Artificial intelligence has finally left the experimental corner and stepped onto the plant floor, the trading desk, and the supply-chain control room. That leap is exposing messy data, widening skills gaps, and proofs of concept that never should have left the lab. In this straight-up Q&A, AJ Abdallat, CEO and Co-Founder of Beyond Limits, explains how disciplined data stewardship, proof-of-value framing, and Hybrid AI can turn those headaches into competitive gains. He shares real examples, from clip-on IIoT sensors spotting motor faults before they happen to on-premise language models that guard hard-won know-how. If you want AI that delivers measurable reliability, efficiency, and safety, start reading here.
The interesting dynamic is that the AI shift has been mirrored by Digital Transformation initiatives which have exposed a number of issues and risks. A major one is data readiness. Many companies are realizing their data isn’t clean or structured enough to support digital initiatives, which is prompting a wave of Master Data and data quality projects.
At the same time, companies are facing the loss of staff through market pressures and retirement, and it is getting more difficult to backfill knowledge loss from the company. In some cases, this is leading to production and quality issues, and in more serious scenarios, safety and loss of containment events. To counter these issues, AI deployments are now expanding beyond traditional numeric modeling. We’re seeing more companies turn to knowledge- and agentic-based solutions leveraging the data and numeric AI as inputs.
AI projects tend to succeed when they follow the same principles as other Digital Transformation programs that start with strong business engagement and an understanding of the problem from a holistic end-to-end perspective (e.g grounding the project in the current ‘As-Is’ state). This engagement should start with the sponsor to understand and align with their vision of the ‘To-Be’ state and defining risk managed phases establishing value and metrics to support the rollout and successful sustainment of the solutions.
One reason POCs (proof of concept) have limited success is because they are technologically looking to prove the AI capability rather than solving a meaningful business problem. In comparison, a POV (proof of value) is typically framed around the business problem first, using an AI-driven solution and other technologies to address a holistic business problem. The shift in focus from tech demonstration to business impact is critical.
In manufacturing, one of the most widespread applications of AI has been in numeric models that support forecasting and predictive maintenance. More accurate production planning and scheduling allows organizations to issue more effective operational directives, ultimately improving the financial performance of assets. When predictive maintenance is layered in – helping to proactively address issues before failures occur – it increases asset availability and reliability. The combination of smarter planning and higher uptime is delivering tangible, measurable gains across manufacturing operations.
One strong example is the shift from traditional “Plan vs. Actual comparisons to an “Operations Directed vs. Actual” approach that plays a key role in reducing inherent noise and provide higher fidelity feedback. By feeding more accurate performance data into planning systems, companies can generate higher-fidelity yield vectors, leading to smarter planning and better execution.
Another rising star is the advent of next generation inferentials. These AI-powered models can deliver real-time predictions of any process characteristic, such as product quality, or process measurements, such as online analyzer values or column temperature, etc. This not only improves controller performance but also helps operators return systems more quickly during process upsets, improving both safety and efficiency.
AI is proving to be a powerful tool for monitoring, identifying, and addressing performance gaps in financial operations. For example, it can come from the traditional waterfall analysis by not only surfacing issues but also recommending corrective actions. AI can also help uncover the true operational constraints that limit capacity or increase asset availability which contribute stronger top line results.
An area that’s often overlooked but equally impactful is energy management. AI can work alongside traditional systems to optimize energy use, and in some cases, has driven fuel cost reductions of up to 3% while also decreasing consumption of other utilities, like hydrogen. These improvements can make a significant financial difference, especially at-scale.
AI is just beginning to make a real impact in transportation and logistics, particularly in improving asset fleet supply chain visibility and performance. AI is now being used to augment traditional linear programming optimizations to better respond to real-world disruptions, like equipment failures, and better able to capitalize on opportunities, such as distressed cargo availability. These are typically open loop advisory solutions today, providing recommendations for human decision-makers. But as confidence and trust in these systems grows, there’s a clear path toward automating more of these decisions to improve speed, consistency, and operational efficiency.
