Saved To My Saved Content
Use BCG Answer to summarize this article

For decades, industrial players have turned to robotics and methodologies, such as lean manufacturing and six sigma, to automate processes and deliver sector-beating productivity gains. This playbook has run its course, however. Since peaking around the mid-2010s, labor productivity in industrials has declined across the largest advanced economies. Among US industrial companies, labor productivity has fallen by an annualized 1.2% since 2014 even as productivity across all other sectors has increased by an average of 1.2%. Manufacturing productivity has also declined across Europe’s largest industrial economies.

AI opens up a new possibility: by changing how decisions are made and how processes work across the organization, companies can fundamentally transform their economics, operating models, and competitive advantage. Our latest survey, the BCG Applied AI Index 2026, shows that the most mature companies across sectors, which we call “future-built for AI,” are achieving strong results from the technology, including the following:

Industrial players, such as automotive, machinery, and process companies, face a tougher challenge than companies in other sectors when it comes to adopting AI. Factors such as fragmented data and complex global supply chains make it harder for them to implement an end-to-end AI transformation. Product durability needs and safety mandates, for instance, require companies to set a high bar for granting AI agents decision-making authority.

However, our modeling shows that industrial companies that meet the challenge can significantly enhance their bottom line. According to our analysis, players in chemicals, automotive, and aerospace and defense that go fully AI-first can increase their EBIT margin by several percentage points, making this a key route to stronger profitability and growth.

AI-first companies in our selected subsectors rely on various AI-enabled levers to improve performance. For chemicals, applying AI to the sales function is a key value driver; in aerospace and defense, the aftermarket plays a central role in creating value; and in automotive, sales and operations planning (S&OP), sales, R&D, and sourcing are important. But despite these differences, each deep dive offers a valuable lesson for other industrial companies: by reshaping and applying multiple AI components—including agents, language models, and an enterprise cortex—players can maximize the technology’s potential.

Reshaping the Organization Is the Key to Value

AI adoption takes different forms that involve varying degrees of difficulty and reward. (See the sidebar, “Three AI Pathways in Industrials: Deploy, Reshape, and Invent.”) Of these pathways, we believe that merely deploying AI tools, such as copilots, into individual workflows does little more than create another layer of automation. To gain broader AI benefits, such as new revenue streams, a faster time to market, and greater efficiencies, it is essential for companies to reshape their organizations first—reengineering processes and functions, helping employees adapt to new ways of working, and placing the technology at the center of their operations. (See Exhibit 1.)

Table showing four ways that industrial companies can create value by reshaping for AI.
Three AI Pathways in Industrials: Deploy, Reshape, and Invent
Today, most industrial companies use AI tools to assist humans in their everyday work. Processes remain unchanged, but key stages now utilize AI tools, such as copilots or assistants—and as a result, tasks take less time. We estimate that deploying AI in this way saves individual workers 10% to 15% of their time on tasks. Much of that gain rarely shows up in the P&L, because the time saved gets absorbed elsewhere.

Introducing the technology into existing workflows has its place, but the approach is becoming table stakes in AI adoption. (See the exhibit.) Creating greater value and competitive advantage entails reimagining workflows end-to-end and transforming organizational structures and operating models so that they fit with AI. In the case of agentic AI, agents execute a far bigger share of the work, and humans take on a more supervisory role. Reshaping core functions in this way can deliver a 30% to 50% improvement in their efficiency and effectiveness.
Chart showing 30% to 50% operational improvement from an AI reshape transformation vs. 10% to 15% time savings from deploy.
Moving beyond simple deployment also enables companies to invent new data-based business models and value pools. For example, adding sensors to its hand sanitizer bottles in offices and using AI to identify replacement needs on the basis of the resulting data enabled a chemical company to transform its corporate hygiene business from simply selling sanitizer into a higher-margin sanitizer-as-a-service operation.

AI leaders in industrials are already doing this. As a result, they grew revenues more than twice as fast as the sector’s AI laggards from 2022 to 2025 and delivered roughly double the three-year TSR. In many cases, leaders apply AI to optimize cost levers. There, too, they outpace the laggards: AI leaders on average achieve over twice the share of AI-enabled cost savings, our survey finds.

Nevertheless, most industrial companies still deploy AI tools and use them to assist humans in everyday tasks, and only a few run dedicated, multiyear AI transformation programs. According to US Census Bureau information, almost 70% of US industrial companies that adopt AI make no organizational, data, or process changes; and only 12% build new workflows, versus nearly one-third of information companies.

