Day: October 9, 2026

The client accepted the final video. We still decided not to work with the production team again.

Before the project, the team had looked like a strong choice. They sent us more than a dozen polished samples, with good image quality and a convincing visual standard.

We hired them to produce a five-minute promotional film for a tea brand. They had about one week to deliver the first cut. The finished video contained roughly 150 shots. More than 10 had clear problems: weak lip-sync, limited visual variety and choices suggesting that parts of the brief had not been fully understood.

We identified the most obvious issues, asked for replacements and spent about two additional days completing the revisions. The client gave some feedback and ultimately accepted the video.

Nothing about the final delivery was disastrous. The problem was what it had taken to get there.

That experience changed how I read a portfolio, especially in AI-assisted production. A portfolio can show what a team has made. It usually cannot show how much judgement someone else had to supply before the work became usable.

A portfolio shows the output, not the process

Commercial portfolios are made from selected outputs. Commercial projects are made from repeated decisions. A team can refine a small number of images or short sequences until they look exceptional. Those samples may genuinely demonstrate strong creative ability.

A five-minute commercial project asks for something else. Can quality survive across 150 connected shots? Can characters, products and environments remain consistent? Does the team understand why the client requested a change rather than simply follow the literal wording? Does someone notice a weak shot before the client does?

AI is widening access to production capability. In the creator community I work with, many people can now produce an attractive frame or polished short sequence with current image and video tools. Far fewer are people I would trust with an entire commercial project.

Once the work expands from a sample into dozens or hundreds of connected decisions, technical ability is only part of the job. Someone must also interpret the brief, maintain continuity, decide what is acceptable and catch failures before they move downstream. Those capabilities are much harder to see in a portfolio.

Also Read: AI won’t just replace jobs. It will redesign how companies work

How much judgement did the client supply?

Our tea project made that gap visible. The final video was acceptable, but reaching that point required our own quality control. We reviewed the first cut, identified the visible problems, decided which shots needed replacing and directed the corrections.

The schedule deserves some weight. One week was tight for a five-minute film containing around 150 shots, and we did not confirm every production stage as fully as we normally would. With more time, we would usually confirm the script, key visual assets and storyboard before reviewing a first cut. Better checkpoints would probably have prevented some of the problems.

I would not attribute every misunderstanding to the supplier. But the compressed schedule also revealed which problems the team caught itself and which ones reached us first.

The client should still make judgements about story, tone and message. Those belong with the brand. It should not routinely be the first party to discover inaccurate lip-sync, obvious continuity problems or a shot that falls well below the quality of the rest of the film.

A first cut does not need to be perfect. What I want to know is how much production judgement has to come back to the client before the work becomes usable.

Two suppliers can therefore deliver equally acceptable final videos while leaving very different amounts of work with the client. A portfolio rarely reveals that difference.

Creative supply is larger than dependable capacity

I see the same gap in the wider creator community I work with. There are several hundred people in the group. I would estimate that 10 to 20 per cent can contribute meaningfully to commercial projects. At the moment, only about 10 are people I would be comfortable asking to take a project from storyboard through image generation, video production and editing with limited intervention.

Also Read: How to turn your founder’s opinions into media-ready narratives

A strong sample may be enough for me to try someone on a first project. It usually takes repeated cooperation to learn what happens when the deadline is shorter, a revision is inconvenient or an early interpretation of the brief turns out to be wrong.

The same applies to a production company. Access to hundreds of creators may make it easier to assemble a team. It does not tell me how many projects the company can supervise consistently or how much senior judgement sits between individual creators and the client.

That is the capacity a buyer is actually purchasing.

Test the process, not another sample

If I were evaluating a new production team again, I would still look at its portfolio. I just would not ask for another polished sample if I wanted more confidence.

A small paid project would tell me more, particularly if it reproduced some of the difficult parts of real commercial work. I would rather see several connected scenes than one perfect shot. I would include an intermediate review and a revision. I would watch what questions the team asks before production, what problems it catches without prompting and how it responds when its first interpretation of the brief is incomplete.

The finished output would still matter. So would the amount of management required to get there.

Our tea video was delivered and accepted. That was enough for the project.

For the next one, I would want to know something the final video could not tell me: how much of the production judgement could stay with the supplier instead of coming back to us.

—

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 Creative supply is large, dependable production capacity is not appeared first on e27.

