Small Language Models: A Game-Changer for Oil and Gas Digital Solutions

Our Perspective

Faced with the dual challenges of an increasingly complex industry and escalating global energy demands, oil and gas companies are turning to Generative AI (GenAI) for smarter, more secure technologies. Companies are actively seeking AI solutions that improve operational efficiency and support long-term sustainability—from optimizing upstream exploration to automating downstream logistics to enhancing safety across the supply chain.

However, with these advancements comes the challenge of finding AI tools that are both precise and industry-specific. Large language models (LLMs) have a lot to offer, but they fall short when it comes to an industry's specialized needs. Small language models (SLMs) offer a promising alternative.

The Problem with Generic LLMs in Oil and Gas Technologies

LLMs like GPT-4 are incredibly powerful because they’re trained on a wide range of data. However, this general knowledge makes them less effective for industries like oil and gas that require specific expertise. For example, asking a typical LLM to generate a detailed drilling plan will give a generic textbook overview rather than a specific drilling plan for the well in question. It is not trained to have the necessary technical understanding or context to give an accurate answer.

Data security is another primary concern, especially for national oil companies. Most LLMs run on remote servers, meaning companies or end users need to share sensitive data off site—something many oil and gas companies prefer to avoid. LLMs also require significant computational power, further increasing the carbon footprint of an already carbon-intensive industry. Oil and gas companies would like to avoid digital investments that would increase their environmental impact.

Why Small Language Models Are the Solution

Unlike LLMs, SLMs can be quickly and cost-effectively trained on industry-specific datasets, allowing them to better understand the specialized terminology, processes, and technical nuances of the oil and gas sector. This targeted training makes them better suited for various industry-specific tasks.

For instance, upstream SLMs can automate the creation of work orders for maintenance requirements specific to the industry, reducing downtime and operational costs. Downstream, SLMs can be used to enhance monitoring and predictive maintenance. While LLMs can analyze asset and performance data, their broader training makes them less efficient at recognizing specific patterns unique to assets like pipelines or functions like flow assurance. SLMs can quickly and accurately detect anomalies or leaks and predict and prevent issues like hydrate formation or wax deposition.

This specificity makes SLMs valuable throughout the oil and gas industry for optimizing everything from inventory management and refinery operations to quality control and environmental monitoring. Businesses can train SLMs on specific ecological regulations and internal targets, enabling them to generate more precise reports, track emissions, and monitor waste management practices. As governments crack down on environmental impact reporting and regulation, this targeted capability can help oil and gas companies remain compliant while keeping processes efficient.

Other aspects that make SLMs particularly well-suited for the oil and gas industry include:

• Data Security: SLMs can be deployed on-premises or in a company’s controlled environment, so sensitive data stays safe and secure.
• Lower Carbon Footprint: Since SLMs are smaller, they use less computational power, which helps reduce their environmental impact and makes them better suited for remote environments with less access to computational power.
• Cost Efficiency: Requiring fewer resources, SLMs are also more affordable to implement and maintain.

Reducing Hallucinations for Technical Use Cases

One major challenge with GenAI, especially in technical industries, is the risk of hallucinations when the model generates inaccurate or misleading information. These errors can have severe consequences in high-stakes sectors like oil and gas. Companies can reduce these risks by training SLMs on domain-specific data and improving accuracy. SLMs can also be designed to credit original sources of information, promoting ethical AI use and proper recognition of industry experts.

The quality of training data is crucial for SLMs to perform well. Training should include a mix of data types: public data for basic industry terminology, subscription data to deepen the model's understanding of technical intricacies, and proprietary data to tailor the model to specific organizational needs. This comprehensive approach ensures the model delivers more accurate and actionable insights, addressing concerns around hallucinations by improving the precision of AI applications in technical fields.

Why Now?

The oil and gas industry is at a pivotal moment where AI agents, data, and platform components are maturing in interdependent ways. Agents, like SLMs, are becoming more sophisticated and capable of handling industry-specific tasks, such as predictive maintenance and drilling optimization. Data is more accessible than ever, enabling broader use of machine learning algorithms trained on specific datasets, such as seismic or production data. Platforms are evolving with open standards, making it easier to integrate these technologies, ensuring a seamless and robust ecosystem for innovation.

Combining SLMs with machine learning (ML) algorithms offers a potent mix for the oil and gas industry. ML can process vast amounts of data, identify patterns, and make predictions. When coupled with SLMs, these predictions become more contextually accurate and actionable. At the same time, the ever-shortening innovation cycle of AI and GenAI foundation models means that these advancements are continuously evolving, driving greater efficiency and accuracy in operations.

Data readiness and technology democratization are crucial in this AI revolution. Open standards and the availability of compute power on demand, through services like edge computing and 5G connectivity, enable real-time data processing and decision-making at scale. This is a game-changer for the oil and gas industry, where timely and accurate data can mean the difference between profit and loss. The pathway to garage innovation is now clearer than ever, with startups and smaller firms able to leverage these technologies to compete with industry giants.

Scalability and Long-Term Potential

SLMs aren’t just a short-term fix; they can grow with the industry. While they may start by addressing specific tasks within oil and gas, SLMs can be fine-tuned to support a broader range of operations. This flexibility ensures that SLMs will remain valuable tools as the industry progresses.

To fully leverage SLMs' potential, companies should consider a phased implementation strategy that allows for gradual scaling and continuous improvement. Businesses can ensure a smooth transition and maximize ROI by starting with high-impact areas and expanding as the models prove their value. Collaborating with AI specialists can provide the necessary expertise to navigate this complex landscape, ensuring that SLMs meet current demands and drive future advancements.

About the Authors

Dr. Lakshmikantha Rao Hosur
Senior Partner – Energy, Resources, and Decarbonization

Lakshmikantha (Kantha) has over 20 years of consulting experience related to energy and the energy transition across Europe and North America. He has worked with clients and assets globally. Drawing on his deep knowledge of the energy value chain, Kantha is a strategist with a proven track record for delivering technology solutions—from ideation to go-to-market—and achieving delivery targets through consulting and portfolio management. Kantha holds a master’s degree in Geotechnical Engineering and a PhD in Soil and Rock Mechanics. He has previously worked with Repsol and Schlumberger (SLB) and is based in Amsterdam.

Pranav Kolachana
Principal Consultant - Energy, Resources, and Decarbonization

Pranav has more than 11 years of experience consulting oil and gas (O&G) companies globally, developing strategies for country chairs, creating go-to-market playbooks for business heads, and performing business risk analyses for senior stakeholders. He has served as a product manager for O&G supermajors in their supply chain, lubricants, decarbonization, and new energies businesses.

Pranav earned an MBA from the Indian Institute of Management Calcutta and a Bachelor of Engineering from the Birla Institute of Technology and Science, Pilani. He previously worked as a cementing engineer with Schlumberger.