Imagine that an equipment problem just brought your factory assembly line to a grinding halt. Production is delayed indefinitely, as technicians struggle to identify and fix the problem. The clock, in this case, is the enemy: when you factor in lost revenue, financial penalties, idle staff time, and restarting lines, you’re looking at losses totaling $532,000 an hour on average.

Clearly, shortening downtime is essential in industrial manufacturing. But doing so is increasingly difficult because of a growing shortage of experienced workers, who are typically spread too thin to fix every equipment breakdown.

Here's where generative AI (GenAI) could play – and in some cases already is playing – a major role. With GenAI's ability to digest hundreds of pages of manuals, technical information, and input from experienced technicians, a junior-level maintenance engineer could get fast answers to problems through a natural-language query – as well as easy-to-understand instructions and tools to fix them.

While industrial manufacturers are being cautious about applying GenAI to their core operations, the majority are actively exploring how this powerful technology could reduce costs, increase efficiency, and improve aftermarket services. For that to happen, they will need to assess and prepare their data, determine which large language models (LLMs) to use in which contexts, and develop governance mechanisms for their companies’ use of AI models and data.

Below, we address questions that manufacturing leaders can consider as they evaluate how best to use GenAI.

Q: Our company hasn’t begun using GenAI. Are we at risk of falling behind?

A: It’s definitely time to get into exploration mode with this powerful technology. Some 78% of industrial manufacturing executives recently surveyed by KPMG named GenAI as the top emerging technology of today – and many are already exploring ways to apply it.

Azaz Faruki, partner at management consultancy Kearney, says 70% of the firm’s clients are “actively considering it and trying to identify use cases.” Cognizant, a provider of consulting and professional services, has over 200 GenAI projects in process around the world, according to Bhoopahti Rapolu, leader of the company’s analytics practice in manufacturing.

But some perspective is needed. Although most manufacturers are investing in identifying use cases for GenAI across their supply chains and trying it in areas such as customer service, “organizations are being cautious about applying GenAI to core operations that could directly impact their P&L,” says Kabali Ganesan, director of business consulting at Cognizant. “But it’s just a matter of time before this technology starts benefiting core operations,” he predicts.

That’s hard to judge, given that most projects are in such early stages. “This is so new that there are very few examples of proven results,” says Naren Agrawal, professor of supply chain management and analytics at Leavey School of Business, Santa Clara University. “People are considering many different use cases, and in some they are making progress, but it’s too early for validated successes.”

Further, many manufacturers may rely on their existing software vendors to fold GenAI capabilities into existing tools that use AI and machine learning. And according to a report by Boston Consulting Group (BCG), the most promising applications for optimizing factory operations will likely use a mix of GenAI, machine learning, and deep learning.

“For certain digital applications in the factory of the future, GenAI is less suitable or offers a less favorable ROI than traditional machine learning/deep learning–based AI,” says the firm’s December 2023 report on GenAI. “Consequently, manufacturers will need to employ both GenAI and traditional AI to optimize their factory operations.”.

Q: Where should we focus our exploration?

A: The most immediate use cases of GenAI for industrial manufacturers will apply the technology’s ability to digest large amounts of data and present information to workers or customers through chatbots. The most likely use cases fall into the following categories:

• Overcoming labor shortages with faster employee training. Kearney’s Faruki says several manufacturers are already deploying GenAI-based tools that speed employee onboarding and training. Doing so will help get new employees up and running more quickly and augment the skills of junior staff – both of which are necessary in the face of industrial manufacturing’s acute worker shortage. The Manufacturing Institute projects that a lack of skilled labor could leave 2.1 million manufacturing jobs unfilled by 2030.
  
  Faruki describes how a Kearney partner is developing a platform that uses two GenAI engines. First, experienced employees can capture their knowledge with smartphones as they do their jobs, and the AI engine can turn that content into how-to videos in multiple languages. Another AI engine ingests existing standard operating procedures and generates video guides. “The result is a unique and extremely valuable library of organizational knowledge that can be effectively transferred to every worker,” Faruki says.
  
  Further, that information can be turned into an interactive chatbot on mobile phones for technicians to use in real time. “It will guide technicians through every single step – what you need to inspect to identify the problems, what buttons to push,” says Faruki. “You can keep asking questions, and it will guide you to the right solution.”

