AI is integral to Southeast Asia’s future, but it is still in an early stage

AI covers a broad set of intelligent applications

AI is the discipline of making analytical machines intelligent, enabling an organization to function appropriately and with foresight. AI can be embedded in a broad range of applications, many of which we considered in our study (see figure 1).

One familiar application is the AI-enabled customer service chatbot. AI also extends to newer human augmentation technologies that enable automating business processes to onboard new employees, provide financial advice, or act as a personal assistant. And, of course, AI will include applications that strive to emulate human behavior in complex, unpredictable environments, such as autonomous driving.

AI adoption is still in nascent stages in Southeast Asia

The advent of Industry 4.0 has spotlighted the need to accelerate technology adoption, especially AI, to futureproof economies. However, our study shows that Southeast Asia is still in the early stages of AI adoption (see figure 2).

The embrace of new technologies typically follows an S-curve, beginning with innovators and early adopters. When adoption hits an inflection point, it triggers exponential growth as the technology’s potential value becomes clearer. Our study indicates that AI adoption in Southeast Asia is still in the early part of the curve, with 30 percent of respondents developing their AI strategies or just starting to invest (see appendix). About 50 percent of companies are at least piloting some AI initiatives. Only 15 percent are in advanced stages of AI implementation, typically those in more service-oriented sectors. When it comes to investment, 83 percent are devoting less than 0.5 percent of their revenues to embedding AI solutions into their operations.

Although investments in AI solution providers have grown steadily, the region is lagging more advanced countries by two to three years (see figure 3). The region received AI investments of only $2 per capita between 2015 and 2019—an amount that is dwarfed by the per capita investments of $155 and $21 respectively for the United States and China. Singapore is the region’s standout with $68 per capita investment.

Even though AI investments are concentrated in Singapore, nine of the top 10 deals were for Singapore-based start-ups with businesses and use cases across the region. For example, Biofourmis, a health analytics platform that analyzes physiology data from clinical-grade wearables, operates in Singapore and Indonesia. Similarly, Tookitaki, an intelligent decision-support system for compliance programs in the financial services sector, operates across the six larger Southeast Asian countries.

The region is also gaining AI research attention from international companies. American software firm Salesforce opened its first overseas AI research center in Singapore last year, joining the ranks of YITU Technology, Alibaba, Dyson, and DataRobot.

AI has the potential to add $1 trillion to the region’s GDP

More than 70 percent of our survey respondents see AI as crucial to Southeast Asia’s future and point out that AI adoption needs to be accelerated.

Our study reveals that AI can have a strong overall impact: a 10 to 18 percent GDP uplift across Southeast Asia by 2030, equivalent to nearly $1 trillion (see figure 4). The impact variance seen across countries is the result of the different sectorial make-ups and the relative maturity of each country’s AI infrastructure and adoption readiness. For instance, Singapore’s GDP is predominantly made up of service sectors and has a highly digitized, automated, and efficient economy that is well positioned to capture the full potential of AI. For example, AI will deliver a bigger economic impact in Singapore than it will in Cambodia, which is more focused on primary sectors and is still at an early stage of AI adoption.

Regional governments have started to recognize how important AI is to their economies. Singapore launched its National AI Strategy in 2019, and Malaysia and Indonesia are expected to follow suit this year. Malaysia already announced a $1 billion project to build its first AI park—a partnership between a local company, G3 Global, and China’s SenseTime and China Harbour Engineering.

Given the value at stake, now is the time for all governments and companies to turn their attention to compelling AI strategies. As AI continues to evolve, it will transform Southeast Asia’s top five sectors (see figure 5).

The benefits of AI can be transformative, but it requires a razor-sharp focus

AI is not a panacea, and the focus needs to be on specific use cases

Given its high-impact potential, AI can be viewed as a panacea for many business problems. As a result, some early adopters are applying AI solutions without carefully considering either the specific problem they are trying to solve or its local context.

Consider the example of a large Southeast Asian telecom group that implemented an AI-powered know-your-customer procedure for its mobile wallet business only to discover it resulted in low autoapproval rates. They later realized that the solution they were using was built using primarily American and European datasets, which did not apply well to the Southeast Asian context. After several iterations with localized datasets, they saw a huge improvement in performance.

