Artificial intelligence is increasingly moving from an experimental technology to an operational tool for small and medium-sized enterprises across Southeast Asia. Yet the more important question is not whether SMEs will use AI, but where AI can create measurable business value without adding disproportionate complexity, cost, or risk.

This distinction matters because Southeast Asia’s SME landscape is unusually diverse. The region includes highly digitised economies such as Singapore alongside markets where many smaller businesses are still building basic digital infrastructure. As a result, AI adoption is unlikely to follow a single regional trajectory.

According to the Asian Development Bank’s Asia Small and Medium-Sized Enterprise Monitor 2025, MSMEs account for an average 99.8 per cent of enterprises and 67.6 per cent of employment across its 26 developing member economies. For Southeast Asia, therefore, the economic significance of AI depends heavily on whether smaller businesses can convert increasingly accessible AI tools into productivity gains.

AI’s first advantage is operational, not transformational

For many SMEs, the immediate value of AI lies in improving existing processes rather than reinventing the business.

Customer service, finance and accounting, marketing, inventory management, demand forecasting and internal knowledge management are among the functions where AI can reduce repetitive work or accelerate decision-making. Singapore provides a useful illustration. Data from the country’s Infocomm Media Development Authority shows that SME AI adoption increased from 4.2 per cent in 2023 to 14.5 per cent in 2024. SMEs using AI-enabled solutions supported through the Productivity Solutions Grant reported average cost savings of 52 per cent in 2024.

Also Read: AI boom pushes chipmakers to expand across Asia Pacific, FedEx report finds

The significance of this example is not the specific percentage. It demonstrates that AI’s commercial case for smaller companies can begin with relatively narrow applications. A company does not necessarily need to build a proprietary AI system. It can use existing models to automate repetitive workflows, analyse information, assist employees or improve customer interactions.

This lowers the technological barrier, but it does not eliminate the organisational one.

The real constraint is digital readiness

AI systems depend on the quality of the processes and data around them. A company with fragmented spreadsheets, inconsistent customer records and largely manual workflows may find it difficult to obtain reliable results from increasingly sophisticated AI tools.

Recent research published by ADB on digital technology adoption among 681 entrepreneurs across Indonesia, Malaysia, the Philippines, Singapore, Thailand and Vietnam found that selected digital applications and business-model digitalisation were associated with greater experimentation and changes in business models.

This suggests an important sequencing issue. AI is often presented as the next step after digitalisation, but for many SMEs, basic digital infrastructure remains the prerequisite for meaningful AI adoption.

A 2026 analysis published through ADB’s Southeast Asia Development Solutions identifies substantial differences in AI readiness across the region, with infrastructure, skills and regulatory maturity influencing adoption. It also notes that Cambodian manufacturing SMEs, for example, continue to face gaps in basic digital systems.

Consequently, an SME that has not yet digitised its core operations may receive more value from improving its data and workflow infrastructure than immediately implementing advanced AI.

Southeast Asia will not have one AI adoption curve

The region’s diversity also means that national AI strategies and enterprise capabilities will shape adoption differently.

Singapore has developed comparatively mature AI infrastructure and policy support, while other Southeast Asian markets continue to address foundational connectivity, skills and digitalisation challenges. The result is likely to be a multi-speed AI economy, rather than a uniform regional transformation.

Also Read: AI chats are becoming the new starting point for online shopping

This matters for enterprise software providers and policymakers. A sophisticated AI platform designed for a highly digitised Singaporean SME may not address the constraints of a smaller business in Cambodia, Laos or parts of Indonesia. Solutions need to account for differences in connectivity, workforce capabilities, language, affordability and existing software systems.

The Asian Development Bank similarly argues that AI readiness, rather than adoption alone, will determine how effectively economies capture the technology’s benefits.

The SME advantage could be speed

There is, however, a structural reason to expect smaller businesses to remain important participants in AI adoption.

Large organisations often have more resources but also have complex legacy systems, multiple approval layers and established processes. SMEs have fewer resources, but they can sometimes introduce new tools across a smaller organisation with less organisational friction.

AI can therefore reduce some of the disadvantages traditionally associated with company size. A small retailer can use automated customer support. A manufacturer can apply predictive analytics to production data. A professional-services firm can use AI to organise documents and accelerate research. A small exporter can use AI-assisted translation and market research to communicate across markets.

This does not make AI a substitute for scale. Rather, it can lower the cost of performing certain activities that previously required specialised staff or external providers.

Also Read: AI labs are chasing a slice of the corporate pie in Southeast Asia

The productivity question is more complicated

There is a tendency to equate AI adoption with productivity automatically. The evidence suggests a more conditional relationship.

AI can reduce the time required for particular tasks, but the resulting productivity gain depends on what employees do with the time released. Poorly implemented automation can simply move work elsewhere, introduce errors or create additional verification requirements.

This is particularly important for SMEs, where a single employee may perform several functions. Automating one task does not necessarily produce a proportional reduction in costs if the employee remains essential for related activities.

ADB’s broader research on digital transformation highlights both the productivity opportunities and the risks, including digital divides, labour-market disruption, cybersecurity, data privacy and governance challenges.

The economic impact of AI should therefore be measured through business outcomes such as processing time, error rates, customer response times, revenue per employee and operating costs, rather than simply the number of AI tools deployed.

Governance will become an SME issue

As AI moves into finance, recruitment, customer service and other decision-making processes, governance becomes increasingly relevant to smaller companies.

ASEAN’s Guide on AI Governance and Ethics provides a regional framework covering issues including transparency, fairness, security, accountability and human oversight. ASEAN subsequently expanded the guidance to address generative AI specifically.

For SMEs, governance does not necessarily require large compliance departments. It can begin with practical questions: What data is being provided to an AI system? Who reviews its outputs? Which decisions can AI make independently? What happens when the system produces an incorrect or biased result?

These questions become more important as AI moves from assisting employees to executing parts of business workflows.

From experimentation to infrastructure

The next phase of AI adoption among Southeast Asian SMEs will be less about experimenting with individual tools and more about integrating AI into core business operations.

The real value will come from connecting AI with customer data, finance, inventory and internal workflows. As adoption grows, success will depend not simply on access to AI, but on digital readiness, skills, governance and practical use cases.

The challenge is no longer accessing AI. It is turning AI into measurable, repeatable productivity.

—

Editor’s note: e27 aims to foster thought leadership by publishing views from the community. You can also share your perspective by submitting an article, video, podcast, or infographic.

The views expressed in this article are those of the author and do not necessarily reflect the official policy or position of e27.

Join us on WhatsApp, Instagram, Facebook, X, and LinkedIn to stay connected.

The post Beyond automation: How AI is reshaping the SME landscape in Southeast Asia appeared first on e27.

Malaysia has put a number on its startup ambitions: MYR450 million (US$110.4 million). That is how much the National Fund of Funds Jelawang Capital and Dana Perintis, the startup vehicle of the Retirement Fund (KWAP), will invest in developing startups, Prime Minister Anwar Ibrahim said in his Budget 2027 speech to parliament on Friday.

“Startups need support at every stage, from ideas to seed capital and scaling up,” said Anwar, who is also Finance Minister.

