problems.

·        Ofgem’s ED3 process and associated digital-engineering expectations were seen as potential drivers of more coordinated investment.

Benefits: Regulation can create common requirements and make digital investment part of infrastructure funding.

Limitations: Without a direct and measurable connection to regulated outcomes, preventive investment in data to support AI implementations may be deferred until a major failure creates urgency.

Action: Organisations should share and promote joint-working on industry-level AI initiatives that solve specific problems and not wait for whole-system solutions to manifest via regulation.

We explored further into specific industry areas to capture and highlight use-cases for AI adoption and the current market feedback on AI implementation progress.

Predictive maintenance and asset reliability

AI is strongly linked to the problem of maintaining ageing, geographically distributed energy infrastructure.

·        Power and water utilities: Engineers and data scientists used forecasting, scenario analysis and asset data to help a distribution company move from reactive and schedule-based maintenance towards predictive maintenance. The reported outcome was significant capital and operating-cost savings.

·        Transmission predictive maintenance: Autonomous drones collect consistent images and LiDAR data from transmission towers. The repeatability of the data makes it possible to measure and predict component wear, corrosion and insulator or shackle deterioration.

·        Transmission and distribution networks: Machine learning classifies LiDAR, imagery, GIS and network data within physics-enabled digital twins. This supports storm modelling, asset-risk analysis, capacity planning and vegetation management.

·        Investment in renewable-energy assets: Better use of operational and maintenance data was discussed as a means of improving maintenance after contractor warranties expire and activities are brought in-house.

Benefits: Earlier identification of failure risks, more targeted inspections, reduced downtime, lower maintenance expenditure and better prioritisation of capital investment.

Limitations: Predictive models depend on consistent historical data. Cheap, irregular or differently framed inspection images maybe adequate for a one-off assessment but unusable for measuring change overtime. Human engineers must still decide what action to take.

Forecasting and optimisation

AI and advanced modelling are being used where decisions involve many complex variables.

·        Major engineering projects: Predictive workforce models estimate the availability, cost and movement of specialist labour up to 18 months ahead. One panel member reported significant and accurately measurable seven-figure savings in agency and workforce costs, full supply-chain fulfilment and attrition below 1%. This was not possible without AI-augmented reporting.

·        Offshore Wind: Automated project modelling explored how turbine prices and other variables affected simulation of energy costs and auction-bid strategy.

·        Energy policy and whole-system energy planning: Hyper local forecasting was cited as an example of AI supporting energy-system balancing and renewable-generation planning.

·        Digital Twins: simulations allow network operators to explore “what-if” questions involving storms, flooding, asset failures and future capacity.

Benefits: Better anticipation of constraints, more informed bidding and investment decisions, improved resource planning and the ability to test scenarios before committing money.

Limitations: Forecasts can create false confidence if assumptions, input quality or uncertainty are not understood. In some cases, deterministic modelling may be more repeatable and explainable than AI.

Engineering, construction and project-delivery productivity

Several companies are using or exploring AI to reduce repetitive workand increase delivery capacity.

·        Specialist energy solutions: Early applications include Microsoft Copilot for design review, design validation, project-delivery support and process optimisation.

·        Utilities / civil infrastructure: AI is being explored for safety communications, drawing take-offs, estimating and report preparation. For a smaller company, it could provide capabilities that would otherwise require a larger team.

·        Substations and data centre infrastructure: A cross-functional adoption initiative is examining copilots and agents as project-delivery tools, while retaining engineers’ critical thinking.

·        Consulting models: AI supports a lean consultancy model based on senior specialists rather than the traditional junior-heavy consulting pyramid.

·        Recruitment and skills management: AI monitors companies and labour-market signals to identify recruitment pressures, persistent vacancies and potential workforce opportunities.

Benefits: Faster document/evidence production and review, reduced administrative effort, lower delivery overhead, greater capacity for small teams and more efficient consulting models.

Limitations: Prompting can cause outputs to drift from the original intention. Generated reports may contain incorrect figures, unsupported references or a plausible narrative that is not supported by the evidence. Benefits are therefore easier to realise in drafting than in safety, engineering- or commercially critical decisions.

AI is also creating new infrastructure problems

The relationship between AI and the grid runs in both directions. Our panel member’s data centre and substation experience highlighted that AI workloads are increasing demand for data-centre capacity, grid connections and electricity supply. Another expert also noted the feedback loop in which more AI processing requires more data-centre energy.

Benefits: AI can help plan and deliver the infrastructure required to support growing demand.

Limitations: AI adoption itself contributes to power demand, capacity constraints and pressure on already stretched grid-delivery resources.

Conclusion

The roundtable indicates that AI adoption across energy, utilities and infrastructure is still at an early operational stage. The strongest use cases are those that connect AI directly to:

1. Predicting asset deterioration and failure.
2. Forecasting labour, costs, generation and network conditions.
3. Simulating infrastructure scenarios through digital twins.
4. Accelerating engineering, reporting and project-delivery work.
5. Classifying and extracting value from large inspection datasets.

Benefits:  Lower costs, improved reliability, faster analysis, better resource allocation and increased organisational capacity.

Limitations: Poor or fragmented legacy data, limited interoperability, weak analytical skills, uncertain assurance, regulatory constraints and difficulty demonstrating value before operational failures occur.

Actions: Invest in accessibility, interoperability and prioritisation of human expert skills to re-organise teams focussed on measurable outcomes using AI-enabled solutions.

The overarching message is that AI should follow the problem: define the decision that needs to improve, establish trustworthy and interoperable data, retain accountable human judgement, and use the simplest analytical method capable of producing the required outcome.