Most importantly, AI solutions need to be focused on specific use cases that solve business problems and deliver economic value in the near term, such as contact center intelligence, personalization, forecasting, and intelligent document management. Our study indicates that 80 percent of AI’s potential value comes from less than 20 percent of use cases. Many companies are implementing AI without clearly identifying high-impact use cases or getting business stakeholders on board. In fact, full involvement of business owners at each step of the use case deployment is crucial to maximizing the value of AI.

At the aggregate level, use cases in sales and marketing as well as in supply chain management deliver the biggest impact. For example, many consumer packaged goods companies forecast inaccuracies because of high demand volatility from stock-keeping unit proliferation. To address this, one major company in Southeast Asia used machine learning-based predictive analytics and pattern recognition to augment its statistical model with various external data such as point of service, price, promotions, and social media posts. This improved forecast accuracy by 4 to 20 percent and lowered inventory safety stock by more than 12 percent.

Demonstrating a near-term impact is also vital to creating the organizational conviction and commitment needed to scale up and deploy AI more widely across the entire business value chain.

Revenue management is the starting point for Southeast Asia

Our study reveals that Southeast Asia has an overwhelming focus on using AI to manage revenues. In fact, revenue management is getting two to three times more attention than cost management across all sectors (see figure 6). This contrasts with other regions such as Europe, where around 70 percent of companies focus on using AI for productivity and efficiency. This seems to be because of both the nascent stage of AI adoption in Southeast Asia and the region’s fewer opportunities to optimize labor costs. Consequently, AI-led topline solutions tend to attract more attention.

Going forward, we expect to see the region’s AI priorities shift toward optimizing the cost base as organizations familiarize themselves with AI and become more aware of its potential to enhance other value chain areas. Not surprisingly, this will also lead to greater use case differentiation by industry. For example, the agriculture and retail industries will focus more on supply chain, operations, and procurement, while energy and public-sector organizations will pay more attention to backend optimization in HR and IT. AI adoption in manufacturing is in a nascent stage, but this might reflect Southeast Asia’s high number of small to medium-size enterprises (SMEs), which make up 95 percent of all manufacturing companies. SMEs tend to be less able to adopt more advanced solutions.

Even within the same function, AI investments are distributed unevenly. In operations, for example, the region’s manufacturing firms are keen on using AI for robotics and automation. However, AI-enabled predictive maintenance can make a bigger impact. This reinforces the need for both a more exhaustive business-focused cost–benefit analysis of all current and new use cases and a comprehensive implementation road map.

AI can impact the entire value chain

Although there are specific industries and use cases that can benefit, AI has the potential to deliver a substantial impact across the entire value chain. Based on the use cases we examined, AI is likely to have the most impact in two areas: supply chain management and sales and marketing. More than 70 percent of all incremental AI value in Southeast Asia could be attributed to these two areas (see figure 7).

However, within each function, some use cases have more potential than others. AI adopters have reported significant benefits from implementing AI across the value chain (see figure 8). Leaders in Southeast Asia have seen a significant upside from sales and marketing use cases, in particular from predictive upselling and cross-selling, customer life-cycle management, and the right product recommendations.

In sales and marketing, NPTB and individualized offerings are huge drivers of value. Replacing basic recommendation engines, which cannot scale, with personalized recommendations powered by machine learning can greatly improve digital engagements, conversions, and revenues. Machine learning can help create and deliver high-quality recommendations that respond to specific needs, preferences, and behaviors of customers. When successfully implemented, AI can increase revenues by 21 to 30 percent (see Case study: Tokopedia and Appier).

One of the highest-impact use cases for supply chain management is inventory optimization (see Case study: Singapore Airlines). By implementing machine learning, businesses can more accurately forecast customer demand for sales and capacity planning, which can also help reduce waste. Machine learning-based forecasting, which combines historical data with a continual influx of new information, is a significant improvement over simple spreadsheets, which rely solely on historical data. AI-based inventory optimization solutions can lead to a 21 to 30 percent reduction in stock-outs.