Also Read: Malaysia’s digital economy’s second wave looks nothing like the first

It is a welcome commitment. It is also, by some distance, not the biggest cheque in the speech. Read the budget end to end and a clear pecking order emerges: mid-tier companies and semiconductors first, research and AI skills next, and early-stage startups a little further down the list.

Follow the money

Anwar said MYR2.1 billion (about US$515 million) from the NIMP fund (linked to the New Industrial Master Plan 2030), KWAP’s Dana Pemacu and Khazanah’s MTC Fund will go towards strengthening mid-tier companies in electrical and electronics (E&E), digital technology, aerospace and medical devices. These firms, he said, “must rapidly become self-reliant to lead in high-value industries.”

Set side by side, the startup allocation is roughly a fifth of the mid-tier pot. Add the MYR1.2 billion (about US$294 million) Khazanah and KWAP will put into local semiconductor companies, and the MYR6 billion (about US$1.47 billion) earmarked for research, development and commercialisation across ministries, and the startup line starts to look like a footnote.

That is not necessarily a mistake. Mid-tier firms hire at scale, export and anchor supply chains, and Malaysia’s growth story rests on climbing value chains rather than minting unicorns. But for founders, the size of the fund matters less than who manages it and how quickly the money actually moves.

KWAP’s presence on both sides of the ledger is worth watching. The pension fund spent part of this year containing the damage from the eFishery fraud shock. Retirement money in venture capital invites political scrutiny, and whether that makes Dana Perintis a more disciplined investor or simply a slower one remains to be seen.

Nudging private capital off the sidelines

The more interesting moves may be the ones designed to pull private money in. Anwar announced MYR270 million (about US$66 million) to support SMEs, mid-tier companies and social enterprises raising through equity crowdfunding (ECF) and peer-to-peer (P2P) lending platforms.

Also Read: AI demand lifts Malaysia’s chip sector, but not every player wins

Income tax exemptions for individual investors on ECF platforms, and for angel investors backing early-stage tech startups, have been extended to December 31, 2030. That four-year runway matters more than it sounds. Angels plan around portfolios and holding periods, not single budget cycles, and a sunset clause that keeps moving is a poor basis for a long-term bet.

Cradle Fund, meanwhile, gets MYR41 million (about US$10 million) to implement the Malaysia Startup Ecosystem Roadmap and to hand out grants for prototype development and commercialisation. For a roadmap meant to reshape the national ecosystem, it is a slim line item. Its impact will depend on how tightly Cradle targets the grants.

Semiconductors, again

If there is a single obsession running through the speech, it is chips. MTDC’s SemiconStart programme will provide follow-on equity for startups in IC design, advanced packaging and semiconductor materials. Khazanah is partnering with the Selangor state government on a fund to develop the state’s semiconductor industry. Khazanah and InvestPenang will also set up a MYR100 million (about US$24.5 million) Strategic Investment Fund for early-stage semiconductor and advanced manufacturing firms.

The follow-on detail is the smart part. In deeptech, first cheques are rarely the problem. Startups stall at the second and third rounds, through long and capital-hungry design cycles. Firms such as GreatAsic, which is trying to pivot Malaysia from chip assembly to indigenous design, are exactly the kind that need patient capital. The risk is the old one: the sector has already faced questions over governance and credibility, and more state money will not settle them on its own.

AI for everyone, on someone else’s model

On AI, the budget leans heavily on access. The free AI subscription initiative will be expanded by another 100,000 subscriptions and extended to new applications such as ChatGPT. Rakyat Digital, meanwhile, offers free digital courses through a partnership with Coursera.

There is an irony here. Paying for subscriptions to an American chatbot sits awkwardly beside Putrajaya’s sovereign AI rhetoric. Local AI startups may wonder why public money is building habits on a foreign platform rather than theirs.

Also Read: Why Malaysia’s AI Nation 2030 plan matters for B2B startups

AI Malaysia Bhd will receive almost MYR15 million (about US$3.7 million) to reinforce a safe and ethical AI ecosystem, towards a target of 200,000 AI-skilled workers. MDEC’s MYR30 million (about US$7.4 million) fund will help 4,000 MSMEs adopt AI and automate their operations, and will also train and certify 5,000 AI professionals. Do the maths and that is at most MYR7,500 (about US$1,840) per business, before the training component takes its share. Given how wide Malaysia’s AI skills gap already is, that is a starter kit rather than a transformation.

Platforms, gig workers and quantum threats

The government and Grab will jointly fund a MYR160 million (about US$39 million) package from 2027 to boost net earnings and improve the welfare of e-hailing and p-hailing (parcel delivery) workers. It is a pragmatic arrangement, though one in which the regulator and the dominant platform now share the bill for gig worker protection. Riders will judge it by what lands in their pockets, not by the press release.

On e-commerce, the government is examining legislative and enforcement measures to shield local MSMEs from unfair competition from foreign platforms and to ensure product safety. An E-commerce Bill will be tabled in the next parliamentary sitting. It continues a broader pattern in which Malaysia regulates while Singapore funds.

For growth-stage companies eyeing an exit, the Securities Commission Malaysia and Hong Kong’s securities regulator have formalised cooperation to help Malaysian companies list in both markets, with matching grants to cover listing costs. Forest City’s Single Family Office scheme offers zero tax on qualifying income, and a Multi-Family Office model will let licensed fund managers serve several families.

Green tech tax incentives have been enhanced and extended to December 31, 2030. They include an Investment Tax Allowance of up to 100 per cent for green and sustainable technology projects, EV charging stations, and green assets bought for own use.

On cybersecurity, Malaysia says it is now recognised at Tier 1 for readiness to transition to post-quantum cryptography. The National Cyber Security Agency (NACSA) and the Malaysian Communications and Multimedia Commission (MCMC) will develop home-grown data protection technologies built to withstand attacks from quantum computers, while CSCDC Bhd’s capabilities are strengthened. “Home-grown” is the operative word. The country’s cyber startups have been battling a funding drought, and they will want to know whether that work flows to them or stays inside government agencies.

The real test is deployment

Anwar projected growth of 4.2 to 5.2 per cent in 2027, a touch softer than the 4.8 to 5.3 per cent expected this year. In a slower year, capital that sits in committees is capital that does not exist.

Also Read: Mid-year 2026 look at Malaysia’s digital and technology regulatory updates

Budgets are statements of intent. This one is clear about its priorities: build the middle, double down on chips, and give everyone a chatbot. Startups got a seat at the table. Whether they get fed depends on how fast Jelawang, KWAP and Cradle turn announcements into term sheets.

The post Malaysia’s Budget 2027 hands startups US$110.4M but the bigger cheques go elsewhere appeared first on e27.

Ask a Singapore small business owner what AI has done for them, and the answer is time. Ask whether their systems are ready to let AI run more of the business, and the answer gets a lot less certain.

Also Read: Beyond automation: How AI is reshaping the SME landscape in Southeast Asia

That tension runs through “The State of AI in Small Business 2026 – Singapore”, a new report from small business accounting platform Xero. Research firm Pureprofile conducted the survey between July 29 and August 5. It polled 1,018 business owners and senior decision-makers at Singapore companies with fewer than 200 employees and annual revenue of up to SGD100 million (about US$77 million).