The crypto market woke up to a familiar tension this morning. Bitcoin trades at US$83,079.42, down 1.02 per cent over the past 24 hours, while trading volume has surged by roughly 40 per cent. That combination of falling price and rising volume rarely signals calm. It tells a story of forced exits, not quiet repositioning. The world’s largest digital asset slipped below US$84,000 overnight and tested the US$83,000 support level, a zone that has repeatedly acted as a floor since the September recovery.

The immediate trigger for this pullback stems from macro forces unrelated to blockchain technology. The Federal Reserve released minutes from its September 15 to 16 meeting on October 7, and the document carried a distinctly hawkish tone. The FOMC voted unanimously to raise the federal funds rate by 25 basis points to a target range of 3.75 per cent to 4.00 per cent, marking the first rate hike since July 2023.

More importantly, most participants assessed that another increase in the target range would likely be appropriate by year-end. Officials pointed to persistent geopolitical tensions that have pushed up crude oil and refined fuel prices, alongside a surge in artificial intelligence-related investment that has added to inflation pressures. The CME FedWatch tool now puts the probability of a December hike at 70.5 per cent, up sharply from earlier expectations.

Oil markets have amplified this pressure. Brent crude futures climbed above US$101 per barrel this week, gaining 93 cents or 0.92 per cent to reach US$101.51 by early Wednesday trading. The International Monetary Fund has warned that high energy prices could persist into 2027 even if current hostilities in the Gulf region end quickly.

Higher crude feeds directly into inflation expectations, which in turn keeps the Fed under pressure to maintain tight policy and keeps bond yields elevated. The 10-year Treasury yield has climbed above 5.3 per cent while the 30-year sits above 5.6 per cent. A stronger US dollar has accompanied this move, pushing the dollar index above 102 against major peers and tightening financial conditions across risk assets.

The liquidation data tells the human story behind these numbers. Between US$550 million and US$690 million in crypto positions were liquidated over the past 24 hours, with the overwhelming majority coming from leveraged long positions. Bitcoin had already fallen from approximately US$85,341 to US$83,790 in roughly 20 minutes during the first major liquidation wave. This mechanism matters because forced liquidations create additional selling regardless of whether holders actually want to exit.

One particularly striking detail emerged from on-chain tracking: four separate wallets opened short positions against 148.49 BTC with 40x leverage on the decentralised exchange Hyperliquid just before the rapid price drop. The cascade that followed wiped out hundreds of millions in bullish bets, with roughly US$487 million coming from long positions alone.

Also Read: Light liquidations and flat funding: Is Bitcoin about to explode?

Beneath this violent surface, something more constructive is taking place. Exchange outflows hit a 7-month high on October 5, with approximately 24,073 BTC moving from centralised trading venues into private storage or custody solutions. This marks the largest single-day net withdrawal since March 1, pushing exchange-held Bitcoin down to roughly 6.50 per cent of the total supply.

A shrinking liquid supply can support prices over time because fewer coins sit on order books ready for immediate sale. The destination of these withdrawn coins remains unconfirmed. They may reflect long-term cold storage by institutional holders, repositioning by large wallets ahead of an anticipated move, or routine withdrawals by retail holders. The data establish that a meaningful portion of the available float left trading platforms precisely when macro headwinds intensified.

The technical picture now hinges on a few critical levels. Support sits in the US$81,300 to US$83,000 range, with the lower bound representing a level that analysts at Bitfinex have identified as the point where sustained trading below would change the market structure and bring the US$77,000 region back into play. Resistance is positioned at US$84,000 to US$86,500, and Bitcoin has now failed three times to break above the US$87,000 area in recent weeks.

The next major catalyst arrives on October 14 with the release of the US CPI inflation report for September. That data point will serve as the first genuine test of whether disinflation dynamics are reasserting or stalling, and markets will build positioning accordingly in the days ahead.

Also Read: Can Bitcoin defend US$85,000 support, or will weakening bids send it toward US$83,000?

Three additional developments deserve attention as they shape the broader narrative.

First, on October 7, US government wallets transferred approximately US$470 million in Bitcoin, wrapped Bitcoin, and USDT to addresses that Arkham Intelligence identified as likely Coinbase Prime deposit addresses. These assets link back to the 2016 Bitfinex hack and to Alameda Research. The move has reignited speculation about potential government sales, though analysts note it could equally signify a change in custody or routine administrative work. A transfer of this size to an institutional venue represents a meaningful inflow to an exchange’s custody pool regardless of whether a sale follows immediately.

