AI Applied: Successful Use-Cases in APEC's Traditional Industries

According to Microsoft’s 2026 AI Diffusion Index, Singapore leads APEC at 63.4 per cent in the share of working-age population using AI, followed by New Zealand, Australia, Canada, and Korea. At the other end of the distribution, several economies remain below 20 per cent, with Papua New Guinea at 7.7 per cent.  

These figures measure individual use rather than business adoption, but they illustrate the very different environments in which APEC firms are operating. Workers' familiarity with AI, digital connectivity, and skills all influence how readily businesses can put new technologies to work.

The same challenge is visible at the firm level. Large enterprises can invest in data infrastructure, specialist talent, and customized AI applications. Smaller businesses are more likely to depend on AI capabilities embedded in the platforms, software, and business tools they already use.

For most MSMEs, the path to AI will therefore be adoption, not development.  

The challenge is to make that pathway affordable, accessible, and trusted across APEC. 

The AI divide

As in the wider digital economy, the divide in AI is less about access to the technology and more about the conditions for using it: data, skills, connectivity, and capital required for adoption. Closing that gap is where co-ordinated APEC action can do the most for regional economic growth. 

Why APEC action matters

The private sector will develop and deploy much of the technology. But businesses cannot create all of the conditions for adoption on their own.

Data interoperability, regulatory and reimbursement frameworks, workforce development, digital infrastructure, and measures that enable smaller firms to participate all have a public-policy dimension. Co-operation across APEC can also help economies learn from successful deployments rather than each starting from scratch.

The opportunity is to turn uneven experimentation into broad-based adoption, allowing more businesses, in more traditional sectors and across more APEC economies, to translate AI into productivity, competitiveness, and growth.

This report examines what is already working and identifies the practical lessons for businesses and policymakers seeking to make that transition. 

This report documents 13 AI case studies across nine APEC economies, spanning traditional sectors including mining, logistics and maritime transport, health care, agriculture and aquaculture, forestry, manufacturing, and new energy.

The cases were deliberately selected to capture a range of economies, sectors, firm sizes, and approaches to AI adoption. They include large enterprises deploying AI at scale, smaller businesses and producers adopting AI-enabled solutions, and technology providers developing applications for traditional industries.

Each case study was developed primarily through interviews with participating firms, supplemented by publicly available information where relevant. Participating firms reviewed and approved their respective case studies for accuracy.

The analysis focuses not simply on the technology deployed, but on the business problem or opportunity being addressed, the conditions required for adoption, the results achieved, and the lessons that can be applied elsewhere across APEC.

The cases are not intended to provide a statistically representative assessment of AI adoption across the region. Rather, they provide practical evidence of how AI is being deployed today, what is enabling successful adoption, and where barriers remain.

Taken together, the cases provide a window into what it takes to take AI from technological possibility to practical business value.

Two cases, Banyu and FisTx, were developed with the Kadin Indonesia Institute, covering seaweed and shrimp aquaculture. Three cases were developed with the APEC Study Center at Nankai University, covering manufacturing, low-altitude drone logistics, and new energy. These are woven through the analysis rather than treated separately. They illustrate patterns that recur across the wider set, including the distinction between firms that adopt AI and firms that build it.

A note on evidence

The findings rest on what firms reported in interview, corroborated by public sources where available. Quantified outcomes of AI adoption are attributed to their source and, where a figure could not be traced to an approved source, it is described qualitatively rather than stated as confirmed. The full case studies, with their evidence and sourcing, are provided in the annex. 

Start with the business problem or opportunity

The strongest deployments began with a business outcome, not an AI ambition.

Successful adopters first identified a problem to solve or an opportunity to capture and then assessed whether AI was the appropriate tool. This gave firms a clear objective against which the value of deployment could be measured.

Rio Tinto described the risk of reversing this sequence: when applying AI becomes the objective, the underlying business problem can be lost. The company therefore focuses on applying AI to defined business needs, with applications spanning areas including productivity, operational performance, equipment reliability, and mine planning.

UPS followed the same principle from a different starting point. Changes in customs requirements created an operational problem that made manual classification increasingly difficult to sustain. Machine learning was deployed in response to that need, helping customs clearance recover to around 99 per cent.  

