Despite this enthusiasm, hurdles such as unstructured data, uncertain model accuracy, and gaps in governance are preventing many organizations from getting the full benefit of AI tools.
Firms that rely on large-scale, around-the-clock operations know the stakes are high. A system outage or an unforeseen failure can cost millions, disrupt the supply chain, and harm overall competitiveness. A study by Aberdeen Research found unplanned downtime in manufacturing can cost up to US$260,000 per hour.
Against this backdrop, AI can predict issues before they occur, helping companies repair equipment proactively and keep downtime to a minimum. Yet concerns linger around data reliability, potential algorithmic bias, and whether AI’s recommendations are truly explainable and safe. A carefully planned approach is key to overcoming these challenges, so that AI becomes a true value driver.
Chief Technologist at Aspen Technology.
Identifying key risks and overcoming them
When setting up a new asset to work in the field, organizations will necessarily have no data to draw on. That’s where they can access learning from first principles models, coupled with simulation models to ensure a balanced set of data, availability of unlikely scenarios, and therefore enable extrapolation to new regimes of operation.
Data from the field can then be used to refine the model (close the simulation reality gap), or to predict future outcomes based on historical observations. With predictive maintenance technology, it is also possible to identify abnormalities by building models from normal modes of operation.
To enable this, companies need strong governance policies, as well as processes for labelling, storing, and updating data. While a sizable upfront investment may be needed, the payoff is significant: well-organized data fuels accurate models that deliver meaningful results.
Another challenge involves explainability. Some AI-generated recommendations can seem like a “black box” e.g., when models rely on complex neural networks. For day-to-day industrial operations, trust is crucial, as operators must be able to understand how and why decisions are made.
Including interpretable features and highlighting key decision drivers helps build that trust. When people know the rationale behind AI findings, they become more willing to follow them, improving adoption rates.
Well-designed dashboards that map input factors to recommend outputs have their part to play here. However, they will be not sufficient in themselves in achieving trust. Organizations should ensure that they select the right tool for the job in hand.
A complex model can be necessary for complex nonlinear behavior. However, while a complex model can address simple use cases, this comes at a cost, for example explainability, challenges with extrapolation, risk of overfitting, large data requirements etc. It is therefore important to select the right tool for the job. Generally, the simplest approach that solves the problem is preferable.
Trust can be further assured by the use of first principles guardrails that provide peace of mind and highlight that a provider has a thoughtful approach to AI.
In addition, there is the question of bias. Historical data sometimes reflects outdated practices or inconsistent recording methods, and if this data is used without scrutiny, algorithms may carry forward biases into their predictions.
Regular auditing of model performance, along with diverse data sets and ongoing feedback from subject matter experts, can mitigate these risks. Proactively revisiting the data strategy and staying aware of evolving regulations also helps organizations stay one step ahead.
Finally, AI’s integration with existing workflows demands attention. Even the most advanced algorithms will struggle if they fail to mesh with established processes. For instance, if plant operators need to switch between multiple tools or can’t easily act on an AI-driven alert, the system’s value quickly diminishes. Seamless product integration, visualizing AI insights, training operators on new procedures, and ensuring IT infrastructure can handle added data loads are often a make-or-break factor for success.
Practical steps to harness industrial AI
A strategic roadmap for AI adoption starts with identifying use cases that promise strong returns. Many companies find early success in areas like predictive maintenance, where AI models spot signals of potential future breakdowns and enable timely fixes. Another example are hybrid models that allow the creation and sustainment of models from data in the field.
This accelerates the model building for complex processes and improves the representation for design optimization or control, thus supporting efficiency and sustainability improvements. Another best practice is merging automation with human expertise. While AI excels at e.g., sorting through large data sets to pinpoint trends or anomalies, seasoned operators understand the practical nuances of running a plant.
Collaboration between people and technology ensures that strategic decisions blend intuitive knowledge with data-driven recommendations. By keeping humans in the loop, organizations lessen the odds of unexpected failures and maintain trust among the workforce.
To secure buy-in across management levels, pilot programs need to show quick, tangible benefits. If a narrow project using AI for quality checks significantly reduces scrap in a factory, the cost savings and improved customer satisfaction help support broader initiatives. Documenting these early gains and calculating the return on investment helps justify scaling AI across multiple sites, which often involves more complex budgeting and approvals.
As expansion proceeds, robust model governance becomes essential. Models must be monitored for “drift,” when real-world conditions change. Deploying self-adapting AI technology or regular checkpoints with engineering and data science teams helps ensure the technology performs as expected.
Enterprises can also set up review boards or specialized groups to vet new AI solutions, confirm compliance with regulations, and measure alignment with corporate targets such as safety or sustainability.
Lastly, organizations should build long-term expertise within their teams. Successful AI adoption isn’t a one-time event; it’s a continuous journey of refinement, learning, and adaptation. Training employees to understand AI fundamentals, interpret analytics, and collaborate with data scientists goes a long way toward embedding AI into the corporate culture. This level of internal capability also positions companies to pivot faster as new technologies emerge.
We’ve featured the best AI chatbot for business.
This article was produced as part of TechRadarPro’s Expert Insights channel where we feature the best and brightest minds in the technology industry today. The views expressed here are those of the author and are not necessarily those of TechRadarPro or Future plc. If you are interested in contributing find out more here: https://www.techradar.com/news/submit-your-story-to-techradar-pro