Second, Bitcoin closed September 2026 at US$83,556, marking a 6.4 per cent gain and its best September performance on record. This defied the historical pattern in which September averages a 2.87 per cent decline, and every positive August since 2013 has been followed by a red September.

The resilience came despite the 10-year Treasury yield climbing to its highest level since 2007 and oil trading above US$100 per barrel. Robust institutional demand drove this outperformance, with spot Bitcoin ETFs recording a 7-day inflow streak adding approximately US$6.6 billion in late September. The buyer’s identity has changed, and that shift has proven more powerful than seasonal tendencies.

Third, on-chain analytics firm Glassnode has flagged a structural shift in Bitcoin’s intraday demand pattern. The US trading-hour bid, which measures net price contribution during New York session hours from 9:30 a.m. to 4:00 p.m. Eastern Time, has flipped direction since the September breakout. This means US session buyers have become the dominant marginal force, a reversal from previous patterns where offshore sessions carried more weight.

The flip is not a single-session anomaly but a sustained change that Glassnode characterises as a structural feature of the post-breakout environment. This points to increased net buying pressure from US-based participants, potentially linked to the timing of institutional ETF flows, and signals evolving participation that could support price stability during domestic hours.

Also Read: Bitcoin jumped 1.81% to US$86,350.24. Is this a real breakout or a short squeeze?

The market now finds itself caught between two powerful forces. On one side stands institutional demand through ETF channels that has proven capable of overriding traditional macro headwinds and seasonal weakness. On the other side stands a hawkish Federal Reserve, surging oil prices, elevated Treasury yields, and a stronger $ that together create a challenging environment for non-yielding speculative assets.

The US$83,000 support level has held for now, but the volume accompanying this decline suggests that conviction remains fragile. All eyes turn to October 14 and the CPI report, which will determine whether the disinflation narrative that supported Bitcoin’s September rally can withstand the latest inflation data.

Active forecasting pools directly on the Polymarket Crypto Hub show how traders are positioning. In the main tracking pool asking what price Bitcoin will hit in October, sentiment has shifted sharply following the drop to US$83,000, with participants pricing in nearly 100 per cent certainty that Bitcoin will break below the US$85,000 threshold during October’s broader multi-week timeframe.

Short-term directional contracts, such as the Bitcoin Up or Down Daily Contracts, reflect highly contested intraday sentiment, with an uncertain 51 per cent chance of an upward close. Traders are positioning capital ahead of mid-month data releases in the Bitcoin price on the October 14 pool, which is a major focal point for bets on whether post-CPI inflation data will trigger a market recovery or deeper corrections.

My stance remains the same. Watch and see. Find a good entry point with a higher win rate. There is no need to rush to lose money.

—

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 Bitcoin’s US$83,000 test: Can institutional demand hold the line? appeared first on e27.

For many companies, the harder part of artificial intelligence is no longer running a pilot. It is turning that pilot into something reliable enough to sit inside daily operations.

That gap is what Singtel’s RE:AI and the Singapore Institute of Technology (SIT) are trying to address through a new partnership with the SIT x NVIDIA AI Technology Centre.

The two parties have signed a memorandum of understanding to help enterprises and government agencies co-develop AI applications that can move from applied research into production, while also training more AI practitioners in Singapore.

Also Read: Why Southeast Asia cannot build sovereign AI on borrowed choices

Under the partnership, enterprises participating in Singtel Digital InfraCo’s Centre of Excellence for Applied AI with NVIDIA will work with SIT students, researchers from the SIT x NVIDIA AI Technology Centre (known as SNAIC) and the NVIDIA AI Technology Centre. The aim is to build purpose-specific AI tools for industry use cases, rather than rely only on off-the-shelf products that may not fit an organisation’s workflows, data requirements or regulatory obligations.

At the centre of the collaboration is RE:AI, Singtel Digital InfraCo’s sovereign AI cloud business. Sovereign cloud refers to infrastructure designed to meet local requirements around data residency, security, compliance and control. This is becoming more important as companies and public agencies experiment with AI models that may process sensitive commercial, operational or citizen data.