This is where most of the value is lost. In the AI transformations we have worked on, value splits roughly 10-20-70. About 10% comes from the algorithms and 20% from the technology and data underneath them. The remaining 70% comes from people, organization, and processes. We call this “the 70,” and it is the part that most US industrial adopters leave untouched. Without it, an AI program only delivers tools and the P&L stays where it was.

Why a Reshape Is Harder for Industrial Players—but Worth the Effort

For most industrial companies, reshaping the organization for AI is a tough task, but it’s well worth undertaking. Here are the main reasons why it’s difficult.

The Data Issue. To make effective decisions, AI depends on high quality, well-organized data. Although large industrial companies generate a lot of data, accessing that data is often difficult. Industrial companies are asset-heavy organizations with high fixed costs that focus on delivering highly engineered physical products. In the past, to make their data usable, players had to make costly investments in sensors and smart machinery. Furthermore, information is often siloed. Practical insights reside in senior managers’ heads rather than being documented. Put simply, data is badly organized and poorly connected across functions and locations and is not contained in a data layer that supports AI.

Multitiered supply chains and value chain complexity further complicate the data challenge. Vendors may not share information about factors that impact component availability. At the same time, industrial companies operate across multiple markets and contend with extensive bills of materials, requiring them to navigate wide variations in pricing and different regulatory contexts. With good visibility enabled by strong data, AI tools can optimize these activities. But without them, AI struggles to make a difference.

The Trust Issue. For most industrial subsectors, uncertainty about whether AI agents will make the right decisions is of particular concern. Safety and durability requirements and reputational issues mean that companies cannot afford to take a leap of faith. They need to set a high bar for agents and adopt a cautious approach toward granting them decision-making authority, rather than rushing to match the pace of adoption of other sectors.

The depth of these challenges is borne out by our survey data. We found that only 12% of industrial companies have the necessary data capabilities to be future-built for AI, compared with an average of 18% across all other sectors, with the biggest gap occurring in the sub-score of data strategy. (See Exhibit 2.) Similarly, industrial companies are less likely to expect AI agents to have decision-making autonomy by 2030, although this lack of autonomy hampers their ability to create value.

Table showing AI capability gaps between industrial firms and companies in other sectors.

The Organization Issue. Decades of lean and six sigma have enabled industrial companies to excel at improving stable, standardized processes in small, carefully controlled steps. But AI-first operations require a different rhythm—one in which companies redesign a process end to end and then redesign it again as the technology evolves. Responsibility is split by function and site, so the end-to-end process that an agent runs rarely has a single owner. In our experience, the speed at which an organization can adapt plays a bigger role in setting the pace of AI value creation in industrials than the technology does.

The Enterprise Cortex’s Central Role in Becoming AI-First

As demand for AI grows, off-the-shelf, external models are becoming increasingly commonplace. But as they influence the way organizations solve problems, these models threaten to dictate companies’ decision making—a situation that we call cognitive lock-in. On top of this problem, industrial companies are grappling with siloed information and definitions.

Creating an enterprise cortex is a key element of the solution. It forms part of the AI tech stack and acts both as a connective layer that links business functions and as a depository for the company’s differentiated thinking and know-how—its business rules, decision logic, and operational context.

Working in tandem with large language models (LLMs), the enterprise cortex can help solve the sector’s AI-related challenges. Today, LLMs can read engineers’ messy notes, maintenance logs, and computer-aided design annotations, and use them to create a centralized pool of data. But the enterprise cortex creates a codified set of definitions and decision rules that apply across the organization and provide vital company-specific context and guardrails for agentic action. Owing to their siloed information, and the harsher consequences if an agent takes a wrong turn, we believe that industrial companies need an enterprise cortex even more than most other sectors do.

Having explicit rules also gives players something tangible to check agents’ reasoning against. Through this process, companies can limit agents’ authority initially and then slowly expand it as trust builds.

To decide how much autonomy an agent should have to start with, organizations should answer two questions: whether AI can carry out a task, and whether it should. This is the AI-first matrix. The first question addresses how well AI performs relative to people; the second considers what is at stake if an agent gets the task wrong. In situations where both answers are yes, agents can run the work under human supervision. In cases where AI is capable but the stakes are high, as with a certificate of release, the right mode is “AI flies, humans land.” Agents do the analysis and prepare the action, but a qualified person always makes the final call.

Creating an end-to-end enterprise cortex layer is just as important a prerequisite as reshaping operations for becoming an AI-first player. (See Exhibit 3.) This objective isn’t achievable overnight and needs to be achieved gradually. Consequently, as they build out the cortex, companies should incrementally expand the scope and depth of the tasks that they delegate to AI agents.

Schematic diagram showing the role of an enterprise cortex in creating lasting value from AI.