Beyond these two areas, we also see substantial pockets of value in other functions. In HR, for example, AI can help organizations increase productivity in recruitment, engage employees, and reduce employee attrition (see Case study: Pulsifi). AI can also be used to improve customer experiences (see Case study: Parcel Platform).

Given the value at stake, the cost of focusing on the wrong AI use cases can be significant. Forward-thinking organizations reassess their AI efforts and prioritize the high-impact use cases that directly address their specific problems.

The challenges to AI adoption are well-founded, but the walls are crumbling

Five issues are hindering AI adoption in Southeast Asia

The benefits from AI use cases are clear. The current adoption rate, however, says otherwise. We have identified five issues that are prevalent across all sectors (see figure 9).

1. Overemphasized concerns for a talent gap

Respondents and interviewees report a mismatch between skills and AI requirements. More than 85 percent highlight difficulties in finding and attracting technical talent that can work on AI-related activities, especially as they are competing with technology companies, especially hyperscalers and start-ups. Interviews also reveal that technical talent often lacks the domain understanding to apply AI to crucial business problems. However, we believe this talent gap is overemphasized: the actual technical skill requirements for developing and implementing AI solutions have become less complex, given the increasing availability of easy-to-deploy AI models from leading AI solution providers.

2. Fragmented and nascent AI ecosystem

More than three-quarters of respondents (76 percent) see the AI ecosystem in Southeast Asia as immature and fragmented—particularly outside Singapore— making it difficult for most companies to find and work with suitable AI providers. This impacts both companies’ willingness to experiment with use cases and AI providers’ ability to develop new applications.

3. Evolving data governance and infrastructure

More than 50 percent of respondents feel that companies’ data quality, infrastructure, and governance are inadequate. Disjointed data collection, lack of quality standards, and antiquated data infrastructure cause challenges in harmonizing and extracting data, creating potential security concerns. This fundamentally slows down the adoption of AI.

4. New regulations that impact AI development

Regulatory constraints—including privacy protection, data-sharing restrictions, on-premise storage requirements, and mandated AI model transparency— are an understandable yet significant bottleneck to AI adoption, especially in Indonesia. Three-quarters of respondents believe that more harmonized regulations across Southeast Asia will accelerate AI investments, addressing in-country AI adoption and facilitating cross-border data flows for the region’s interconnected businesses and value chains.

5. User resistance to AI

AI will impact jobs, and the pace of change can be worrisome for some. In fact, our interviews exposed two opposing scenarios: either a gradual replacement in line with previous technology adoptions or a catalytic displacement driven by an unprecedented, wide-ranging impact of AI on most industries. Regardless of the scenario, 33 percent of respondents are concerned about employee backlash from job displacement.

A comprehensive improvement agenda will help capture AI’s full potential

Resolving these issues will require concerted actions at both the organizational and national level. Inspiration can be found in how some companies and governments are already addressing them.

1. Democratize AI to resolve the talent gap

The AI talent gap is often overestimated. While it is true that data scientists are in short supply, many AI solution providers have ready-made, easy-to-deploy models that do not require deep expertise or depend on external talent. With appropriate training, many science, technology, engineering, and mathematics (STEM) graduates will be able to transition to AI-related jobs such as data preparation, data visualization, and data management. Upskilling the workforce as well as acquiring external talent for select roles can also address the talent gap.

For example, the national initiative AI Singapore has an apprenticeship program designed to facilitate reskilling and outplacement into appropriate AI-related jobs. The program grooms talent with a nine-month classroom-based program, followed by on-the-job training with an industry partner to create a real-world AI use case. In addition, there are a number of publicly available training courses that allow anyone to easily acquire AI-related knowledge and improve their AI literacy.

Many interviewees pointed to a lack of business or domain knowledge as a bigger challenge than technical skills. AI leaders focus on continuously co-creating new use cases with immersion sessions that combine business and technical staff. For example, DBS Bank collaborated with AWS DeepRacer League to educate 3,000 staff about AI and machine learning. By raising awareness about AI and combining business understanding with technical know-how, the company was able to identify and develop more new AI use cases.

Rather than focusing on searching and competing for deep technical AI talent, companies would do well to invest in upskilling their existing talent who are interested in AI because their understanding of the business issues is invaluable.

