Decommissioning is a significant but unavoidable expense, theoretically offering no revenue potential but significant liability risks. Globally, there are an estimated 2,600 offshore oil and gas platforms due to be retired by 2040, translating into around $210bn in cost for operators and governments.
The compliance process is notoriously long and arduousness. Accuracy is impeded by legacy data, which is fragmented and analogue, often stored in handwritten paper documents in decades-old filing systems. Innes Grant, chief operating officer at AI-for-decommissioning company rahd.AI, describes this “unstructured data” as a fundamental decommissioning challenge.
As a result, AI is finding a convincing use case in the digitalisation of documents and derisking of application processes. Chevron has developed ApEX, its proprietary AI platform to simplify data retrieval, while Halliburton’s Boots & Coots platform uses AI to identify at-risk wells and enable compliance auditing.
Historically, decommissioning has also represented a blot on operators’ environmental, social and governance (ESG) strategies. Ageing assets present safety risks – decades of exposure to seawater means platforms experience corrosion – while emissions from vessels and seabed disturbance harm operators’ environmental profile.
AI is finding use cases for ESG-compliance too. Algorithms can assess human risk and provide continuous, intelligent surveillance, while digital twin technology enables operators to model and predict safety and environmental risks. For example, major operators such as Repsol use Baker Hughes’ Leucipa – an automated field production platform – to pull data from operational systems and build digital twins of production assets.
AI is changing the shape of oil and gas decommissioning, providing cost, labour- and time-saving opportunities across the value chain. As Aditya Singh, founder and CEO of decommissioning company Promethean Energy, puts it: “Technology, data and AI best practices can be used to decommission fields safely, efficiently and at the lowest possible cost, with regulatory compliance and environmental responsibility.”
Accuracy and accessibility: AI digitalisation for legacy data
Oilfields can operate for decades; the UK Continental Shelf, for example, has abandoned wells dating back to the 1960s. Legacy data is often stored in manual, paper filing systems organised by now-retired staff. “What that means is that we don’t have the institutional knowledge or continuity in terms of people,” says Singh.
The first challenge therefore comes in the form of identifying and accessing well data, required for plugging and abandonment (P&A) planning. Singh explains that when data is very spread out, it takes weeks, if not months, to plan how to decommission a field.
This issue is exacerbated by business changes, mergers and acquisitions. Well stock information made and stored under one owner might not have been transferred to new data management systems. The problem grows when decommissioning is contracted out and data finds itself again in new hands.
“That is where P&A engineering becomes tedious, because an engineer has to go through all these wells, extract key data, then use that to design the P&A job,” explains Jayabrata Kolay, drilling engineer at bp and previously lead designer for an AI project at SLB.
He says AI is fast-tracking digitalisation, as well as sorting and organising well-specific information such as casing configuration and cement placement. The trend is notable across oil and gas processes broadly – legacy data is a widespread pain point – but offers “direct savings”, as Grant describes it, in streamlining expensive and time-consuming decommissioning planning.
SLB has invested heavily in the use case, deploying generative AI (genAI) models to streamline P&A planning in its Wellbarrier Well Integrity Management System (WIMS) – a digital platform used by operators including Petronas, Equnior and VNG Norge. This was the focus of Kolay’s project, which used off-the-shelf genAI models to digitalise archival documents.
Kolay says archival documents were fed through ChatGPT, Gemini and other genAI models, which used optical character recognition (OCR) to systematically extract well data from historical manuscripts, well reports and archival documents. “OCR takes the image – it can be handwritten or typed – and recognises characters using image recognition,” explains Kolay. “Some of the existing OCR engines are very powerful. They can understand even a doctor’s handwriting.”
Technical oil and gas terminology created complications, and models required additional training to address returned bad data and hallucinations. However, Kolay says users can now use WIMS to quickly access data and cross-check AI’s findings with the relevant digitalised PDF documents.
“For users to read through a 200-page report, they would have needed a minimum of one day to extract the facts. We boiled it down to a matter of one hour,” says Kolay.
Derisking compliance: AI-organised data supports P&A planning
P&A and decommissioning compliance risks lie in the detail. Operators face a lengthy process of comparative assessment and stakeholder consultation before they can submit a decommissioning programme, and data must accurately identify the current condition and integrity status of infrastructure, alongside cost estimates and timescales.
Even when legacy data has been digitalised, organised and AI-optimised, most operators face problems with incomplete or inaccurate well data, due to legacy ownership changes, inconsistent internal reporting, or the physical, logistical challenges associated with physical surveying of wells.
As a result, applications for regulatory approval are usually fraught with errors and rejections, which can tip approval timescales from months into years.