The starting point is familiar to anyone who has run a small company. Some 76 per cent of respondents said their leadership and finance teams lose anywhere from three to more than 10 hours a week to manual administrative work. AI, the report argues, is buying that time back.

Where the reclaimed hours go is the more flattering part of the story. Nearly half (46 per cent) of SMEs said they channel them into growth strategies and market expansion, 38 per cent into product and service innovation, and 33 per cent into upskilling existing staff for higher-value tasks.

“The real dividend of AI isn’t simply doing the same work faster. It’s the capacity it gives back, and business owners are clear they’d reinvest that time in growth, innovation and their people,” said Koren Wines, Managing Director of Xero APAC.

It is a tidy narrative, and a convenient one for a company that sells AI-embedded accounting software. It is also worth noting what the survey measures: where SMEs say the hours go, not whether those hours turned into new customers or revenue.

The 82 per cent problem

On sentiment, Singapore’s SMEs are bullish. Some 82 per cent said AI returns have met or exceeded expectations, and 69 per cent plan to increase total AI spending over the next 12 months.

Dig into the operational numbers, though, and the confidence thins out. Only 18 per cent feel fully confident that their current software has enough context to support automated decision-making, such as data history, clean ledgers and connected systems. Nearly a third (31 per cent) named poor data quality as the main reason AI pilots never made it into regular production.

Also Read: Why Singapore SMEs should treat cybersecurity as a sales issue

Cost and talent remain sticking points too. Integration costs (38 per cent) and upskilling staff (36 per cent) are the top hurdles to expanding adoption.

Read together, the figures suggest SMEs are pleased with what AI does today, which is largely task-level admin relief. They are far less sure it can be trusted with anything more consequential, and high satisfaction may simply reflect modest expectations.

The harder question is what happens after the pilot phase, when AI is expected to make decisions rather than draft emails. As one e27 analysis put it, the region’s small businesses arguably have a record problem before they have an AI problem.

Guardrails, at least on paper

Xero frames Singapore’s SMEs as leaders in responsible AI use. Some 67 per cent have established formal internal usage frameworks, and 90 per cent say vendor data privacy is a priority.

The split within that 90 per cent is telling. Only 36 per cent will work exclusively with vendors that explicitly guarantee customer data will not be used to train external models. The remaining 54 per cent prefer vendors that offer a clear opt-out. That quietly shifts the burden onto the business owner to find, and remember to flip, the right setting.

A framework on file is also not the same as one that is enforced. Larger Singapore firms adopting agentic AI have been found to keep thin audit trails, and many fear data sovereignty failures while remaining underprepared for them. There is little reason to assume a 30-person trading company is more rigorous.

The grant question

For Xero at least, the report’s most commercially useful finding is that 52 per cent of SMEs now use AI built directly into core software they already rely on. That includes cloud accounting, inventory tracking and customer management tools. It is precisely the pitch embedded-software vendors make, so the figure is best read through that lens.

Also Read: The Philippines uses AI as a burnout shield; Singapore uses it for deep work

The other number that deserves attention is 63 per cent: the share of SMEs using government digital grants to fund technology deployment and training. Singapore has made AI adoption a national priority, most visibly in its AI-first Budget 2026, and the grants clearly lower the barrier to entry.

They also raise a fair question. How much of that planned increase in AI spending, reported by 69 per cent of SMEs, is driven by their own conviction, and how much by the availability of subsidies? The answer will matter if grant priorities shift.

Ahead of the region, still short of the finish line

By regional standards, Singapore’s small businesses are in a strong position. A similar Xero study found Malaysian SMEs grappling with a growing confidence gap in AI adoption. Globally, SMEs have been adopting AI four times slower than large enterprises.

But being ahead of the pack is not the same as being ready. For founders building software for Southeast Asia’s SMEs, the report contains a clear signal: the bottleneck is plumbing, not models. Clean books, connected systems and trustworthy data handling are what stand between an SME saving a few hours a week and an SME letting AI make real decisions. It is also, not coincidentally, exactly where Xero wants to sell.

“What’s encouraging is that AI is reshaping roles rather than replacing them,” Wines said. “For Singapore’s SMEs, it’s no longer about adoption. It’s about what happens next.”

Also Read: Why Singapore, Indonesia, and Vietnam are losing the AI race they think they are winning

She is right about that. The first phase of SME AI was measured in hours saved. The next will be measured in revenue earned, and on that score, Singapore’s small businesses have yet to show their workings.

The post Singapore SMEs buy back time with AI, but messy data keeps pilots stuck appeared first on e27.

Before a Vietnamese shopper adds anything to a cart these days, there is a good chance a chatbot has been consulted. As many as 62 per cent of Vietnamese buyers use AI, including generative AI, to research purchases before they buy, according to a new survey by Ho Chi Minh City-based market research firm Q&Me.

Only 3 per cent of respondents said they “never” use AI for pre-purchase research. Frequent use peaks among people in their 30s (68 per cent), while those in their 40s are more likely to say they use it “sometimes” (44 per cent).

Also Read: The invisible shopper rewriting Asia’s e-commerce playbook

On the surface, it is another data point for the thesis that AI chats are becoming the new starting point for online shopping. Read the rest of the survey, though, and a more nuanced picture emerges: AI has joined the shopping journey, but it has not taken it over.

The old guard still runs the funnel

When respondents were asked which research methods they actually use, generative AI came in fifth, cited by 58 per cent. Search engines such as Google led with 86 per cent, followed by e-commerce platforms such as Shopee and Lazada (77 per cent), Facebook (59 per cent) and TikTok (57 per cent).

In other words, the chatbot has become one more tab among many rather than a replacement for them. A typical Vietnamese buyer is likely to ask ChatGPT what to look for, check Google for reviews, compare prices on Shopee, and watch a TikTok creator unbox the product before paying. That fragmentation is a headache for brands, which now have to show up credibly in all five places.

It also matters for the platforms. TikTok, which is investing US$980 million in a logistics project in Ho Chi Minh City, is betting that discovery and checkout happen in the same app. A shopper, who does the thinking inside a chatbot and only arrives at the marketplace to pay, weakens that bet, and hands the most valuable part of the funnel — the moment of consideration — to someone else.

What people actually ask the bots

Shopping is not even the main reason Vietnamese users turn to generative AI. According to Q&Me, 78 per cent use it for information search, 62 per cent for personal advice, 58 per cent for product research and 57 per cent for work tasks.

The information-search figure is the one to watch. Chatbots are increasingly doing the job search engines used to do, a shift regulators elsewhere are starting to notice; Europe’s decision to classify ChatGPT as a search engine is one sign of where that leads. For shoppers, the risk is that a confident-sounding answer replaces a comparison. Product data feeding AI systems is often incomplete or simply wrong, and a chatbot rarely tells a buyer when it is guessing.

Also Read: How Vietnam is emerging as a leading AI builder ecosystem in Southeast Asia

The 62 per cent seeking personal advice deserves its own caution. Advice on what to buy shades quickly into advice on whether to buy, and few of these tools disclose how products end up in their recommendations.

ChatGPT leads, Google closes in

Among generative AI services, ChatGPT is the clear leader, used by 80 per cent of respondents, followed by Gemini (68 per cent) and Meta AI (25 per cent). Claude, Copilot, DeepSeek, Grok and Perplexity trailed behind.