Finally, technology partners play a significant role in helping companies deploy AI solutions faster and more effectively, reducing the need for specialized talent across domains. Leading providers have established relationships with systems integrators, software vendors, and software-as-a-service providers offering a wide range of tools, features, and models.

As a result, the perceived and often overemphasized talent gap is emerging as less of an obstacle to AI adoption in many businesses.

2. Systematically stimulate AI ecosystem and testbed development

The immaturity and fragmentation of Southeast Asia’s AI vendor ecosystem make it difficult for many companies to identify providers that have the relevant AI models and business understanding. Furthermore, many potential AI users are inexperienced and looking for solutions that have proven to be successful before. This leads to a chicken-and-egg situation, particularly in parts of the value chain where use cases are nascent or specific to local business problems. Finally, large enterprises are predisposed to working with large vendors and are often not set up to work with smaller providers because of their procurement protocols. Larger enterprises can address these challenges with their in-house teams and infrastructures, but most SMEs cannot afford to do this. Cloud providers help lowering infrastructure costs, yet operational issues still require investing in local use cases.

Here, government-supported testbeds (test and development setups) can play a crucial role in stimulating local use cases, generating rich and context-specific datasets for iterative testing of models, and giving SMEs access to the AI ecosystem without immediate or major financial commitments. Testbeds also demonstrate the intricacies of AI’s impact on society by enabling data-driven decision-making.

Building a testbed is difficult because it needs to support the development of real, scalable solutions and not merely serve as a technology showcase. By studying successful testbeds such as Dublin Smart Docklands, Masdar City, Britain’s National Health Service, Manufacturing Quality Management in China, and Water Security in the United States, we discovered five common factors:

a. Real commitment. Dedicated test areas, financial contributions, and regulatory exemptions

b. Results-orientation. Not just a showcase, but focused on actual business outcomes

c. Multi-stakeholder. Active and coordinated involvement from government, companies, start-ups, academia, and society

d. Scalable approach. Progressively expanding from sandboxes to broader, public environments

e. Trial of real ecosystem changes. Intended to test and revise existing rules, regulations, and methods while continuing to ensure safety and security

A good example of such a comprehensive testbed is Singapore’s autonomous vehicle testing (see Case study: Autonomous vehicle testing in Singapore).

For factors such as regulations and quality, relevant data are especially crucial to AI testbeds and will be discussed further below.

3. Change the approach to data in the near term

Given that about 80 percent of AI activity is spent on basic data processing, having an adequate data structure and governance will accelerate the ability to adopt AI.

Data collection is often complex and time consuming. Companies should focus on collecting only the data that business stakeholders need (see Case study: AWS Registry of Open Data and Amazon SageMaker Ground Truth). Data infrastructure projects should no longer be managed in a traditional IT way—as multiyear “fixing of building blocks.” Instead, companies should follow a two-speed approach of balancing business requirements with systematic infrastructure building and legacy overhaul. Other important factors such as security, legal, and regulatory requirements should be addressed proactively.

4. Proactively engage regulators to address AI adoption

Our interviews reveal a number of regulatory areas that need to be addressed. Regulators need to be updated on the latest AI developments so they can revise regulations accordingly. Understandably, they are concerned about privacy protection, security, and data sharing, but they need to take a forward-looking approach to support the development of new AI solutions.

AI governance efforts should take a risk-based approach to addressing the responsible use of AI, taking into account the nature and context of the AI deployment. A one-size-fits-all approach risks underregulating highly impactful use cases, which creates potential for real harm, and overregulating the inconsequential, which hinders innovation and adoption of technology that can benefit people’s lives. This also reaffirms the fundamental importance of testbeds, not only for developing innovative solutions, but also for regulators to understand and test new regulatory boundaries.

Some countries, such as Singapore and Malaysia, have started to develop national AI strategies to balance ecosystem developments and regulatory approaches while relying on the private sector for input and risk understanding. Given that 23 percent of trade is inter-ASEAN, AI regulations should be aligned and harmonized across the region, similar to the approaches being pursued for cybersecurity and cross-border data flows. The European Union’s General Data Protection Rule, which is standardized across its 28 member countries, takes such an approach by giving citizens control of their personal data while simplifying the regulatory environment for businesses.