RE:AI will host an advanced AI testbed at SIT’s campus in the Punggol Digital District. The testbed will be powered by NVIDIA GPU infrastructure and AI solutions, giving enterprises, researchers and students a shared environment to build, test and validate applications before deploying them more widely.

Why the pilot-to-production gap matters

The announcement lands at a time when Southeast Asian enterprises are under pressure to show that AI can deliver real productivity gains, not just proof-of-concept demonstrations. Banks, telcos, transport operators, hospitals, retailers and government agencies are all exploring generative AI and machine learning, but many projects stall once they move beyond a small internal trial.

The reasons are familiar to enterprise technology teams. Models need to be tuned to sector-specific data. AI systems must integrate with legacy software. Outputs need to be explainable enough for regulated industries. Data cannot always leave the country or be sent to public cloud environments without additional controls. Most importantly, the organisation needs people who understand both the technology and the business problem.

Singapore has tried to position itself as a regional testbed for this next phase of AI adoption. Its National AI Strategy 2.0, launched in 2023, placed stronger emphasis on industry deployment, talent development and trusted infrastructure. The government’s Research, Innovation and Enterprise 2030 plan also identifies AI as a priority area for long-term economic competitiveness.

Also Read: Singapore firms embrace agentic AI but audit trails remain thin

This partnership fits into that wider policy direction. SIT brings an applied learning model and access to students, faculty, postgraduate talent and researchers. SNAIC, officially opened in October 2025 as a joint initiative between SIT and NVIDIA AI Technology Centre, focuses on applied research and industry collaboration. Singtel brings cloud infrastructure, enterprise relationships and its growing digital infrastructure business.

Manoj Prasanna Kumar, Chief Technology and Information Officer at Singtel Digital InfraCo, said the partnership is designed to bring together industry, academia and digital infrastructure to help enterprises move “from research to production deployment”.

A campus testbed for industry problems

The location of the AI testbed is significant. SIT’s centralised campus sits inside the Punggol Digital District, Singapore’s attempt to create a tighter link between industry, research and talent development. By placing the testbed on campus, the partners are trying to make enterprise AI work less abstract for students and researchers, and less isolated for companies.

For enterprises, this could mean access to multidisciplinary teams that can prototype and validate AI tools before they are integrated into live systems. For students, it offers exposure to real business problems rather than classroom-only AI exercises. That is especially relevant in Southeast Asia, where demand for AI talent is rising faster than the supply of practitioners who understand deployment constraints in sectors such as transport, finance, logistics and public services.

Professor Susanna Leong, Deputy President for Academic and Provost at SIT, said the collaboration will create opportunities for students and academic staff to tackle “complex business issues” and gain hands-on experience applying AI to practical problems.

The early examples cited by the partners come from SNAIC’s work with public transport operator SMRT. One project, GENESIS (Generative AI Aided Safety Investigation System), was developed to automate parts of safety reporting and investigation. Instead of manually searching through past records, staff can retrieve relevant incident histories, rulebook-based guidance and possible mitigation measures.

Another project, AiDiSA (AI-Driven Intelligent Situation Awareness System), helps analyse commuter feedback from selected sources. It classifies and routes cases based on severity and urgency, tracks public sentiment, and alerts teams to issues that may need closer attention.

Also Read: 163,000 workers, 37% training: Malaysia’s AI skills gap in focus

These are not flashy consumer AI applications, but they show where enterprise AI may gain traction first: in repetitive, data-heavy, operational tasks where faster retrieval, classification and escalation can improve decision-making.

The competitive landscape

RE:AI enters a crowded and fast-moving market. Global cloud providers such as Amazon Web Services, Microsoft Azure, Google Cloud and Oracle are all pushing enterprise AI infrastructure and tooling across Southeast Asia. Microsoft and Google, in particular, have tied their cloud strategies closely to generative AI services and productivity software used by large organisations.

At the same time, regional telecom and infrastructure players are trying to capture demand for sovereign cloud, GPU capacity and AI-ready data centres. Singtel’s advantage lies in its combination of connectivity, enterprise relationships, data centre exposure through Nxera, and its digital services arm, NCS. Its challenge will be to show that RE:AI can offer more than infrastructure: enterprises will judge the platform by whether it shortens deployment timelines, supports compliance and delivers measurable business outcomes.