Gemini’s strong second place is no accident. Google has been pushing hard on distribution and local-language capability across Southeast Asia, and it owns both the Android phones most Vietnamese carry and the search engine 86 per cent still rely on.

Meta AI’s 25 per cent, by contrast, looks modest given how deeply Facebook is woven into Vietnamese commerce, which suggests users have not yet learned to treat the assistant inside their social feed as a shopping adviser.

Notably absent from the list of popular services is any homegrown Vietnamese assistant, despite the country’s ambitions to become a regional AI builder and a growing body of work on models trained for Vietnamese, Thai and Indonesian. Vietnamese consumers, for now, are doing their AI-assisted shopping on American platforms.

Read the fine print

The survey’s headline numbers should be handled with care. Q&Me polled just 200 men and women aged 20 to 49 in Ho Chi Minh City and Hanoi. A sample that size carries a margin of error of roughly seven percentage points either way, and the age-group splits are drawn from smaller pools still, so the gap between people in their 30s and 40s may be narrower or wider than it looks.

The sample is also urban, working-age and, almost by definition, online, which is exactly the slice of Vietnam most likely to have embraced AI. It says little about older shoppers or buyers outside the two biggest cities, where much of the next wave of e-commerce growth is expected to come from.

There is a definitional wrinkle too. The survey says 62 per cent use AI for pre-purchase research, yet only 58 per cent name generative AI as one of their research methods, and the same 58 per cent use it for product research. The difference may come down to how respondents interpreted “AI”: some may be counting the AI features already built into search results and shopping apps rather than a standalone chatbot.

What it means for brands and founders

None of this undercuts the direction of travel. Even with a small sample, near-universal exposure among urban buyers is a strong signal, and the 30-something cohort, the one with the most purchasing power, is the heaviest user.

For brands selling into Vietnam, the practical lesson is that being findable now means being legible to machines as well as to people: accurate product data, consistent reviews and clear specifications that a chatbot can parse and repeat. For startups, the gap is in the middle of the funnel, building tools that help merchants show up correctly inside AI answers, and that help shoppers verify what those answers tell them.

Also Read: Why Singapore, Indonesia, and Vietnam are losing the AI race they think they are winning

For now, Vietnamese shoppers are using AI the way they use a well-read friend: worth asking, rarely the final word. Google, Shopee and TikTok still close the sale. The question for the next survey is how long that lasts.

The post Vietnam’s shoppers are asking AI first, but Google still gets the last word appeared first on e27.

Generative AI has made video creation remarkably fast. A scene that once required a camera crew, actors, locations and hours of post-production can now begin with a prompt and an idea.

But after working with AI-generated cinematic videos, I have learned that generating a video quickly and producing a video that is actually usable are two very different things.

An AI model can create an impressive scene on the first attempt. At first glance, everything may look great: cinematic lighting, realistic environments, convincing characters and smooth camera movement.

Then you look closer. A car is moving in the wrong direction. A character suddenly looks different. An important object disappears. The environment does not accurately represent the intended location. A person appears twice. The camera moves differently from what the story requires.

The video may look impressive, but it is not necessarily usable.

Good-looking is not the same as correct

One of the most interesting things about AI video is how easily visual quality can hide mistakes. A generated scene can look highly realistic while containing small problems that make the entire shot unusable.

This becomes particularly important when creating content based on real locations, products or brands. If a scene is supposed to represent Singapore, for example, recognisable landmarks, roads, architecture and the overall environment need to make sense. A beautiful futuristic city is not enough if it no longer feels like Singapore.

The same applies to branded content. A logo that changes shape, text that becomes unreadable or a product that changes appearance can quickly turn a visually impressive generation into something that cannot be published.

This means creators need to evaluate AI-generated footage beyond visual quality. Accuracy matters too.

Also Read: AI boom pushes chipmakers to expand across Asia Pacific, FedEx report finds

Consistency is one of the biggest challenges

Creating one impressive AI-generated shot is becoming increasingly easy. Creating 10 shots that feel like they belong to the same film is much harder.

A character may look correct in one scene and slightly different in the next. Clothing can change. Buildings can move. Objects can disappear. Lighting can suddenly shift.

Individually, each generation might look good. Together, however, they may not feel like one continuous story.

This has changed the way I approach AI video creation. Instead of thinking only about individual prompts, I increasingly think about continuity: What must remain the same? What can change? Where should the character be? What direction should objects move? What should the camera be doing? What details from the previous scene need to remain visible?

Consistency becomes part of the creative process.

Prompting is becoming more like directing

Early AI prompts could be relatively simple. “Create a futuristic city.” Today, producing more controlled cinematic content often requires much more thought.

A creator may need to define the subject, action, environment, camera movement, framing, lighting, visual style, timing and restrictions. Instead of simply asking AI to create something beautiful, you begin directing what should happen. The difference is important.

A prompt might specify that the camera slowly pushes forward while the main character remains visible, vehicles continue moving naturally in the correct direction, the environment stays consistent and the shot ends at a particular visual moment. At that point, prompting starts to feel less like giving a computer a sentence and more like communicating instructions to a virtual production team.

The ability to describe an idea clearly becomes increasingly valuable.

Also Read: AI chats are becoming the new starting point for online shopping

Human judgement still decides what is usable

AI can generate alternatives very quickly, but it does not always understand why one result works better than another. A human still needs to notice when something feels wrong.

Is the movement natural? Does the location look accurate? Is the character consistent? Does the camera movement support the story? Is the branding correct? Does the scene communicate the intended message?

These decisions require judgement rather than generation. This is why I do not see AI video simply as a replacement for traditional creative work. It changes where some of the work happens.

Instead of manually producing every visual element, creators increasingly spend time defining ideas, setting constraints, reviewing generations, correcting problems and deciding which results deserve to become part of the final production.

Businesses should look beyond generation speed

For businesses considering AI-generated video, speed is understandably attractive. But “How quickly can AI generate a video?” may not be the most useful question. A better set of questions might be:

• How many generations are required before the result is usable?
• Can characters, products and brand elements remain consistent?
• How much human review is required?
• Can the technology accurately communicate the intended message?
• Does the final content actually feel appropriate for the brand?

AI can dramatically reduce some production barriers, particularly for ideas that would otherwise require expensive locations, visual effects or complex production. But businesses should still treat AI-generated content as creative production rather than a one-click solution.

The role of the creator is changing

AI video technology will continue to improve. Many problems that are difficult today will probably become easier.

Characters will become more consistent. Physical movement will improve. Models will understand environments better. Creators will gain more precise control.

But better technology does not eliminate the need for creative decisions. Someone still needs to decide what the audience should see, how the story should unfold, what feels authentic and whether the final result communicates the right idea.

The tools are changing rapidly, but storytelling, direction and judgement still matter.

AI can generate the footage.

Someone still has to direct it.

—

Editor’s note: e27 aims to foster thought leadership by publishing views from the community. You can also share your perspective by submitting an article, video, podcast, or infographic.

The views expressed in this article are those of the author and do not necessarily reflect the official policy or position of e27.

Join us on WhatsApp, Instagram, Facebook, X, and LinkedIn to stay connected.

The post AI can generate a video in minutes. Making it usable is the hard part appeared first on e27.

