AI-generated livestream hosts are becoming more realistic and more widely used in Southeast Asian commerce, but the harder question for brands is where they actually add value compared with human presenters.
AnyMind Group recently updated its AnyLive platform with more lifelike AI avatars and analytics that compare human-led and AI-led livestreams. TNGlobal previously covered the AnyLive update, including its focus on longer operating hours, cross-stream analytics, and video-based avatar generation.
In this TNGlobal Q&A, Kubo Akinori, Managing Director of Global E-Commerce at AnyMind Group, discusses how brands should compare human and AI hosts, where disclosure matters, how localization should work across Southeast Asia, and which safeguards should remain human-led.
AnyLive now brings analytics across both human-led and AI-led livestreams. Which metrics are most useful for comparing the two fairly, and where do you see AI hosts currently outperforming or underperforming human hosts?
Evaluating AI avatars against human hosts strictly on peak hourly output is flawed. Our take is that human hosts should be active during prime-time campaigns and promotional periods, while AI avatars operate during hours with lower traffic but still possess latent consumer demand. So rather than asking which host generates more GMV per hour, brands should look at the bigger picture: how much additional coverage and value are they getting from having both?
Metrics like incremental off-peak GMV, cost per livestream hour, channel coverage, and customer interactions can give a more complete view. AnyLive brings these signals together so brands can look at the overall impact rather than treating it as a competition between human and AI hosts.
The two also have different strengths. AI avatars are particularly useful when brands need consistent, scalable coverage, for example, keeping a livestream running for longer hours, delivering consistent product information, or expanding across markets without having to scale up the same level of production resources. Meanwhile, human hosts are especially valuable during high-impact moments where personality, storytelling, and real-time judgment can make a difference.
So rather than choosing one over the other, brands should leverage human creators for connection-driven moments while using AI to maintain continuous storefront availability.
Are there particular product categories, audience segments, or stages of the shopping journey where AI avatars work especially well, and others where human presenters still have a clear advantage?
One area where AI avatars can really stand out is in interactive product expertise. Brands already have a wealth of detailed knowledge about their products, from skincare ingredients and formulations to the technical specifications of electronics. AI can make that knowledge more accessible during a livestream by responding to highly specific questions in real time and consistently drawing from the information the brand has provided.
This becomes especially interesting when you combine that capability with off-peak audiences. A consumer browsing at 2 AM may have very specific questions about whether a product is suitable for them or how two products compare. If there is no human host online at that moment, that opportunity can easily be lost. AI gives brands a way to make their product expertise available whenever those consumers are ready to engage, rather than simply keeping a livestream running for longer.
Human hosts remain particularly important for audiences looking for entertainment, trend inspiration, and social connection. Viewers who come for a creator’s personality, community, humor, or personal recommendations respond differently to a human host, who can read the room, react spontaneously to comments, and create the energy and urgency that are especially important during major campaigns.
As AI avatars become more lifelike, how should brands approach disclosure and consumer trust? Do you think viewers should always be told explicitly when a livestream host is AI-generated?
Transparency should be part of the experience as AI livestreaming becomes more lifelike. The good thing is that platforms have already started enabling this disclosure. Viewers should be able to know when they are interacting with an AI host, particularly when the host is answering questions, making product recommendations, or guiding purchasing decisions.
For brands, disclosure does not have to disrupt the shopping experience. A clear on-screen label or brief introduction can establish transparency while allowing the AI avatar to focus on product discovery, education, and real-time engagement. The goal is to make AI feel like a trusted part of the shopping experience, rather than something being presented as human without explanation.
For platforms like AnyLive, this also means pairing AI scalability with responsible brand oversight. Brands should define clear boundaries around what the AI host can say, ensure product information is accurate, and continuously monitor interactions and performance. Trust ultimately comes from transparency, accuracy, and giving consumers a clear understanding of who, or what, they are engaging with.
Live commerce in Southeast Asia spans different languages, shopping behaviors, and cultural expectations. What does effective localization involve beyond translating the script or changing an avatar’s appearance?
Brands should adapt the selling approach to each market, from product recommendations, promotions, and shopping moments to cultural references and consumer needs. What resonates with shoppers in Malaysia may not work the same way in Thailand, Indonesia, or Vietnam.
For example, a skincare brand running an AnyLive session across Southeast Asia could adapt product recommendations by market. The AI host in Malaysia might focus on skincare concerns relevant to a hot, humid climate and recommend lightweight products, while in another market it could prioritize different concerns, product combinations, or promotional bundles based on local consumer preferences and purchasing behavior.
Additionally, using market-specific consumer and performance data to continuously refine the experience is crucial when localizing experiences for shoppers. Brands can look at the questions consumers ask, the products they engage with, viewing behavior, conversion patterns, and promotional responses in each market, then use these insights to optimize future livestreams. For instance, if data from a livestream in Thailand shows that viewers frequently ask about skincare product combinations, while Malaysian viewers have concerns about daily wear in humid weather, AnyLive can adjust automated responses and product emphasis for each channel.
Ultimately, effective localization is about making the livestream feel locally relevant while maintaining a scalable regional operation, combining AI’s ability to adapt content and operate across markets with local market knowledge and human oversight.
AnyLive can generate avatars from existing video footage and create digital twins of mascots. What consent, intellectual-property, and identity safeguards should brands put in place before creating and deploying those digital representations?
Before creating a digital representation, whether a digital twin of a human host or a brand mascot, brands should establish clear consent and usage rights with stakeholders and relevant parties whose likeness, voice, performance, or creative work is being used. This should cover where and how the digital twin can be used, the duration and markets involved, whether it can be modified or reused for future campaigns, and what happens when the relationship or license ends.
For mascots and other brand assets, brands should also confirm that they have the appropriate intellectual-property rights to create derivative digital versions.