The lesson is straightforward: AI adoption should begin with the value a business is trying to create, not with the technology it wants to deploy.

Get the data ready

Data readiness sets the ceiling on AI value.

Across the cases, the quality, accessibility, and consistency of data determined what AI could achieve. In several instances, preparing the data required as much attention as selecting or developing the technology itself.

Mitsui described data readiness, rather than the availability of AI tools, as setting the pace of adoption. Rio Tinto faces the challenge and opportunity of applying AI across decades of operational and legacy data. Yeji Data Lab found that operational metrics first had to be consistently defined and re-engineered before effective modelling could begin. eShepherd similarly rebuilt its inquiry database before applying AI to customer support and warranty processes.

These experiences point to a practical sequencing issue. Firms may be attracted by increasingly capable AI models, but sophisticated technology cannot compensate for fragmented, inaccessible, or poorly defined data.

For many businesses, the most important first investment in AI may therefore be the data foundations that make AI usable.

Create an enabling environment

Technically viable AI does not automatically translate into commercially viable adoption.

The cases demonstrate the importance of the policy and market environment surrounding the technology. Regulation, approval processes, reimbursement, and assurance frameworks can determine whether an AI application can progress from successful demonstration to sustained deployment.

EverEx provides the clearest example. Its AI-enabled home rehabilitation technology had demonstrated technical and clinical potential, but adoption depended on regulatory approval and a mechanism to reimburse the clinical activity supporting the service. Once a reimbursement pathway became available in the United States, its revenue there overtook its Korean revenue in 2024.  

UPS similarly identified regulatory complexity and incomplete adoption of paperless customs processes as constraints on further automation.

The implication extends beyond these individual cases.  

When policies, regulations, or frameworks do not evolve alongside technology, viable AI applications can remain stranded between innovation and adoption.

Governments therefore have an important role not in choosing the technologies businesses should use, but in creating predictable conditions in which proven applications can be responsibly deployed and scaled.

Prepare the workforce

AI adoption is not simply a technology investment. It changes how work is performed, how decisions are made, and how employees contribute value.

Successful adoption depends on combining AI capabilities with deep industry knowledge. The challenge is not simply to develop more AI specialists, but to equip people who understand their industries to apply AI effectively.

Across the cases, AI reduced or automated some routine tasks while increasing demand for workers able to combine sector knowledge with digital and AI capabilities.

At UPS, machine learning substantially reduced reliance on specialist manual customs classification. Rio Tinto identifies one of the constraints on adoption in the resources sector as the availability of people who understand both complex industrial operations and AI. eShepherd developed new data-science capabilities as AI became more deeply integrated into its operations.

Antofagasta Minerals illustrates how firms can respond deliberately. Its Digital Academy provides employees with new capabilities as digital tools and AI become integrated into mining operations.

The cases therefore point primarily to task redesign, augmentation, and redeployment, while also demonstrating the need for substantial upskilling. The challenge is not simply to teach workers how to use AI tools. It is to develop people who can combine AI capability with the deep sector expertise required to apply those tools effectively.

For traditional industries, this combination of operational experience and digital capability will become an increasingly important source of competitive advantage.

Expand the capacity to adopt

The most consequential finding for APEC is the divide in the capacity to adopt AI.

Large enterprises generally have greater access to the data, specialist skills, infrastructure, and capital needed to experiment with and deploy AI. The experience of Rio Tinto, Mitsui, UPS, and Antofagasta demonstrates what becomes possible when these foundations are already present.

The starting point is very different for many smaller businesses and grassroots producers.

Banyu and FisTx work with Indonesian aquaculture producers where connectivity, data collection, and digital capability may need to be strengthened before AI can add value. Their experience demonstrates why AI sometimes needs to be sequenced behind more fundamental capacity-building rather than introduced as the first intervention.

The challenge also exists in more AI-ready economies. eShepherd identified affordability, understanding of the technology, and limited public support for initial deployment as barriers for smaller agricultural producers.

The cases therefore suggest that the emerging AI divide is not simply about access to technology. It is increasingly about whether businesses have the capacity to put that technology to work.