NVIDIA’s role is also important. The chipmaker has become the backbone of much of the global AI infrastructure buildout, with its GPUs widely used for training and running AI models. Its participation gives the partnership technical weight, but it also reflects a broader regional reality: Southeast Asia wants to build AI applications, yet much of the underlying compute stack is still shaped by global technology suppliers.

A Singapore play with regional implications

Although the partnership is Singapore-based, its implications reach beyond the city-state. Many Southeast Asian markets face similar barriers to enterprise AI adoption: limited specialist talent, fragmented data systems, regulatory uncertainty and difficulty moving pilots into production. Singapore often acts as the regional headquarters for multinationals and a proving ground for regulated technologies, so successful deployment models can influence how companies roll out AI elsewhere in the region.

For Singtel, the partnership also reflects a broader shift in the telecom sector. Connectivity remains core, but telcos are increasingly trying to move up the stack into cloud, cybersecurity, data centres and enterprise AI. The reason is straightforward: as AI workloads grow, demand for secure infrastructure, low-latency networks and trusted deployment environments will grow with them.

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

The question now is whether collaborations like this can avoid becoming yet another layer of innovation theatre. Enterprises do not need more AI showcases. They need systems that work under real constraints, with clear accountability, trained users and measurable value.

If RE:AI, SIT and NVIDIA can help companies make that jump, the partnership could become a useful model for applied AI in Southeast Asia: less about building AI for its own sake, and more about embedding it into the industries that keep the region moving.

The post Singtel’s RE:AI partners SIT-NVIDIA centre to tackle enterprise AI deployment gap appeared first on e27.

Singapore’s financial regulator has set out how banks, insurers, payment companies and other financial institutions should govern artificial intelligence, as AI moves from back-office experiments into systems that can influence customer outcomes, risk decisions and even execution.

The Monetary Authority of Singapore (MAS) has issued its Guidelines on Artificial Intelligence Risk Management, a principles-based framework that will take effect on 7 October 2027. Financial institutions will be allowed to implement the rules in phases, with core expectations around governance and risk management due in 2027, and additional requirements to be met by 7 October 2028.

Also Read: AI governance is moving from promises to proof

The guidelines apply to all financial institutions and all forms of AI technology. MAS is not prescribing a single compliance model. Instead, it is asking firms to calibrate their controls based on how extensively they use AI, the complexity of those systems, and the potential harm if something goes wrong.

That distinction matters. A bank using AI to summarise internal documents does not carry the same risk as one deploying AI to approve loans, detect fraud, price insurance or interact with customers. MAS’s message is that financial firms can innovate, but they must know where AI sits in their operations, who is accountable for it, how it is tested, and what happens when it fails.

“AI has significant potential to improve financial services, from enhancing customer outcomes and strengthening risk management to improving productivity and enabling new products and services,” said Ho Hern Shin, Deputy Managing Director at MAS. “Realising these benefits sustainably requires financial institutions to understand and manage the risks that come with increasingly capable AI systems.”

A risk-based rulebook, not a blanket ban

The MAS guidelines follow a public consultation held in November 2025, during which respondents supported a principles-based and risk-proportionate approach. In plain terms, this means the regulator is not trying to stop financial institutions from using AI, nor is it asking every firm to build the same governance machinery regardless of size or risk.

Financial institutions may use their existing governance structures if those structures provide adequate oversight and cross-functional coordination. They do not need to create a dedicated AI committee simply to satisfy MAS. This will be welcomed by smaller firms and fintechs, which often lack the resources of large banks but still use AI in customer support, compliance, data analysis or product personalisation.

At the same time, MAS is making clear that AI cannot be treated as a side project controlled only by technology teams. Boards and senior management are expected to provide effective oversight, set clear roles and responsibilities, define risk appetite, and ensure policies and procedures are in place.

Also Read: The compliance paradox: More checks, more fraud

This reflects a broader shift in how regulators view AI. Early AI governance discussions often focused on ethical principles such as fairness, explainability and accountability. Those still matter, but the rise of generative AI and agentic AI systems (tools that can generate outputs, make decisions or take actions with greater autonomy) has made operational resilience, cyber risk, third-party dependency and model failure much more urgent.

What financial institutions must do

The MAS framework expects firms to manage AI risks at two levels: across the enterprise and at the level of individual use cases.

At the enterprise level, financial institutions will need to understand their overall AI exposure. This means identifying where AI is being used, maintaining inventories with an appropriate level of detail, and assessing which applications are material from a risk perspective.