In addition to aligning on AI regulations, countries should commit to enabling the cross-border flow of data. Data localization or residency requirements hinder the development and use of AI, especially where data scientists need to have access to significant computing power that may be hosted in another country or require access to a large variety of datasets from multiple jurisdictions.

Depending on the circumstances, a business may decide to keep its data in one location, in which case federated learning can help (see Case study: Federated learning). With federated learning, AI models are trained with decentralized data: the AI model is downloaded and trained locally, and then updates are consolidated to improve the shared model.

5. Build a future-ready workforce

Although AI can have a catalytic effect on the workforce, we believe its impact will be more gradual. For the most part, AI will replace tasks, not jobs, augmenting human actions so that people can be more productive and make better, more data-driven decisions and will give rise to other “builder” jobs such as data scientists and translators, which are in scarce supply in ASEAN. Tomorrow’s workforce must be prepared to harness the power of AI in their daily activities.

Policymakers and business leaders both have a role to play in ensuring that employees remain relevant. Reskilling and upskilling programs are essential to help employees prepare to work with AI. Government-backed initiatives such as the United Kingdom’s National Retraining Scheme help prepare people for the future of work by strengthening their technical skills via both online and in-person training. Businesses can also tap online providers such as Udemy or Coursera to educate their employees on the fundamentals of AI.

Although reskilling and upskilling are important, what’s even more important is bringing about long-term changes in the education system. Unleashing AI will require a broad spectrum of talent that can research, build, train, maintain, and monitor the technology. This will entail systemic changes aimed at expanding and strengthening the STEM education system.

Across the region, governments have been encouraging STEM studies in a bid to bridge the talent gap. In Malaysia, the Ministry of Education’s STEM4All initiative aims to fuel students’ interest, expand access to STEM subjects, and evolve STEM to STREAM (science, technology, reading, arts, and mathematics). In Vietnam, demand-oriented technical and vocational education and training programs provide an alternative to college. STREAM programs designed for formal skills development in areas such as programming will result in a strong technology base. One challenge is local languages, particularly in areas such as conversational AI. Most platforms and code bases are still English-centric, limiting their usefulness for otherwise well-educated technical workforces in Thailand and Vietnam.

While governments are well aware of the talent gap, most efforts continue to be focused on reskilling. Addressing the issue will require a more comprehensive effort to change countries’ education systems.

Every stakeholder has a role to play in unlocking the benefits of AI

To resolve the issues, national AI strategies and skill-building programs are a good start, but the momentum should continue. Both AI users and solution providers need to be more focused on the business impact. Investors can also play a more active role in disseminating AI across their portfolio companies, and academia should focus research on industry applications.

1. Governments

• Develop a national AI agenda that prioritizes the adoption of AI in key sectors. Governments can lead the charge in AI adoption to ensure the efficient allocation of resources and extract maximum value while creating a strong ecosystem to deploy AI across the wider economy.
• Promote long-term changes in education to build a future-ready workforce. To address the talent gaps and reskill low-skilled workers who may be at higher risk of displacement, future-proofing the workforce is essential through an adaptive education system and enabling a systematic shift to activities that add more value.
• Expand the use of testbeds to stimulate AI development while minimizing risks. Testbeds are the most effective ways to address the uncertain implications of AI, foster innovation, and facilitate adoption. Governments should work closely with regulators, users, and solution providers to ensure these testbeds are effective for piloting business driven use cases and to help make them pervasive across businesses, sectors, and society.

2. Business users of AI

• Prioritize AI initiatives and focus on the near-term business impact. Businesses need to prioritize use cases that solve operational issues and have the potential for a tangible, near-term impact. This is the only effective way to both foster quick business buy-in and build momentum for deploying AI at scale.
• Streamline and establish new ways of working. Adopting and scaling AI requires organizational changes, including agile operating models, aggregation of feature sets to scale use cases, automated closed-loop test beds, and linkage of AI-models to other automated workflows.
• Fix data infrastructure and governance issues. A more radical approach is recommended for this; businesses cannot wait for a step-by-step overhaul of legacy systems. Improving data needs to start with a focus on use cases that deliver near-term value.
• Adopt a business- and employee-centric approach to AI. AI should not be sparking a search only for data scientists. Rather, it is about building, reskilling, and augmenting the relevant AI skills in the existing workforce to help them solve crucial business problems. This approach also ensures that employees remain relevant and do not resist the adoption of AI.