This distinction matters particularly for MSMEs. As Section 3 shows, many smaller firms will access AI through platforms, software, and specialist providers rather than developing it themselves. That can significantly lower the technical barrier to adoption, but it does not eliminate the need for digital capability, skills, appropriate data, and the resources to change business processes.

Without those foundations, the firms and economies already best equipped to adopt AI are likely to capture its benefits faster than those starting further behind.

Closing the adoption-capacity gap is therefore central to ensuring that AI contributes to broad-based productivity and growth across APEC.

From technology readiness to adoption readiness

Taken together, the five findings shift the focus of the AI debate.

The availability and capability of AI technology will continue to advance rapidly. But the cases in this report suggest that technology alone is not the principal determinant of whether businesses capture its value.

The critical question is increasingly one of adoption readiness: Does the business have a problem worth solving, data that can support the solution, an enabling policy environment, people equipped to use it, and access to the capabilities required to deploy it?

These conditions are mutually reinforcing. Strong data without skilled people will not produce sustained adoption. A technically successful application may not scale without an appropriate regulatory or commercial pathway. And smaller firms may struggle to participate even when proven AI solutions are readily available.

The next phase of AI adoption across APEC will therefore depend not simply on making AI more powerful, but on making more businesses ready and able to use it.

Section 5 examines how these conditions play out across individual sectors before Section 6 turns to the actions APEC Leaders and Digital Ministers can take to accelerate adoption. 

The conditions for successful AI adoption are broadly consistent across the case studies, but how those conditions are applied differs significantly by sector.

In mining and heavy industry, safety, accountability, and operational resilience are critical. In logistics and maritime transport, interoperability and the movement of data across complex systems matter greatly. In health care, regulation and reimbursement can determine whether adoption is commercially viable. In agriculture and aquaculture, affordability, connectivity, and producer readiness set the pace of uptake. In manufacturing and other industrial applications, AI is increasingly being embedded directly into production systems.

There is therefore no single model for AI adoption across APEC. The common conditions identified in Section 4 need to be applied to the operational realities of individual sectors.

Mining: Productivity gains require strong operational guardrails

Mining demonstrates both the potential value of AI in complex industrial environments and the importance of governing where and how it is used.

Antofagasta Minerals provides a clear example of AI applied at scale in mining. Its SIRO recommendation systems support the optimization of processing steps, such as leaching and grinding. Its Digital Academy also illustrates the workforce dimension of adoption, helping operators develop the capabilities needed to work alongside new digital systems.  

Rio Tinto is building on decades of automation and technological innovation to apply AI across mining operations. Applications include modernization of systems, machine learning to support mineral exploration, AI to optimize equipment dispatch and autonomous truck movements, and predictive analytics to improve equipment reliability. These examples demonstrate how AI can support value creation and productivity improvements in complex industrial environments where safety and risk management are critical.  

The sector lesson is that AI can create significant value in heavy industry, but adoption must be matched by governance appropriate to the operational consequences of failure.

Logistics and maritime transport: Interoperability is central to scaling adoption

Logistics and maritime transport demonstrate the value of AI in managing complex flows of goods, information, and decisions across organizations and borders.

UPS introduced machine learning into customs classification as United States import requirements became more complex, helping customs clearance recover to around 99 per cent. The company also reports that 97 per cent of its shipments clear customs on the first day of entry. UPS identifies greater adoption of paperless customs as an important means of enabling further automation across international logistics.  

Mitsui O.S.K. Lines takes a broader transformation approach. AI is set within its long-term management strategy, with applications including voyage optimization and knowledge transfer. The company has set a target of reducing time spent on repeated tasks by 20 per cent by 2030 as part of efforts to transfer knowledge from experienced crews to younger employees.  

These cases point to a broader sector issue. Logistics and maritime transport are inherently interconnected and cross-border. The value of AI therefore depends not only on what an individual firm can automate, but also on the quality and interoperability of the data and processes connecting firms, ports, customs authorities, and other parts of the supply chain.

In logistics, AI adoption can progress only as fast as the systems around it allow.