Grant points out that the data inaccuracy problem is compounded by “discrepancies” between regulator and operator data. This is typically because regulators hold older, baseline data from initial licensing or permitting, while operators track the evolving picture. Reporting requirements and data formats have also changed over the decades, and older regulatory filings may not map cleanly onto modern methods of tracking well integrity.
According to Grant, derisking compliance is where AI’s “real value is”. DecomGPT, rahd.AI’s proprietary large language model, is trained not only on operators’ data but on regulators’ data too, enabling decommissioning teams to plan around “data gaps” and “giving them the confidence that there are not any hidden issues or risks they hadn’t identified”.
By combining both regulators’ and operators’ data, Grant says rahd.AI “serves up” the big picture to engineers. Each data point is traceable to its original source document, enabling a transparent audit trail to build trust and enable compliance. By avoiding lengthy rework processes, AI offers efficiency and quicker approval timelines.
Yet the administrative side of compliance represents only a fraction of the cost associated with compliance. Carrying out approved plans is capital-intensive and small errors can translate into millions of dollars of capital.
Drawing from Promethean Energy’s use of decomGPT in its P&A planning, Singh says AI insights can support project duration estimates and enable engineering teams to identify where avoidable costs might lie. He adds that the company has successfully reduced decommissioning costs by 30 –50% in some projects.
“This isn’t just about incurring hundreds of dollars of an engineer’s time to assess data,” says Grant. Instead, it is about the millions spent on a decommissioning team, vessel and equipment. If the needs of the decommissioning site are not fully understood, and the wrong equipment is rented and transported to a remote offshore site, “you might have to turn around and come back”.
So far, rahd.AI says its platform has been validated to save up to 15% on total decommissioning bills and is targeting 35% savings by 2027.
AI technologies for safety and environmental compliance
AI is also changing on-the-ground delivery, aligning well retirement more closely with operators’ ESG strategies.
This is particularly the case in safety – a prominent consideration in decommissioning. Even with the best planning, ageing structures, degraded well barriers and undocumented modifications mean inevitable on-site risks.
Visual intelligence (VI), a subset of AI that can process visual inputs, is one solution already seeing rollout in high-risk environments. Singh says Promethean Energy uses VI as a “safety assistant” to provide alerts in real time. Cameras are installed on the platforms, and the AI is trained to recognise and flag specific hazards, including if an employee is not wearing a helmet, does not maintain three points of contact or approaches an area of the platform without authorisation.
“AI is part of the reason that we [Promethean Energy] had a total recordable incident rate of zero in 2025,” says Singh.
Decommissioning also brings a host of environmental risks including seabed disturbance, vessel emissions, waste generation and the longer-term risk of methane leakage. AI insights can enable decommissioning teams to avoid or mitigate potential risks, either through enhanced P&A or long-term monitoring.
Methane leaks are particularly likely to occur in poorly plugged or orphaned wells, but Grant says that AI can “give operators confidence that their P&A plans are robust, because they have uncovered all the risks”.
“It all comes back to the quality of the P&A planning that is done, which is dependent on the quality of the operator’s information, which is where we come in by applying AI,” he adds.
AI is also being used to address the methane leak risk from historical oil and gas infrastructure, including forgotten or undocumented orphaned wells (UOWs), of which there are an estimated 310,000–800,000 in the US alone.
In 2024, researchers at the Lawrence Berkeley National Laboratory trained AI to recognise UOW symbols on historical US Geological Survey topographic maps. The method of well identification itself wasn’t innovative, but the scale was: “Until recently, [the human eye] was the only available method to extract information from these maps – but that strategy does not scale well if we want to apply it to thousands of maps,” explained Fabio Ciulla, a postdoctoral fellow who helped develop the technology at the Berkeley Lab.
Elsewhere, AI-enhanced digital twins are also supporting both environmental and safety considerations for decommissioning. Digital twin company eserv developed AS-TEG for decommissioning oil and gas platforms and reports that the software reduces the need for surveys by 60%. In a case study, eserv says that it developed two digital twins for two offshore oil rigs due for decommissioning, allowing an unnamed North Sea operator to determine where drains and vents were required.
By modelling an asset’s condition before decommissioning commences, digital twins also allow operators to predict and plan around environmental disturbance and to model emissions, providing engineers with further insights for ESG compliance planning.
Promethean Energy has already adopted digital twins as a necessary technology. The company uses 3D cameras in inspections to build up a digital twin “in great detail”, according to Singh, which it uses to optimise site safety.
AI is already reshaping the decommissioning process, from data to delivery. As AI technologies mature, its use cases will too evolve, but Singh warns that it is not whether AI is used that will count but how: “There is no silver bullet. You have to be an integrator of technologies and you have to integrate AI across the ecosystem that you operate in.”