Over the past year, the question I hear from operators shipping into three or four ASEAN markets has quietly changed. It used to be about the tariff. Now it is about the paperwork. Which certificate does this lane need? Who in our supply chain can prove where this component was made? Why does the same product clear in two days in one market and sit for a week in the next?

That question just got sharper. On 19 September in Manila, ASEAN’s 10 signatories adopted a declaration allowing the upgraded ASEAN Trade in Goods Agreement, the bloc’s core treaty on goods trade, to apply provisionally ahead of its formal entry into force on 1 June 2027. The economic ministers described it as bringing the upgrade into force early, within 2026. The upgrade itself is substantial: relaxed rules of origin, broader self-certification, stronger digital documentation and a new chapter for small businesses.

Read the mechanism, though, and the picture changes. Provisional application is voluntary and reciprocal. It applies only between member states that have each notified the ASEAN Secretary-General, from the dates each one specifies. The upgrade does not switch on across the region. It switches on lane by lane.

One rulebook, many lanes

For the next eight months at least, there is no single upgraded ASEAN. There is a matrix of upgraded lanes, legacy lanes, and lanes where one side has moved and the other has not. A business shipping across four markets will scale at the speed of its slowest notifying partner.

And the tariff was never the real barrier. ASEAN reported as far back as 2019 that 98.6 per cent of import duties across the bloc had already been eliminated. Yet the ministers’ statement this month put ATIGA’s utilisation rate, the share of eligible trade that actually claims the preferential tariff, at 62.7 per cent for 2025 across the nine members that reported, up from 51.7 per cent the year before. Real progress. It also means more than a third of qualifying trade still does not claim the benefit it is entitled to. The gap is not access. It is the capacity to prove eligibility, shipment after shipment.

That gap behaves differently at each scale. A micro exporter will not notice the upgrade at all and will keep filing the old forms, or ride on a larger buyer’s paperwork. An SME selling into two or three markets faces a dual regime that taxes management attention, and will either overpay duty to stay safe or claim wrongly and invite post-clearance audits. A regional operator that can see which lanes are live gets a temporary moat, pricing and positioning inventory ahead of competitors who still treat ASEAN as one market. At national level, notification becomes a competitiveness decision. A government that notifies late quietly taxes its own exporters relative to its neighbours.

Also Read: Your safest market is a crowded trade: Why I moved to Dubai six weeks after the ceasefire

The fast lane is already reserved

The detail that matters most is who benefits first. Since 30 June 2026, ASEAN’s mutual recognition arrangement for Authorised Economic Operators has been implemented by every member state except Timor-Leste. An AEO is a trader certified by its own customs authority as low risk, and under the arrangement that status earns expedited clearance and priority treatment across the region. Certification demands audited processes, security controls and a clean record. Few SMEs hold it. The upgrade stacks new facilitation on top of a club most of them have not joined.

The second trap sits upstream. Proving a product is ASEAN-made usually means showing regional value content, the share of its value created inside the region, of at least 40 per cent, or a qualifying change in tariff classification. That evidence lives in your suppliers’ invoices and bills of materials, not in your own files. If a tier-two supplier in another market cannot produce it, your claim fails, regardless of how the treaty reads.

The digital rail is not the problem. Ministers reported that 1,346,971 electronic certificates of origin moved through the ASEAN Single Window in 2025. The rail works. What travels on it is only as good as the record behind it.

Two paths through the patchwork

For businesses directly exposed, the move is architectural. Build a lane map: every origin and destination pair you ship, its notification status, and the origin rule for your top products. Treat it as a live operating document owned by someone senior, not a compliance memo filed and forgotten. Start AEO or approved exporter preparation now, because the window is short and certification is slow. Bring suppliers into the record. Origin evidence is shared data, and paying fairly for clean documentation costs less than a failed claim. Most of the businesses that use my scaling framework have stopped asking which market to enter next. They are asking which lanes their records can carry.

Also Read: Tourists first, trade later: Why Southeast Asia’s next boom is coming from the Gulf

For businesses less exposed, services-heavy or still largely single-market, this is a window to buy capability rather than build it. Distributors with good lanes but weak systems will struggle through the patchwork, and some will be open to partnership or acquisition on fair terms. Slower-notifying markets will see competitors hold back inventory. That is where patient positioning pays.

Korea already ran this experiment

South Korea learned early that signing trade agreements is not the same as using them. Its own officials, presenting to APEC, named complicated rules of origin and a shortage of trained staff as the main barriers for SMEs. The response was an institutional layer: the Korea International Trade Association’s FTA and Trade Support Center, a public-private one-stop service for smaller exporters, a free consulting hotline, visiting advisers, and from 2018 a trade ministry programme that linked exporters’ ERP systems directly with their suppliers’ data, supporting parent companies and their partners as a group.

That last detail is the lesson. Korea treated origin as a supply chain data problem, not a customs form problem. ASEAN now has the treaty and the digital rail. It does not yet have that layer coordinated across 10 markets during the patchwork. Operators will have to build their own.

The upgraded ATIGA will be fully in force by mid-2027. By then, the lanes worth having will already belong to operators who treated this year as an architecture problem. The treaty is regional. The advantage is architectural.

—

Editor’s note: e27 aims to foster thought leadership by publishing views from the community. You can also share your perspective by submitting an article, video, podcast, or infographic.

The views expressed in this article are those of the author and do not necessarily reflect the official policy or position of e27.

Join us on WhatsApp, Instagram, Facebook, X, and LinkedIn to stay connected.

The post ASEAN just upgraded its trade rulebook. Most operators will scale on the old one appeared first on e27.

Insurance has spent the past year experimenting with generative AI in the most obvious corners of the enterprise: summarising emails, extracting details from claims forms, drafting responses for call-centre agents, and speeding up document-heavy back-office work.

Those are useful gains. But a new whitepaper by Google Cloud and CoverGo argues that they may also be a distraction.

Also Read: OneByZero raises US$20M Series A to help enterprises move AI from pilots to production

The paper, authored by Maxim Afanasyev and Tomas Holub, frames the industry’s current AI moment as a potential “Kodak moment”: a point at which incumbents optimise an existing business model while the market quietly moves somewhere else. As management thinker Oren Harari put it, “The electric light did not come from continuous improvement of candles.” For insurers, the warning is that better workflow automation will not be enough if the fundamentals of claims, underwriting and distribution are being rewritten by autonomous software agents.

The paper points to studies from MIT and BCG suggesting that 95 per cent of organisations struggle to generate tangible returns from agentic AI. Its diagnosis is blunt: many institutions still treat AI as a series of isolated productivity projects, rather than a new operating layer for the business.

For Southeast Asia’s insurance sector, where mobile-first customers, digital wallets, embedded finance and platform ecosystems are already reshaping financial services, that distinction matters.

When every market participant has an AI agent

Agentic AI refers to systems that do more than answer prompts. These agents can plan tasks, call tools, retrieve data, make decisions within defined limits, and interact with other agents or systems. In insurance, that could mean an underwriting agent checking data sources, a claims agent assessing evidence, or a broker agent negotiating terms.

The bigger shift begins when insurers are no longer the only ones deploying these tools.

Customers may use personal AI agents to read policy wording and optimise claims submissions. Healthcare providers, repair workshops, brokers and even regulators may use agents to submit, verify, challenge or audit information in near real time. At that point, insurance stops being a slow exchange of forms between humans and becomes a machine-speed negotiation between multiple software actors.