Health care: The adoption bottleneck may lie outside the technology

Health care provides the clearest example in the study of technically successful AI depending on the surrounding regulatory and financial environment.

EverEx's MORA platform uses AI-based pose estimation through the camera of a phone or tablet to guide and monitor home rehabilitation without additional hardware. A randomized controlled trial involving more than 200 patients in Korea showed significantly improved outcomes compared with usual treatment.  

The technology alone, however, did not determine commercial viability. Adoption depended on a reimbursement pathway that funded the clinician effort supporting the service. Once such a pathway became available, it helped turn a technically and clinically viable application into a viable delivery model.  

The health care lesson is particularly important: successful innovation can stall if regulation, approval, and payment models do not evolve alongside the technology.

For AI-enabled health care, adoption therefore depends not only on technical performance and clinical evidence, but also on regulatory pathways, assurance, and mechanisms that allow providers to be compensated for providing the service.

Agriculture and aquaculture: Adoption must start from the producer's reality

Agriculture and aquaculture show both the potential for AI to reach grassroots producers and the importance of understanding their starting conditions.

eShepherd demonstrates how AI can be integrated into an existing agricultural platform. It uses AI across customer support, warranty assessment, and multilingual onboarding as it scales its virtual-fencing platform internationally. Warranty processing fell from four weeks to five days, while the platform supports farmers in more than 10 languages.  

Banyu and FisTx illustrate a different adoption pathway in Indonesian aquaculture. Banyu sequenced AI behind more fundamental work on seaweed biotechnology and improved seedlings, which raised farmer yields by about 20 per cent. FisTx first applied AI internally in areas such as budgeting, procurement, and product research before placing AI-supported tools in farmers' hands. Its farm success rate rose by roughly 20 percentage points.  

Both cases involve producers with limited connectivity and relatively little prior exposure to digital tools. Their experience shows why adoption may need to begin with foundational capability, demonstration of value, and trusted support rather than with the technology itself.  

In grassroots sectors, the first barrier to AI adoption may not be AI at all. It may be connectivity, data, skills, affordability, or confidence in the value of the technology.

This makes agriculture and aquaculture particularly important to the report's inclusion argument. The businesses and producers that could benefit significantly from AI may also be those least able to adopt it without external support or enabling platforms.

Manufacturing, drone logistics, and new energy: AI is moving into core operations

The manufacturing, drone logistics, and new-energy cases show AI moving beyond experimentation and into daily industrial operations.

Xiaomi applies AI across materials discovery, quality inspection, and production control in automotive manufacturing. Its quality-inspection process reduced inspection time from 20 seconds to two seconds, achieved accuracy above 99.9 per cent, and contributed to an approximately eight per cent increase in manufacturing efficiency.  

Phoenix Wings applies deep-learning perception, vision-language models for scene understanding, and a large-language-model-based flight assistant in low-altitude drone logistics. SmartMore develops IndustryGPT and industrial inspection systems for new-energy battery production, including blue-film inspection operating at 0.5 seconds per unit and reducing the key-defect miss rate from about 0.3 per cent to near zero.  

These examples illustrate a different stage of adoption. AI is no longer simply an isolated tool or pilot; it is increasingly being integrated directly into production, inspection, and operational systems.

At an industrial scale, the value of AI comes increasingly from embedding it into the workflow rather than treating it as a standalone technology project.

Sector differences, common foundations

The cases demonstrate that the fundamentals of successful adoption are consistent, but their relative importance varies by sector.

Safety and accountability are particularly important in mining and industrial operations. Interoperability and cross-border processes dominate logistics and maritime transport. Regulatory and reimbursement pathways can determine viability in health care. Affordability, connectivity, and digital readiness are especially important in agriculture and aquaculture. In manufacturing and new-energy applications, integration into operational systems becomes increasingly important as AI moves to scale.

APEC's challenge is therefore to establish common conditions for AI adoption while allowing those conditions to be applied in ways that reflect the realities of individual sectors.

This sector perspective also reinforces a broader conclusion from the case studies: adoption policy should not assume that every industry will follow the same pathway or face the same bottleneck. What is common is the need to identify and address the constraint that prevents proven technology from being put to productive use.