Brands should also put identity and content safeguards in place before deployment. This can include defining what the AI avatar is permitted to say or do, reviewing generated content before it goes live where appropriate, protecting the underlying source footage and identity data, and clearly documenting who owns and controls the resulting digital asset. These safeguards become particularly important when the same avatar is deployed across multiple markets and campaigns.
Ultimately, the technology should operate within a clear framework of consent, ownership, permitted use, and brand oversight. This allows brands to scale the use of digital representations while maintaining control over how their creators, mascots, and intellectual property are represented.
Cross-stream analytics can be difficult because platforms expose different data and define engagement or conversion differently. How does AnyLive normalize those signals, and what should brands be careful about when attributing sales or performance to an AI-hosted stream?
Because each platform can define metrics differently, we have established a consistent measurement framework across platforms on AnyLive. For example, AOV, efficiency, add-to-cart, GMV per hour, and similar measures should be mapped to common definitions wherever possible, while platform-specific metrics are kept separate rather than treated as directly comparable. AnyLive can then use these normalized signals to evaluate performance across different livestreams, markets, and operating periods.
When it comes to attribution, it really depends on the attribution model the brand is using. For example, if a brand uses last-touch attribution, a purchase that comes directly after an AI-hosted livestream interaction may be attributed to that stream. With first-touch, linear, or other attribution models, the same purchase may be credited differently across multiple touchpoints.
So rather than saying that AI should or should not receive attribution, brands should be clear about the methodology they are using and apply it consistently when comparing AI-hosted and human-led commerce.
The goal is to understand the additional value created by extending livestream coverage, rather than simply assigning all sales during an AI session to the AI host. This gives brands a more realistic view of how AI livestreaming contributes to the wider commerce operation and fits into the wider customer journey.
Where should human intervention remain part of an AI-led live-commerce workflow, particularly when a viewer asks an unexpected question, raises a complaint, or needs information that could create a financial, legal, or reputational risk for the brand?
AI avatars can handle typical product questions, demonstrations, and FAQs during off-peak hours, but situations involving complaints, financial claims, legal questions, or potential reputational issues should be handled by a human. The AI should not attempt to respond beyond its defined scope. Instead, the customer should be directed to the appropriate customer-service channel for further assistance.
Humans also need to continuously review AI-run livestreams and improve the systems behind them so that the experience becomes more accurate and effective over time.
That’s where the human-in-the-loop approach becomes valuable. AI can manage routine interactions and keep commerce running during long-tail hours, while human teams remain responsible for cases that require dedicated customer service and for continuously improving the AI experience. This allows brands to scale live commerce while maintaining appropriate safeguards for customer trust and brand protection.
Over the next 12 to 18 months, what evidence would convince you that AI-hosted live commerce has become a durable operating model rather than primarily a cost-saving or novelty tool? Which measures should brands watch beyond stream volume or production cost?
To show that AI livestreaming is not just a passing trend, we have to look at whether it actually drives long-term growth across the whole business. In Southeast Asia’s growing content-commerce market, brands need to look beyond simple metrics like total hours streamed or hosting-cost reduction. What matters more is whether AI-led livestreaming is actually helping them generate additional revenue, reach more consumers, and build longer-term customer value.
Over the next 12 to 18 months, I’d look at a few things:
- Are brands generating incremental GMV during off-peak hours that previously had little or no coverage? This is a good indicator of whether AI livestreaming is actually capturing demand that the brand would otherwise have missed.
- Are they actually using the consumer insights from those streams? For example, if viewers repeatedly ask the same questions or raise similar objections, are those insights being used to improve creator briefs, product messaging, or future campaigns? Are they also being used to improve the actual product or go-to-market strategy across other channels?
- Are those customers coming back and purchasing again? Repeat purchases and customer retention can help show whether AI-led streams are attracting valuable customers, rather than simply driving one-off transactions.
- How efficiently can brands extend their live-commerce coverage across platforms such as e-commerce marketplaces and brand.com sites without having to scale their operating resources at the same rate? This helps brands understand whether AI is creating a more scalable operating model.
If AI livestreaming can consistently generate additional GMV while also giving human creators better insights and allowing them to focus on the moments where they have the most impact, that’s when I think it becomes more than a cost-saving tool. It becomes a sustainable part of how brands operate live commerce.
What is your key takeaway on AI livestreaming?
Rather than defining the roles of human hosts and AI avatars by when they generate the most sales, I believe human hosts should increasingly focus on delivering emotional value, such as engagement, connection, and entertainment. As AI avatars become more prevalent, the emotional value that human hosts can provide to audiences will become even more important.
The reason we want AI avatars to look and behave more naturally is to provide viewers with an experience as seamless as possible compared with traditional human-hosted livestreams. We’ll continue improving how lifelike our AI avatars are, but what’s even more important is how quickly we can run A/B tests, gather feedback and market signals at scale, and use those insights to continuously improve content-commerce performance.
The goal of AI Live is not to replace human hosts with AI hosts. Our broader vision is to use AI technology across livestream commerce, including human-hosted sessions, to better understand audiences, optimize content, and ultimately improve the shopping experience and purchasing value for consumers.
Kubo Akinori is Managing Director of Global E-Commerce at AnyMind Group. He leads the company’s digital-commerce strategy across global markets, including live streaming and automated commerce.
Editor’s note: This Q&A has been lightly edited for clarity and TNGlobal house style. The substance of the interviewee’s responses has been preserved.
Share your perspective: TNGlobal welcomes contributed insights and expert commentary from across Asia’s technology and innovation ecosystem. Submit a contribution for editorial consideration, or explore more conversations in our TNGlobal Q&A and Interviews archive.
AnyMind updates AnyLive with more lifelike AI avatars and cross-stream analytics