The whitepaper gives a simple example: if a dental policy reimburses treatment linked to acute tooth pain, a customer’s AI agent may learn to frame a routine visit around that symptom. In underwriting, agents could identify weak spots in pricing models and generate applications that exploit them. In distribution, broker agents may flood the market with automated outreach, while customer agents filter or block them.

The result is not just more volume. It is a change in market behaviour.

Also Read: Health, wealth, and legacy planning converge as new wave of SEA insurtechs emerges

Economists have long discussed signalling: the idea that reliable information is costly to fake. AI weakens that assumption. If polished applications, claims narratives and supporting documents can be generated cheaply, insurers need to think beyond traditional risk scoring. They need mechanisms that encourage truthful behaviour, detect strategic manipulation, and decide when a human should intervene.

Why Southeast Asia cannot treat this as a distant problem

In mature Western markets, insurers often carry decades of legacy infrastructure and rigid internal rulebooks. Southeast Asia has its own constraints, including fragmented regulation, uneven data quality and protection gaps, but it also has a different adoption curve.

The region’s consumers are young, mobile-first and used to digital financial services moving quickly. In markets such as Singapore, Indonesia, Vietnam, Thailand and the Philippines, customers increasingly expect insurance to be embedded into payments, e-commerce, lending, travel, logistics and healthcare journeys. The next insurance interaction may not start on an insurer’s website. It may start inside a superapp, a hospital system, a broker platform or a small business finance tool.

That makes the agentic shift especially relevant. If every partner in the value chain deploys AI, insurers cannot rely on isolated internal pilots. A claims chatbot here and an underwriting assistant there will not be enough. The core systems must be able to coordinate agents, enforce authority limits, maintain audit trails and connect with external systems safely.

This is where the paper introduces the idea of an Agentic Core Operating System.

From AI tools to an agentic operating layer

The argument is not that insurers should replace humans with autonomous machines. It is that they need a technology foundation where human experts and AI agents work as coordinated teams.

In practical terms, this means moving away from monolithic, hardcoded systems towards modular architecture. One building block highlighted in the paper is the Model Context Protocol (MCP), often described as a “USB-C for AI”. The idea is to give agents a standardised way to access context from systems such as customer databases, policy administration platforms and claims records, without building a custom integration every time.

Other emerging layers include Agent-to-Agent protocols, which allow different agents to exchange information and negotiate tasks, and agentic payment protocols that could enable transactions under strict controls. Workflow engines based on directed acyclic graphs can help manage the recurring loop of plan, act, observe and reflect that agentic systems use.

Also Read: Why Southeast Asian enterprises need AI governance before scaling generative AI

For many Southeast Asian insurers, the appeal is clear. Legacy replacement projects are expensive and risky. A protocol-driven architecture could allow incumbents to build around old systems first, rather than rip everything out at once. But the governance challenge is just as important as the technology.

The new failure modes

A traditional AI tool can make a bad recommendation. A network of agents can create system-level failures.

The whitepaper identifies several risks. One is an agent loop, where two or more agents keep triggering each other without resolution. Another is agent disagreement, where different agents reach conflicting conclusions on the same case. A third is authority violation, where an agent acts beyond its approved mandate.

These risks matter in insurance because decisions affect money, compliance and trust. A wrongly denied claim can damage a customer relationship. A mispriced portfolio can create financial losses. A poorly governed automated process can draw regulatory scrutiny.

The proposed answer is a schema-driven agent model, where business rules, authority limits and escalation triggers are defined in configurable schemas rather than buried in code. Human-in-the-loop controls remain central: confidence scoring, mandatory escalation for sensitive cases, audit logs, and clear accountability for decisions.

For regulators in Southeast Asia, this could become a key question. The issue will not simply be whether an insurer uses AI, but whether it can explain how autonomous decisions are made, challenged and reversed.

The road beyond pilots

The paper’s modernisation roadmap is deliberately incremental. It begins with intelligent document processing to extract logic from legacy documentation, then separates solution design from execution so business and technology teams can validate workflows before coding. New agentic platforms can be built in parallel with legacy systems, master data can be loaded first, and policies can gradually move to the new core before old systems are retired.

That may sound unglamorous, but it addresses a real problem in insurance technology: transformation often fails when companies try to jump straight from legacy complexity to a grand new platform.

Also Read: What I’m learning about the second wave of insurance digital transformation in Indonesia

The deeper message for founders, investors and operators is that AI strategy in insurance can no longer be measured only by productivity gains. The question is whether the business is prepared for a market where customers, intermediaries and service providers also have intelligent software acting on their behalf.

Peter Drucker’s warning remains apt: “There is nothing so useless as doing effectively that which should not be done at all.”

For Southeast Asia’s insurers, the risk is not that they fail to automate enough tasks. It is that they automate yesterday’s business model while the market shifts to a new one. The next phase of insurtech will belong to those that build not just AI use cases, but the operating system for an agentic economy.

The post SEA’s insurers face a new question: what happens when customers have agents? appeared first on e27.

Healthcare digitisation usually becomes visible through the applications people interact with: patient portals, telehealth platforms, digital prescriptions and, increasingly, digital health wallets. But the interface is rarely the difficult part.

In March 2026, the World Health Organisation (WHO), the Alliance for Health Policy and Systems Research and Temasek Foundation launched a three-year initiative to help ASEAN Member States move from paper-based health records towards secure, interoperable digital health wallets. The programme plans to use global standards including HL7 FHIR and the International Patient Summary so essential health information can move more consistently between systems. That initiative represents something bigger than another healthcare app.

It raises a question that Southeast Asia’s healthtech ecosystem will increasingly have to answer: Can health information remain useful when it moves beyond the system, provider or country where it was originally created?

Digital health is becoming infrastructure

Digital health is no longer confined to individual hospital transformation projects. Globally, WHO reported in 2026 that 129 Member States have national digital health strategies. Its Global Digital Health Certification Network already serves 1.8 billion people across 80 countries.

The broader South-East Asia region is moving in the same direction. WHO says at least 80 per cent of countries in its South-East Asia Region have drafted or completed national digital health strategies, with interoperability among the areas these strategies are intended to strengthen.

As this infrastructure expands, simply digitising records becomes less of a differentiator. The harder problem becomes making those records usable outside the system that created them.

Imagine a traveller receives treatment in one country and later needs care elsewhere. A digital wallet may be capable of presenting a medication, allergy, diagnosis or laboratory result to the second provider. Technically delivering that information is one problem. Making sure the second system understands it correctly is another.

Moving data is not the same as understanding it

Healthcare interoperability is often discussed as an API problem. Connect System A to System B. Exchange information using FHIR. Validate the payload. The integration works.

Except healthcare data carries meaning as well as structure. One hospital may represent a laboratory observation using an international terminology. Another may use a local code created years ago. Medication names can differ. Units can vary. Clinical concepts can be recorded at different levels of specificity.

The receiving platform might therefore successfully receive the data while still being unable to use it reliably. That distinction becomes especially important when health information starts travelling between organisations and eventually across borders.