At the use case level, firms must apply controls across the AI life cycle. These include data governance, testing, human oversight, cybersecurity, monitoring and change management. The life-cycle approach is important because AI risk does not end once a model is launched. Models can degrade over time, behave differently as data changes, or produce unexpected results when integrated into new workflows.

The guidelines also cover third-party AI, one of the most difficult issues facing financial institutions. Many firms do not build their own AI systems from scratch. They rely on cloud providers, software vendors, embedded AI features in enterprise tools, and external model providers. MAS says financial institutions remain accountable for AI used in the services they deliver, even when that AI is developed, operated or supplied by third parties.

Firms must therefore obtain sufficient assurance from providers, assess whether third-party AI is suitable for their intended use, and apply compensating controls where there are gaps. If risks cannot be brought within the institution’s risk appetite, MAS says the firm should consider limiting, suspending or replacing the third-party AI service.

That is a notable signal to the market. As banks and fintechs race to integrate AI copilots, fraud detection tools and automated customer engagement systems, vendor due diligence will become more demanding. AI procurement will no longer be only a technology or commercial decision; it will become a regulatory and risk management issue.

Why this matters for Southeast Asia’s fintech sector

Although the guidelines apply to Singapore-regulated financial institutions, they are likely to influence AI governance beyond the city-state. Singapore remains a regional base for many banks, insurers, payment firms, digital asset companies and fintech startups operating across Southeast Asia. When MAS raises supervisory expectations, regional compliance teams often take notice.

Also Read: The EU called ChatGPT a search engine. SEA’s AI startups should worry about what comes next

This is especially relevant because Southeast Asia’s financial services market is highly digital but unevenly regulated. Digital banks, e-wallets, buy-now-pay-later providers, remittance platforms and lending startups serve large underbanked populations, often using alternative data and automated decisioning to manage cost and scale. AI can improve fraud detection, credit scoring and customer service, but it can also create risks around bias, opaque decisions, data misuse and over-automation.

For startups, the immediate challenge will be documentation and discipline. Many young companies use AI tools informally across product, engineering, compliance and support functions. The MAS guidelines point towards a future in which financial startups will need a clearer inventory of AI use, stronger vendor controls, and evidence that higher-risk systems have been tested and monitored.

This could raise compliance costs, particularly for smaller fintechs. But it may also give serious players a clearer path to enterprise partnerships and regulatory trust. In financial services, the ability to demonstrate responsible AI governance could become a competitive advantage, especially when selling to banks or expanding into regulated markets.

The next frontier: agentic AI

MAS also flagged agentic AI as an area for further attention. Agentic systems can operate with more autonomy, access tools and execute tasks across software environments. In finance, that could eventually mean AI agents that help with portfolio management, compliance investigations, customer servicing, treasury operations or claims processing.

The upside is productivity. The risk is that autonomous systems may take actions that are hard to predict, explain or reverse. In regulated financial markets, small failures can cascade quickly if they affect transactions, customer decisions or market behaviour.

MAS plans to consult the financial sector in 2027 on what additional guidance on agentic AI would be useful. This suggests the current guidelines are not the final word, but a foundation on which more specific expectations may be built.

Also Read: SEA’s insurers face a new question: what happens when customers have agents?

The phased timeline gives financial institutions room to prepare. But the direction is clear: AI adoption in finance is moving from experimentation to supervision. Singapore wants firms to use the technology, but not at the expense of customer trust or financial stability.

For Southeast Asia’s financial sector, that may become the defining balance of the next few years: how to capture AI’s productivity gains while proving that automated systems can be governed as carefully as any other part of the financial infrastructure.

The post MAS gives Singapore’s financial firms one year to prepare for AI risk rules appeared first on e27.

Every mature startup ecosystem has eventually faced a period the venture community refers to, after the fact and usually with discomfort, as its accountability phase. The United States went through one in the early 2000s, after Enron, WorldCom, and the broader dot-com governance cleanup. China went through one in the late 2010s, when Luckin Coffee and a wave of regulatory enforcement actions reset expectations around financial controls inside high-growth tech companies. India entered one in 2022-2023. ASEAN is now entering its version.

I have spent 15 years inside Indonesian risk functions, and the structural patterns that produce accountability phases in startup ecosystems are not new to anyone who has worked in regulated finance. The governance gaps that surface during these moments are not technical failures. They are the predictable consequences of growth velocity outpacing institutional maturity.