• Speeding and predicting equipment maintenance needs. Traditional AI software can use both historical data and real-time sensor data embedded in factory equipment to predict when that equipment needs maintenance to avoid potential breakdowns, says Randal Kenworthy, senior partner and consumer & industrial products global practice leader at West Monroe Partners, a business and technology consultancy. Add a GenAI chatbot to the mix, and the maintenance engineer could then ask, through a natural-language query, exactly what to do.
  
  “The chatbot could identify three things that may need adjustment or recalibration, then pull up the documentation showing the engineer how to do it,” Kenworthy says. “Now you have a more informed and natural way of responding to the maintenance needs.”
  
  Using chatbots to instruct maintenance technicians “is a big efficiency gain, especially when it’s embedded in existing software,” Kenworthy notes. “We’ve built and piloted this capability, but it’s hard to get engineers to use it as a stand-alone application. If it’s embedded, it has a lot more traction.”
  
  GenAI could also improve spare parts management, adds Faruki. By tracking which parts on which machines fail and how often, GenAI could predict how many parts to keep on hand for certain machines.

• Delivering information to factory workers to improve production workflow decisions. The combination of GenAI with cloud and edge computing means information can be delivered to factory workers in a way that’s easy to act on, notes Brian Legan, industrial products leader for advanced manufacturing and mobility in the Americas at EY. Currently, workers may carry rugged tablets with complex dashboards, he says. GenAI can not only present information in ways that are easier to understand but can also interactively answer questions.
  
  “Now you’re not just pushing information to frontline workers; you’re enabling them to have a conversation with the AI about ways to improve throughput and quality,” he explains. “If a worker sees some sort of disruption, they can query the AI about options and tradeoffs to make the best decision.”
  
  Sharad Prasad, business leader in manufacturing at Cognizant, calls this the democratization of Internet of Things data. “Not everyone needs to be a data specialist,” he says. Rather, GenAI “could pull the data from yesterday and tell the operator, ‘Here are the top five issues in the plant that need your attention,’” he says.

• Creating custom configurations without reinventing the wheel. In addition, GenAI could help makers of complex, custom products become more efficient. Expensive machinery, such as semiconductor fabrication equipment, is custom-built for each customer. These machines may have millions of parts, with specific configurations for specific customers. Today, those machines are specified, designed, and built practically from scratch. GenAI can help manufacturers by crunching data on past designs and identifying how to configure new equipment for certain customers.
  
  “Manufacturers may be able to develop better solutions more quickly by training GenAI with full data sets of all the solutions their engineers have developed in the past and then using that knowledge to quickly configure the solutions customers want,” says a recent KPMG report.

• Enabling quality control even on equipment that’s never been in the market. GenAI could also help fill gaps in data used to train quality control systems, says Agrawal. When manufacturing a new and novel product, a visual quality control system may not have enough real-world data to identify all possible product defects, he explains. Specifically, there may be too few images of those defects to train the AI models. “If defects are rare, you can’t wait till you have a sufficient number of defects to train the model. So generative AI can be used to create synthetic data on defects.”

• Improving aftermarket services. Companies could also employ GenAI to gather and use data for after-sales services and support. Again, this application relies on the ability of conversational GenAI to deliver the right information to the right person at the right time.
  
  When a complex product fails in the field, it can be challenging to figure out what went wrong and how to fix it, explains Cognizant’s Rapolu. “There can be thousands of pages of manuals, and they can change every year as new technologies and models come in,” he says. With a chatbot trained with user manuals, service manuals, and data on previous service calls and repairs, maintenance engineers have an automatic assistant. For instance, they can ask the chatbot how many times a problem has happened in the past and what specific part was replaced.
  
  Cognizant is working with an elevator manufacturer to develop a maintenance advisory system. The tool ingests thousands of pages of product specs, log data, and FAQs on hundreds of elevator subsystems, says Rapolu. The application, which is in the pilot phase, tells field technicians how to diagnose and resolve problems quickly. Rapolu says the application has already cut overall repair time by more than 90%.
  