What Becoming an AI-First Player Looks Like in Three Industrial Subsectors

We’ve explored in depth how an AI-first company is likely to look in chemicals, automotive, and aerospace and defense. AI leaders in these three subsectors are already taking some of the actions that we outline below.

Each of our three deep dives has relevance for other industrial areas. The chemicals deep dive contains lessons for similar asset-intensive, process-based industries, such as metals, mining, cement, and forest products, paper, and packaging. The aerospace and defense deep dive provides AI insights for engineering-intensive and complex manufacturing environments, including machinery and industrial equipment. And the automotive OEM deep dive has implications for players involved with complex product development, intricate supplier ecosystems, and large commercial operations.

In each deep dive, we have also modeled an aspirational P&L showing the uplift and enhancements in the bottom line and individual line items that an AI-first company in these three subsectors may be able to achieve by reshaping for the technology.

Deep Dive: The AI-First Chemicals Company
The chemical industry faces a structural crisis, not a cyclical dip. TSRs sit in the bottom quartile among asset-heavy industries, EBIT margins and return on capital employed have declined for over a decade, and overcapacity in core building blocks such as ethylene, propylene, and aromatics is expected to persist through 2030. In this environment, AI is not a margin add-on for good times; it is the lever that enables AI-first players to weather the downturn and emerge stronger, rather than simply following the industry’s default playbook of cutting capacity.

The path to becoming AI-first players is different for chemical companies than for other industrial sectors. The chemical industry’s cost base is dominated by raw materials and energy, which typically account for 65% to 75% of total costs. Furthermore, process safety and product liability set a higher bar for granting agents decision-making authority. Three functions carry most of the sector-specific value: R&D, manufacturing operations (including supply chain), and marketing and sales.

R&D. AI generates and ranks orders of magnitude more molecule and formulation candidates than physical experimentation can, helping deliver target properties needed for customer applications. This narrows the field to a handful of formulations that are worthy of further examination in the lab. AI-enabled formulation design builds on patterns in a company’s own historical data, so a new grade starts from known information rather than a blank page. The payoff is both a direct reduction in R&D cost and an indirect revenue impact through faster time-to-market. (See Exhibit 4.)
Chart showing how chemicals companies can use AI agents to cut R&D costs by 10%.
Manufacturing Operations, Including Supply Chain. In operations, plants move from manually tuned to self-optimizing systems. In continuous processes such as crackers and distillation columns, AI keeps the plant operating at its economic optimum by continuously resetting setpoints on the basis of live margin, energy price, and process constraints. In batch processes, AI defines the golden batch from historical operating profiles and steers each run toward it, adjusting changeovers and cycle times to suit the process type.

With the agentic step forward, agents no longer only recommend; they also execute, writing approved setpoint changes directly into control systems within a defined envelope. The human operator’s role shifts from tuning the process to supervising the envelope, handling exceptions, and owning every safety-critical action. Operator override is a standard element of the design, and agents’ authority expands only as they earn greater trust.

Underneath sits a plant- or site-level orchestrator agent that connects previously separate systems, including maintenance, quality, supervisory control and data acquisition, and batch records. Because the agent reads across these sources, it can perform root cause analysis on an anomalous batch or a yield drop, flag bottlenecks, and prepare maintenance reports. Engineers can then review conclusions instead of assembling the data themselves.

Supply chain planning follows the same logic. Planning shifts from a periodic S&OP cycle to an always-on agent that monitors demand, supply, inventory, and logistics; evaluates far more scenarios than a manual process can; and pushes an updated, feasible plan in near real time. Instead of building the plan, human planners handle exceptions and tradeoffs that require judgment, such as allocating volume to customers amid shortages. As in the plant, authority is tiered: agents advise first, then act with approval, and only later act on their own within defined limits. A polymer producer that used AI planning agents to evaluate 50,000 planning scenarios over a 60-day horizon improved gross margin by $8 to $10 per ton and customer service levels by 4%.

Marketing and Sales. AI plans the seller’s day or week and prepares meetings and customer interactions. Preparations include inserting high-potential cross-selling and upselling opportunities into the seller’s pipeline and providing logic and documentation to support the conversation. The opportunities include using micro-segmentation techniques to target customer niches. AI also supports creating bundled and tailored offerings tied to what a customer is likely to purchase next. The result is a revenue boost, especially for specialty players. Using an AI-supported sales approach, one global specialty chemicals distributor generated 30 to 40 new leads per sales rep every month and increased the EBITDA contribution from pilot segments by 3% to 5%.