3. Technology and AI providers

• Adopt a business model-centric approach to AI. Solution providers need to focus on value propositions with a clear near-term business impact instead of developing potentially breakthrough technologies with uncertain paths to monetization.
• Reassess delivery models. Solution providers need to work closely with the business and IT side of their customers to achieve the expected impact. An effective mix of on-site and off-site presence is essential, and solution providers will need to work with the right local partners to effectively deliver and build a local AI setup in the region.

4. Investors

• Broaden investment areas. There are huge pockets of value yet to be captured across various use cases, sectors, and regions. Investors can benefit from widening their investment landscape.
• Play an AI activist role across the portfolio. Investors can add more value by opening their AI networks to portfolio companies, thereby improving their understanding of AI. For example, global investors could stimulate AI development in Southeast Asia by connecting their portfolio companies with AI hubs in China, India, or the United States.

5. Academia

• Align research interests with industry and societal needs. To achieve widespread AI adoption, business and research interests must be closely aligned. Academia should not limit research to current applications; there needs to be a focus on addressing the current and upcoming societal and market needs.
• Adopt an industry-focused curriculum. Students should be educated and equipped with the right knowledge and skills that are aligned with industry needs. It’s not about developing data scientists and AI experts; it’s about developing talent who can use their technical skills to address business problems and opportunities.

Across all of these stakeholder actions, one theme prevails: focus on the business impact. The value of AI is not in the technology itself but in how it can solve business problems and deliver value in the near term.

Given that AI adoption in Southeast Asia is still nascent, focused and concerted action can unlock the potential of AI and capture the projected $1 trillion boost to the region’s GDP by 2030. All stakeholders—governments, businesses, AI solution providers, investors, and academia—will need to work in tandem with a common focus on driving the business and social impact.

Appendix

Methodology

Our model is based on a bottom-up approach that models the potential economic impact at a use-case level and then aggregates the impact up to the economy level. The economic-level aggregation of use-case impact is moderated and adjusted based on industry- and country-specific adoption factors such as digital maturity, network readiness, and economic productivity.

The value impact model has four components:

i. Gross value add. We measured the contribution of key economic sectors based on two primary drivers of value add: sector sales and sector input. Data from the Economist Intelligence Unit, IHS Markit, and World Bank was used to establish the baseline GDP in 2030 at $7.5 trillion (before the full potential of AI is realized).

ii. Impact from AI. We sized the impact on value-add at an industry sector level through two categories of AI use cases: revenue enhancements and cost efficiencies. These estimates are informed by survey responses from more than 110 business executives in the region as well as lessons learned from more than 50 industry use cases. Industry specific adoption rates for each use case are factored in based on current levels of industry focus. We estimated the incremental full economic impact from AI to be $1.2 trillion.

iii. Other effects from AI. We applied additional adjustment factors to account for other positive effects from the implementation of AI, such as additional investments that companies make to support AI implementation, investors’ investments in AI providers, and induced consumption due to the economic value that AI could generate.

iv. Country-specific adjustments. Our model accounts for cross-country variances in digital adoption, network readiness, and economic productivity to simulate the realistic adoption of AI through an S-curve model. Data from World Bank and the World Economic Forum were used as inputs into the S-curve model. As with most technologies, the adoption of AI is likely to follow an S-curve pattern with a slow start followed by accelerated adoption as the technology matures. Applying this moderating factor to our results on the full impact potential of AI, we estimate the final gross economic impact of AI to be around $1 trillion.

This approach assumes that most of the economic value that AI generates in Southeast Asia will remain in the region and that leakage to other regions will be small, given that AI benefits are realized mostly in the local economies and most AI implementation will be carried out locally.

Finally, our approach assumes that the productivity benefits of AI can be fully realized at the national economy level, where labor can shift from activities impacted by AI toward growth sectors and new jobs created by AI.