  This type of solution can be done quickly, notes Rapolu. He says such GenAI applications can typically be developed in 10 to 12 weeks by training LLMs on a company’s manuals and data.
  
  Manufacturers can also use this data to improve the next generation of products, notes EY’s Legan. He suggests that manufacturers could use GenAI to collect data on how products are used in the field and send that back to engineers and designers. “Think of it as a feedback loop. You could allow someone in design or even marketing to query the AI about how the product is being used,” he explains. “Maybe it’s not being used quite as you envisioned. And the company can learn from that.”

With a chatbot trained with user manuals, service manuals, and data on previous service calls and repairs, maintenance engineers have an automatic assistant. For instance, they can ask the chatbot how many times a problem has happened in the past and what specific part was replaced.

• Designing parts and products. GenAI is also being used at the initial design phase of products, although this is a gray area in which the delineation between traditional AI and GenAI blurs. NASA made a splash in 2023 when it announced it had used generative design on certain spacecraft components. Some news reports said the process, called Evolved Structures, was GenAI design; others called it generative design.
  
  Regardless, some manufacturing experts are recognizing GenAI’s design capability. In a Capgemini survey, 30% of industrial manufacturers said they were piloting generative design of parts optimized to meet specific goals and constraints. EY’s Legan says GenAI design is revolutionizing rapid prototyping for manufacturers. “That has really changed the game,” he says. AI design is revolutionizing rapid prototyping for manufacturers. “That has really changed the game,” he says.

Q: Is there anything we can do now to prepare for a GenAI initiative?

A: One is to find out what your software vendors’ planned capabilities are for GenAI. “In the future, we see GenAI becoming embedded in existing platforms,” says West Monroe Partners’ Kenworthy. For example, a vendor of a manufacturing execution system platform that already uses AI will likely incorporate the natural-language processing capabilities of GenAI.

Other recommendations include:

1. Take the time to improve data quality and management. “Manufacturers are not spending enough time on enterprise data management,” says Legan. The best algorithm in the world won’t help if you have inaccurate or conflicting data. Many companies are dealing with a hodgepodge of legacy systems. They often have multiple enterprise resource planning systems and different data warehouses, for example. The data in these silos has to be unified and reconciled before it can be useful, he notes.
   
   This may be one reason that the most immediate application of GenAI is to ingest technical information from manuals. This information is not data taken from key operational systems; presumably, specifications and instructions in various manuals are consistent. Such an application would be of limited use, however, if companies don’t have good data on past repairs.

2. Carefully examine whether and where to use GenAI to make sure it’s adding business value. The aforementioned elevator manufacturer, for example, is cutting repair time, which keeps the customer happy and saves the manufacturer time and money. A semiconductor equipment maker could save time and costs in quoting and building custom machines.

3. Decide on LLM sourcing. According to the BCG study, manufacturers have three options: use a model hosted by a vendor and fine-tune it to their specific needs; use an open-source foundational model, which can also be fine-tuned if required; or develop their own LLMs. The third option is for those who want more control and customization. It’s also expensive and requires huge amounts of data, so businesses usually use one of the first two options.
   
   Survey the field to see if there are LLMs available that are appropriate for your application. Determine what kind of data the LLM requires for training to fit your application and whether you have such data in usable form. And even when using a pretrained model, you should test that model in your application for reliability and accuracy.

4. Develop oversight and security. How much human interaction is needed in this application? What would a mistake by GenAI cost you? For example, some industries or products require more reliability than others – think washing machine failure vs. semiconductor fabrication equipment failure, the latter of which could cost a customer millions of dollars in revenue. If the manufacturing process will be audited – for example, to meet government regulations – companies should require human review and sign-off to ensure traceability.

5. Meet data privacy requirements. Three-quarters of organizations in Capgemini’s research identified a lack of data privacy and protection as a major concern for realizing the full potential of GenAI. Some countries, including Italy, Spain, and Canada, are looking into how OpenAI is collecting its training data. So organizations could be exposed to liability when using OpenAI or other pretrained models from the market, Capgemini says.

It's clear that GenAI offers the chance for manufacturers to augment and improve how they deploy skilled workers, manage production workflows, and serve customers. Achieving these benefits means assessing your data – its quality and how it’s managed – as well as how best to use it with GenAI in your operations.