AI also makes new business models possible. Specialty players that know their customers’ operations well can use AI to track how well their products perform in the customer’s processes and to recommend actions. The result is a shift from selling product to selling outcomes, such as less downtime or higher labor productivity, and the opportunity to earn recurring revenue on top of the product. For example, a water-treatment specialist can monitor a customer’s cooling circuits in real time, adjust dosing recommendations through AI, and charge for avoided downtime rather than for tons of chemicals.

Our modeling suggests that large, global, predominantly strong-performing chemical companies that reshape to become AI-first players may improve EBIT margin by 2.5 to 4.5 percentage points. (See Exhibit 5.) The only cost line that deliberately increases is technology/IT, which funds the platform—including compute, tokens, and agent tooling—necessary for every other AI lever to work.
Stacked bar chart and table showing how a full AI transformation can give chemical companies a 2.5pp to 4.5pp EBIT uplift.
Deep Dive: The AI-First Aerospace and Defense Company
Aerospace and defense has more demand than it can build or service. The commercial backlog stands at roughly 17,000 aircraft—about 12 years of production at current rates. According to projections, demand for engine shop visits will exceed capacity by 15% to 20% by 2030, and the shortage of certified technicians will reach 20% by 2028. Already, roughly three in four programs are missing their ramp-up timelines. AI is therefore a key throughput lever. The hours that agents free up can be redirected toward more aircraft, engines, and shop visits—the things that customers want most. Each 1% of extra commercial output puts about 14 more aircraft a year into service.

Profit in A&D sits mostly in the installed base. Across 12 listed companies, airframer original equipment generates about 20% of revenue and none of the profit, while engines and aftermarket services generate 29% of revenue and more than half of the profit. Airworthiness sets a high bar for agent authority, so agents prepare evidence, findings, and dispositions, and a qualified person still signs off. Commercial R&D is self-funded, so any savings show up in the profit margin. By contrast, defense development is largely customer-funded, and cost-type savings pass to the government.

Engineering and Certification. AI agents draft detailed requirements, test cases, and change packages that are necessary to modify a civil or defense platform. Meanwhile, machine-learning surrogates deliver near-instant test results compared with high-fidelity virtual simulations that can take hours or days. Engine OEMs are already using surrogates to test hundreds of thousands of design combinations in milliseconds. One company reports that it has halved design lead times in some instances. Certification evidence builds up alongside design and testing, so gate reviews start with a more complete package. The payoff is lower development cost and faster entry into service. In defense, AI-enabled gains show up mainly in the schedule and win rate.

Production and Supply Chain. Predictive quality and in-process monitoring raise the first-pass yield metric. Each avoided defect reduces rework and traveled work from constrained production lines. Planning becomes a perpetual S&OP process that holds new-build deliveries, spare engines, spare parts, and shop visits within a single plan. Agents push the resulting demand signal down the supplier tiers and flag subtier risk weeks ahead. During a widebody ramp-up, one airframer linked engineering, production, and supplier data via AI, enabling the company to resolve quality issues faster and cut delivery lead times by a third.

Aftermarket and Services. An AI control tower orchestrates parts, labor, and approvals across engine overhaul, component repair, and backshops to safeguard turnaround. Fleet-health forecasts let shops stage materials before the asset arrives. Copilots answer technicians’ questions in context, supported by remote experts. Non-routine findings become part of standard processes, and AI-enabled scheduling matches people to jobs by skill and certification. Work scope optimization cuts the cost of materials, which is the largest maintenance, repair, and overhaul expense. Models set prices within guardrails and check quotes against contracts. Service agents resolve aircraft-on-ground queries faster, closing value leakage. Engine OEMs are building AI into inspections. One company recently rolled out an AI-enabled blade inspection tool that halved turbine-blade inspection times on its widebody engine. Another has purchased an AI software company that can read borescope video and is extending this application to military engines. (See Exhibit 6.)
Chart showing how AI-first aerospace and defense companies can use AI agents to raise maintenance technician productivity.
The value of being an AI-first player compounds because each link in the chain supports the next one. Good-quality engineering data improves first-pass yield. A better and more predictable yield enhances the credibility of the production plan and encourages suppliers to commit. And a growing fleet generates aftermarket forecasts that feed back into that plan. Conversely, a weak link limits everything downstream, which is one reason why isolated pilots rarely scale.