Also Read: Why Southeast Asia’s next healthtech winners will be built around healthcare workflows, not just AI

FHIR provides a standard way to structure and exchange healthcare information, but consistent exchange does not automatically guarantee consistent interpretation. That is why capabilities such as healthcare terminology services become relevant: clinical vocabularies, mappings and value sets help different systems interpret the same underlying concepts consistently. For regional digital health infrastructure, semantic consistency may prove just as important as connectivity.

Standards are a foundation, not a finished solution

The ASEAN digital health wallet initiative’s use of HL7 FHIR and the International Patient Summary is significant because common standards reduce the number of proprietary formats each participant needs to understand. But adopting a standard does not remove every interoperability challenge.

Healthcare systems still have to resolve questions around patient identity, consent, terminology, data quality, security, access permissions and the clinical context surrounding a record.

There is also a regional dimension. Southeast Asia is not one healthcare market with a common technology stack. Countries differ in digital maturity, healthcare delivery models, public and private sector participation, regulation and existing infrastructure. Even within one country, a modern cloud platform may need to exchange information with hospital systems that were designed years earlier for very different purposes.

This means regional interoperability cannot simply mean making every participant use identical software. A more realistic goal is enabling different systems to communicate through shared standards while preserving the meaning, trust and context of the information being exchanged.

This changes how healthtech products should be designed

For healthtech founders, interoperability is easy to postpone. An early product may integrate with one hospital, one laboratory or one device ecosystem. At that stage, a tightly coupled integration can appear sufficient.

The architecture starts to matter differently once the product expands. A second healthcare provider may use another EHR. A new market may require different consent rules. Clinical terminology may change. A partner may expose FHIR APIs while another still depends on older interfaces.

If these differences are handled individually inside the core product, every expansion adds another layer of dependency. That is why interoperability increasingly needs to be treated as product architecture rather than a later integration task.

Also Read: Vietnam’s healthtech boom has a talent problem nobody is talking about

A scalable healthtech platform should be able to separate its core product logic from the systems around it. Identity, consent, terminology mapping, API management, auditability and data transformation need clear boundaries.

The objective is not to predict every future integration. It is to avoid designing a product where every new integration requires rebuilding the foundation.

Portability also creates a trust problem

Technical interoperability alone will not determine whether portable health records work. Healthcare data is unusually sensitive. Giving individuals greater control over their information also creates questions about authentication, consent, revocation, data provenance and who should be permitted to access what.

A clinically useful record needs to be more than portable. A receiving healthcare organisation needs confidence about where the information came from, whether it has changed and whether the person presenting it is authorised to do so. That makes trust architecture inseparable from interoperability architecture.

This is particularly important for systems intended to operate across multiple organisations or jurisdictions. The further health data travels from its original environment, the more important identity, provenance and governance become.

The opportunity is larger than the wallet

Digital health wallets may eventually become one visible layer of Southeast Asia’s healthcare infrastructure. Their deeper significance is what must exist underneath them.

For a patient’s essential health information to travel with them, hospitals, government platforms, healthtech products and other healthcare systems need ways to exchange data without losing its meaning or trustworthiness. That will not happen because every organisation adopts the same application. It will happen through standards, governance and architectures that allow different applications to participate in the same healthcare ecosystem.

For Southeast Asian healthtech companies, this changes the strategic question. The next generation of regional healthcare products will not only need to ask, “Can our software work in another market?” They will increasingly need to ask: “Can our product become part of a healthcare system where information needs to move beyond us?”

That may ultimately be the more important interoperability test.

—

Editor’s note: e27 aims to foster thought leadership by publishing views from the community. You can also share your perspective by submitting an article, video, podcast, or infographic.

The views expressed in this article are those of the author and do not necessarily reflect the official policy or position of e27.

Join us on WhatsApp, Instagram, Facebook, X, and LinkedIn to stay connected.

The post Southeast Asia is digitising health records. Making them portable is the harder problem appeared first on e27.

Endeavor Catalyst has closed an oversubscribed US$320 million fifth fund, lifting its assets under management to more than US$850 million and giving the global investment platform fresh capital at a time when startup markets outside Silicon Valley are beginning to stir again.

The fund, announced from Singapore, comes after a difficult few years for venture-backed companies across Southeast Asia and the wider region. Late-stage funding became harder to secure, public listings slowed, and investors became more selective about growth claims that were once rewarded during the low-interest-rate era.

Also Read: Grab’s US$1.49B Atome deal signals a deeper race for SEA’s credit economy

For funds with exposure to emerging markets, the question has shifted from whether talent exists in these ecosystems to whether companies can find the capital, governance and international networks needed to scale through tougher cycles.

Endeavor Catalyst is betting that they can. Across its five funds, it has backed 437 companies in 44 markets, including 83 companies valued at more than US$1 billion and 39 exits. Its portfolio includes names familiar to Southeast Asian readers, such as Bukalapak, Advance Intelligence Group and Carsome, alongside global companies including Replit, Moove and Preply.

In Asia, the firm has invested US$46.8 million across 44 companies in six markets. The region accounts for only a slice of its global portfolio today, but Endeavor Catalyst says deal activity has picked up after several quiet years, making Asia a stated pipeline priority for Fund V.

That matters because the reset in Asian tech has not killed founder ambition. It has changed what investors are looking for. The era of subsidised growth has given way to a sharper focus on sustainable unit economics, regional execution and paths to liquidity. For founders, especially those building from Southeast Asia, capital alone is rarely enough. Access to follow-on investors, senior operators and customers across borders can be just as important.

A fund built around the Endeavor network

Endeavor Catalyst is the investment arm of Endeavor, the entrepreneur network that selects and supports founders building high-growth companies outside the most established startup hubs. The fund invests only in companies led by Endeavor Entrepreneurs and usually co-invests alongside other venture capital and growth equity firms.

This gives the platform a different shape from a traditional VC fund. Its sourcing depends heavily on Endeavor’s founder network, while its limited partner (LP) base also reflects that community. Endeavor Entrepreneurs make up about 30 per cent of Fund V’s LPs by headcount and account for 20 per cent of the capital raised.

The pitch is what Endeavor calls the “Multiplier Effect”: founders who build large companies return as mentors, investors and connectors for the next generation. In Southeast Asia, that idea is increasingly visible. Entrepreneurs behind companies such as Carro, Carousell and ShopBack have become part of a broader support system for newer founders, whether through angel investing, operator networks or informal advice.

“Asia has never been short of founders with global ambition,” said Nadir Zafar, Managing Director of Endeavor’s Global Hub in Singapore. “Endeavor Entrepreneurs like Carro, Carousell and ShopBack have shown what the region can build, and many of those founders are now mentoring and investing in the next generation.”

The Singapore hub, launched in 2026, is intended to anchor Endeavor’s regional activity. It also reflects Singapore’s continued role as a base for Southeast Asian scaleups, regional headquarters and cross-border capital — even as founders increasingly build teams, revenue and product depth across Indonesia, Vietnam, the Philippines, Malaysia and Thailand.

Asia after the funding winter

For Endeavor Catalyst, the timing is notable. The region’s venture market has been through a funding winter, particularly at later stages. Many startups cut costs, delayed expansion plans or explored consolidation. Some investors who entered Southeast Asia during the pandemic boom pulled back or slowed deployment.