To me, the value of IOT and AI comes from the ability to fast and flexibly deploy solutions to solve field problems without the need for a capital project. Today, there are a host of WiFi IIoT sensors used in AI projects that can collect data to monitor asset performance and predict future failure.
One example is a clip-on sensor that transmits the current load on a motor lead that can be used to diagnose electrical quality and even mechanical issues. This just requires that the IoT data pipeline be established using site WiFi and an AI platform to develop, deploy, sustain, and scale AI solutions.
Right now, some of the most impactful applications of computer vision AI are in the area of safety. For example, systems can detect when there are unsafe conditions such as personal protective equipment (PPE) not being worn in required zones, or a ladder not being properly secured in-use. These are relatively straightforward use cases with current technology, but they’re paving the way for more advanced operations, such as robotic inspections that can execute high-quality, faster inspection tasks with minimal human intervention. What sets our approach apart at Beyond Limits is our use of Hybrid AI. By combining different types of AI – symbolic, numeric, and reasoning-based – we can significantly reduce deployment time and complexity. Unlike pure machine learning models that require training on every possible scenario, our Hybrid AI can deliver value without needing to account for every permutation upfront.
One major trend we’re seeing is the growing use of Large Language Models (LLMs), like ChatGPT, often without a full understanding of the security implications, ie information placed in these tools are then available to everyone on the Internet. As a result, forward-thinking organizations are now deploying on-premise LLMs. These internal systems ensure that proprietary information stays protected while still enabling employees to access and share knowledge securely across the organization.
Also, many companies are experiencing knowledge loss as staff reductions and retirements impact the workforce. As experienced employees leave, companies risk losing decades of best practices and hard-won operational know-how, which can negatively impact asset performance, availability, and even site safety. Connecting an on-premise LLM to all best practices documents, operator logs, supplier documents and web site scrapes curated by employees prior to departure can capture this knowledge efficiently, allowing them to preserve institutional knowledge and maintain continuity, even as the workforce evolves.
One of the most common practices that I see slowing down is implementing AI technology just to see how it works on a specific technical issue without considering the broader business context. For example, a few months ago a client asked our team for the source code of the algorithm they expected us to deliver. We explained that our ML Operations platform typically involves more than 30 different algorithms, all working together. That opened up a more productive conversation about how different types of AI, from machine learning to rule-based and symbolic reasoning, combine to create a comprehensive, effective solution. Machine learning is just one piece of a much larger puzzle.
Another common mistake is trying to implement AI in a closed loop system before the organization and users understood and trusted how a finding is delivered. Trust is critical, and if the AI can clearly explain why it’s recommending a particular action, users are more likely to accept and validate those recommendations and allow action automation.
One approach could be to start by developing a centralized set of best practices that address common, high-value business challenges. Then, focus on a single work process or use case to deliver a quick POV. The goal is to demonstrate measurable impact and build trust with business leadership early in the process.
In many cases, you may need to run a few POVs across different use cases to showcase how various types of AI, such as machine learning, rules-based systems, or hybrid approaches, can address diverse operational needs. This helps ensure the organization sees the broader potential of AI, while also learning how to scale solutions effectively.
The next wave of AI innovation across manufacturing, finance, energy, and transportation is centered on the move toward autonomous operations in which systems that can self-optimize and adapt with minimal human input. Achieving this will require integrating multiple AI technologies, from machine learning to reasoning engines, into trusted, scalable solutions. Companies that focus now on building trust and proving value with AI will be best positioned to lead in this shift.
Start by identifying high value business problems that align with your company’s strategic goals and focus on use cases that can deliver meaningful impact early on. It’s critical to involve people who understand both the business and the technology, so you can bridge the gap between operational needs and AI capabilities. A clear, collaborative approach from the start sets the foundation for long-term success.
Beyond Limits has worked extensively in oil and gas. What lessons learned from applying AI in such a complex environment could benefit other industries moving toward operational AI?
The beauty of AI is that its value is highly transferable across industries. Lessons learned around reducing emissions and fuel cost in the energy sector can be applied to other energy-intensive industries like power generation, concrete manufacturing, and chemicals to name a few. The core principles of optimizing performance, improving efficiency, and driving sustainability through AI are universally relevant.