The historical pattern

Three things tend to be true of accountability phases across mature ecosystems.

They follow rapid scale. Every ecosystem that has been through one entered the phase after a multi-year period of high-velocity capital deployment, valuation expansion, and founder-led decision-making. The governance infrastructure that should have accompanied that scale was outpaced by operational growth.

They surface in clusters. Accountability phases are rarely about a single company. They are about a cluster of disclosures, investigations, and resignations that arrive within 12 to 24 months of each other, often involving companies with overlapping investors, advisors, and audit relationships. The cluster reveals what individual cases sometimes obscure: the gaps were systemic.

They produce structural change. The ecosystems that handled their accountability phases best did not stop at prosecuting individual cases. They used the moment to install governance infrastructure (board independence requirements, audit committee standards, founder oversight mechanisms, investor diligence norms) that the previous era of growth had skipped.

Also Read: Startups keep scaling ops before they scale data — Here’s why it backfires

Why these phases happen

Three structural forces produce them predictably.

Velocity outpacing governance. Founders are trained to decide quickly. Boards staffed primarily by founder allies and friendly investors do not provide effective challenge. Audit committees in pre-IPO companies are often nominal. Internal financial controls grow more slowly than the revenue figures being reported. Each gap is individually defensible during growth. Together, they create the conditions for failures that only become visible when external pressure forces examination.

Diligence theatre. Late-stage venture investors, particularly in markets with limited public-company comparables, often run diligence processes that test surface metrics more rigorously than underlying operational reality. The same diligence playbook, repeated across deals and across investors, produces consistent blind spots.

Founder isolation. Founders of unicorn-scale companies eventually reach a point where almost everyone in their daily orbit benefits financially from the company’s continued narrative. The mechanisms that should challenge the founder’s interpretation of reality (independent directors, external auditors, internal risk officers with real authority) are often the same mechanisms that have been quietly weakened during growth.

What changed for the ecosystems that handled it well

Investor diligence standards reset publicly. The US after Enron, China after Luckin, India after 2022-2023: each ecosystem saw major institutional investors publicly upgrade their diligence frameworks. The new standards became the baseline, and companies that resisted them faced funding consequences.

Board governance norms changed. The composition, independence, and authority of pre-IPO company boards shifted meaningfully. Independent directors with real veto power became more common. Audit committees got teeth. Founder-CEO dual roles in companies above a certain valuation began carrying additional governance requirements.

Auditor scrutiny increased. The professional services firms that audited and advised during the growth phase faced their own accountability moment shortly after the founder cohort did. Standards tightened. The cost and rigour of independent verification increased.

Also Read: Startups should learn to leave bad markets faster

Lessons learned

Five principles from previous accountability phases are worth carrying into ASEAN’s current one.

The cluster will get bigger before it gets smaller. Accountability phases tend to surface 12 to 24 months of issues within a relatively short period. The ecosystem response is more durable if it treats the cluster as predictable rather than as a series of individual surprises.

Diligence framework upgrades are collective infrastructure. The fastest way to upgrade due diligence is for the largest investors to coordinate publicly on new standards. Individual firms upgrading alone get adverse selection. Coordinated upgrades change the market.

Independent directors are not luxury. Independent board seats with actual authority, not friendly investor representatives, are the single highest-leverage governance investment a unicorn-stage company can make.

Founder-CEO accountability needs explicit structure. The presumption that the founder will act in the company’s long-term interest is necessary but not sufficient. Specific mechanisms (board oversight, audit committee authority, risk officer independence) should be in place before they are needed.

Transparency is the recovery accelerant. The ecosystems that recovered fastest from their accountability phases were the ones where the institutions involved disclosed honestly and quickly. The ones that protected reputations through opacity recovered more slowly.

The macro stakes

ASEAN’s startup ecosystem has spent 15 years building. The capital deployed, the talent attracted, the regional infrastructure built, none of that is at risk from the current moment, unless the ecosystem responds to it with denial rather than with structural improvement.

The accountability phase, when handled well, is not a setback for an ecosystem. It is the moment an ecosystem grows up. The institutions that come out of it stronger will be the ones that treat the period as collective standard-setting rather than individual damage control. The question is whether ASEAN’s response will match the scale of what has been built.

—

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’s startup ecosystem is entering its accountability phase appeared first on e27.