Also Read: When 43 per cent of the world’s funding goes to two firms, where does that leave SEA?

Yet the slowdown has also created a clearer divide between companies built on durable demand and those reliant on cheap capital. Fintech infrastructure, AI applications, logistics, software for small businesses, climate and mobility remain active themes, but investors are asking harder questions about margins, regulation and exit options.

“2025 was a ‘wait and see’ year for investment in Asia, but we’re expecting companies to scale up, merge, list, and exit at a faster pace in the coming years,” said Caela Tanjangco, Head of Asia for Endeavor Catalyst.

The firm has had no liquidity events in Asia since 2025, but globally it has recorded 15 liquidity events since the start of that year. Fund V has already made 32 investments across 30 companies and deployed US$55.7 million. Early investments include Replit, Moove, Preply, Cashea, Respond.io, HappyRobot, Flash and Lunar Outpost.

The current pipeline is strongest in Europe and Latin America, according to Endeavor Catalyst, but Asia is seeing a rise in opportunities. The fund is also tracking deeptech, foundational AI and cloud infrastructure, as well as second-time founders building AI companies at the application layer, areas drawing investor attention globally.

For Southeast Asia, the AI wave presents both opportunity and risk. The region has large digital populations, fast-growing enterprise demand and cost-efficient engineering talent, but it still depends heavily on imported compute infrastructure and global foundation models. The companies most likely to break out may be those solving practical industry problems in finance, commerce, logistics, healthcare or manufacturing, rather than those trying to compete head-on with US and Chinese AI labs.

The competitive landscape

Endeavor Catalyst is not alone in chasing founders outside traditional venture centres. In Southeast Asia, it competes for access and mindshare with regional firms such as East Ventures, Wavemaker Partners, Openspace Ventures, Jungle Ventures, Insignia Ventures Partners and Monk’s Hill Ventures. Global and multi-market investors including 500 Global, Antler, Peak XV Partners, Accel and Lightspeed also maintain exposure to the region in different ways.

What differentiates Endeavor Catalyst is that it does not operate as a broad sourcing fund; it invests through the Endeavor network and into founders already selected by that community. That can narrow the investable universe, but it may also give the fund warmer access to later-stage founders and a built-in support system across markets.

Redrawing the innovation map

Endeavor’s larger argument is that venture capital remains too concentrated in a small number of markets, even as high-quality companies emerge elsewhere. Latin America remains Endeavor Catalyst’s largest region, while Europe accelerated sharply with 12 new investments in the first half of 2026, after 14 in all of 2025. The broader portfolio spans the Middle East, Africa, Asia, the US and Canada.

For Asia, the significance of Fund V is less about a single cheque and more about what it signals. Investors are not rushing back indiscriminately. But funds with patient networks, cross-border reach and a sharper view of emerging-market execution are preparing for the next cycle.

Also Read: Southeast Asia’s biggest tech IPO of the year is a landlord

If the last few years forced Southeast Asian startups to prove they could survive without easy money, the next few may test whether the strongest among them can scale, consolidate and eventually return capital. Endeavor Catalyst’s new fund is a bet that more of those companies will come from places still underrepresented in global venture portfolios — and that Asia will be one of them.

The post Endeavor Catalyst raises US$320M to back the next wave of ‘Elsewhere’ startups appeared first on e27.

Southeast Asia’s retail sector is racing to embed AI into every layer of the customer journey and nowhere is that race more visible than in Singapore. The government announced a 400 per cent tax deduction on AI expenditure to help small and medium-sized retailers adopt the technology, a clear signal that AI is no longer a nice-to-have but a core national retail strategy.

The appetite from shoppers matches the ambition of policymakers. A Lazada study found that 92 per cent of SEA shoppers already use AI-powered personalised recommendations, and 90 per cent say they trust AI-generated product summaries. 

For an industry built on razor-thin margins and fierce competition, AI has become the linchpin for anticipating demand, auditing inventory in real time, and crafting personalised offers at scale.

Customer experience at stake

But the same systems retailers are racing to deploy can just as easily undermine the customer experience they’re meant to elevate. A broad-strokes rollout during peak sales periods like Singles’ Day or a festive season is a high-stakes gamble. 

If AI misreads data and pushes false discounts mid-campaign, the financial and reputational fallout can be immediate and severe. Irrelevant recommendations quietly erode brand loyalty, one dismissed suggestion at a time.

As AI spreads across more retail functions, the margin for error shrinks toward zero. Retail organisations intent on unlocking AI’s full potential should accelerate adoption while building the safeguards and visibility needed to deploy it responsibly.

Also Read: Singapore’s AI dividend will depend on what happens after the pilot phase

Observability as a strategic guardrail

This is where observability enters as a key strategic tool  holding the whole system together. A New Relic report into the sector reveals that 55 per cent of the industry already embed active monitoring into its workflows, indicating that granular visibility, not blind trust, is the path to digital maturity.

AI-strengthened observability gives retail and engineering teams consolidated visibility into application performance, data quality, and system behaviour. This provides insight into not just what AI is doing, but why. By continuously monitoring the full lifecycle of AI requests and responses, teams can catch anomalies early, validate assumptions, and act before problems ever reach a customer’s checkout screen.

Observability is also a frontline defence against bias, hallucination, and quiet performance decay. Continuous monitoring flags the moment a model starts leaning on incomplete, outdated, or skewed data, long before customers notice degraded recommendations or vanished stock.

An example would be a scenario during a peak season sale. Instead of AI confidently surfacing products that are already sold out, observability finds the mismatch between real-time stock signals and inferred availability the moment it appears. When pricing or personalisation logic starts drifting from actual customer behaviour, anomaly detection flags the deviation before it snowballs into a bad Raya or Black Friday customer experience.

Also Read: The hidden economics of autonomous AI agents

Observability also lets teams pre-validate AI applications in staging environments, reducing the risk of disruption once systems go live. Paired with automated, gradual rollouts, updates can be introduced safely even during the highest-traffic festive periods. If something misbehaves, teams can pause or roll back changes fast. This is an especially critical capability, given that high-business-impact outages in retail can cost as much as US$1 million per hour.

Moving ahead with clarity

AI may be the engine driving retail growth, but observability with other practices is a key tool that keeps that engine on the road. Retailers that have already invested in these safeguards are seeing performance improvements deliver up to 2x ROI. In a market where digital customer experience is the real battleground, from SEA’s tech-forward shoppers to the booming quick-commerce apps, organisations need to ensure they are pairing bold AI innovation with the granular oversight needed to make every touchpoint reliable.

With half of retailers already integrating granular browser and mobile monitoring to close visibility gaps, they are certainly moving in the right direction. The next step is correlating front-end user behaviour directly with AI-driven back-end performance. A mature observability practice will enable engineering teams to track model response times, monitor prompt quality, and manage operational costs, all without losing sight of the mobile and browser experience that ultimately drives revenue.

—

Editor’s note: e27 aims to foster thought leadership by publishing views from the community. You can also share your perspective by submitting an article, video, podcast, or infographic.

The views expressed in this article are those of the author and do not necessarily reflect the official policy or position of e27.

Join us on WhatsApp, Instagram, Facebook, X, and LinkedIn to stay connected.

The post Why SEA’s retailers need to protect AI-powered customer experience with observability  appeared first on e27.