Our modeling suggests an EBIT margin improvement of about 3.3 percentage points for established OEMs and 5.6 percentage points for aftermarket businesses, widening the gap between selling the unit and servicing it. Defense-heavy portfolios realize roughly half of this enhancement after cost-type passthrough. Technology is the only cost line deliberately increased, roughly doubling to fund the data layer and agent tooling that every other lever depends on. These figures describe established, profitable players at full maturity. (See Exhibit 7.)
Stacked bar chart and table showing how AI-first aerospace and defense firms can boost OEM and aftermarket player EBIT.
Deep Dive: The AI-First Automotive OEM Company
Agentic transformation in the automotive industry has shifted from an IT initiative to part of the C-suite agenda. Thanks to falling token costs, the range of what agentic AI can do is expanding fast. Agents are becoming an operating layer, using standardized protocols to coordinate across systems. This raises two questions that only the CEO and board can answer: where does the value sit, and can the organization make the holistic changes needed to tap this value?

Value in automotive resides in core processes, not in support functions. The full potential of a 3- to 5-percentage-point EBIT margin uplift comes from a complete holistic transformation and is most likely to be achieved over the long term. (See Exhibit 8.) S&OP, sales, R&D, and sourcing drive the bulk of this potential. The value that AI creates is not primarily about replacing headcount; to a large degree, it involves lowering material costs, which are equivalent to roughly 57% of revenues. The agentic architecture works at the process level within priority processes. For the CEO of an established OEM, the pragmatic path is to reshape the company’s priority processes end to end, redesigning them so that agents do most of the work and people supervise, rather than building an entirely new organization from scratch.
Stacked bar chart showing how automotive companies can increase long-term EBIT by up to 5pp through a full AI transformation.
Capturing value from AI requires investment in company-wide readiness—in ownership and steering capabilities, data and platforms, and the creation of clear decision rights between agents and humans. This involves a multiyear journey that will take place in iterative waves, with self-funding from year two. Key organizational changes typically take 18 to 24 months to land, and they need to start now.

S&OP. The introduction of agentic AI in S&OP enables key elements to move from a sequential pattern to always-on. (See Exhibit 9.) Agents forecast demand and product mix continuously rather than every four weeks. They also check supplier, logistics, and production capacity in parallel with active demand forecasts, instead of waiting for a determination of demand levels, and they flag constraints and propose alternatives if any issues arise. Specialized agents reconcile supply and demand continuously, rather than late in the production cycle, and they test scenarios for potential tradeoffs across the organization—a major change from the siloed, localized planning approach most OEMs take today. Humans set the guidance and guardrails for agents, and they intervene to review material changes and resolve tradeoffs. The result is a 15% to 20% drop in inventory and a 40% reduction in planning effort.
Chart showing how AI-first automotive OEMs can use AI agents to improve sales and operations planning.
R&D. Autonomous agents, working around the clock, write and test the computer code used to design new automobiles and to function as the central component in vehicle electronics, with developers defining the architecture and providing oversight. This depends on harness engineering—meaning the tests, tools, and limits that keep coding agents on track—and on having engineers review any safety-critical code. Using natural language instructions, agentic computer-aided design units model complex components within minutes rather than days. Agents also help streamline modification processes. The payoff is a reduction in R&D engineers’ hours, a 1% to 2% reduction in overall materials costs, and greater speed. Chinese OEMs can launch a new vehicle in approximately 20 to 24 months, whereas some Western incumbents still take three to four years. AI will be a key lever for Western OEMs seeking to match China’s speed-to-market.

Marketing and Sales. AI-driven pricing engines boost revenues by determining the optimal price and incentive package for a vehicle on the basis of indicators including demand forecasting, historic trends, and competitors’ moves. Agents also take a bigger role in generating and handling leads, scheduling test drives and offers, personalizing marketing content, and identifying the next best action with each customer. OEMs that use AI to guide buyers from first search to signed contract should also sell more cars, mainly by taking share from slower rivals. The extra volume can lift profits without changing the margin structure in our P&L.

To maximize the potential of their agentic transformation, large AI-first automotive companies rely on a centralized AI office to own the platform and data, set standards, and manage risk. DevOps teams and specialized units within individual functions are responsible for running agentic processes and for the resulting outcomes.
Weekly Insights Subscription
Stay ahead with BCG insights on industrial goods

Five Must-Do Actions for Senior Leaders

When industrial leaders—CEOs, C-suite executives, and board members—set out to reshape their organizations as AI-first players, five actions are essential:

Most industrial CEOs today are deploying AI tools to extend automation. That approach is no longer a differentiator, however, and it leaves most of the value on the table. Leaders who are pulling ahead are reshaping how their companies decide, organize, and compete. The organizational changes alone take two years, so the time to start is now.

Acknowledgments

The authors thank Adam Rothman, Arun Rajamani, Brent Kelley, Chris Hill, Clemens Nopp, Csilla Szalay, Marc Roman Franke, Michael Grebe, and Philipp Lossie for their contributions to this article.