Live selling is the most human form of e-commerce there is. A host talks for hours, answers questions by name, jokes with regulars, and builds the kind of trust a product page never will.

It's also exhausting, expensive and impossible to scale. Hosts get tired. They speak one or two languages. They can't be live at 3 a.m. when a shift worker in Jakarta or a student in Pune is finally scrolling.

AI avatars are changing that, fast. But the real story isn't "AI replaces the host." It's more interesting than that.

Live commerce is now a core channel in Asia

Southeast Asia shows how big this has become. Momentum Works' Live Commerce in Southeast Asia 2026 report found content commerce GMV across Shopee, TikTok Shop and Lazada reached US$49.7 billion in 2025, up 98% year on year, and is on track for US$77.9 billion in 2026. In the first half of 2026, content commerce made up 37% of the region's platform e-commerce GMV, up from just 20% in 2024.

India is earlier on the curve, but moving. Research firms disagree on its exact size, but most expect Indian live commerce to grow at over 40% a year this decade, driven by mobile-first shoppers and, crucially, regional languages.

China, as usual, shows where this goes next: digital humans now run a large share of e-commerce livestreams there, mostly at night and for catalogue selling, at a fraction of the cost of human-run streams.

The economics are hard to ignore

The cost gap is the first thing every brand notices. Momentum Works found that in some cases, AI livestreams in Southeast Asia cost only 20 to 25% of a human-hosted operation.

The results can be striking too:

  • In China, an AI version of tech personality Luo Yonghao co-hosted a six-hour livestream on Baidu's platform that drew 13 million views and about US$7.6 million in sales, outperforming his own earlier human-hosted session.
  • In Malaysia in 2026, a skincare brand ran an always-on AI host on Shopee Live as a 24/7 "skin advisor": diagnosing concerns, explaining ingredients and suggesting pairings. It broadcast for more than 300 hours in a month, against roughly 20 hours with human hosts, and reported Shopee Live sales up 299%. The human hosts didn't disappear; they moved to big tentpole events and creative direction.
  • In March 2026, Shopee began testing live commerce hosted by licensed, AI-generated digital likenesses of celebrities.

But here's the part most people skip

AI avatars don't automatically sell. A University of British Columbia study of 328 products found human streamers significantly lifted sales, while digital streamers delivered only a small improvement over having no streamer at all.

That matches what you hear from people who run live commerce in Southeast Asia. The format works because of relationships: hosts who talk for hours, take live questions and build a loyal base of repeat viewers. Strip out the human connection and you have a talking catalogue.

So the question isn't "avatar or human?" It's "what job is each one best at?"

Where AI avatars genuinely win

From the work I've seen across markets, the strongest use cases fall into seven groups:

  1. Always-on selling. Avatars cover the long, quiet hours (overnight, weekdays, off-season) so the store never goes dark. Humans take the prime-time, high-energy slots.
  2. Every language, one host. India has 22 official languages; Southeast Asia speaks Bahasa, Thai, Vietnamese, Tagalog and more. A single avatar can sell in all of them, so regional shoppers get a host who speaks their language, not a subtitle.
  3. Licensed creator and celebrity twins. Creators can "be" in more places at once, under contract and with consent. A celebrity's digital likeness can host a launch in five markets on the same night.
  4. Virtual brand custodians. An AI version of a brand's face or ambassador that lives inside ads, apps and websites and answers questions in real time. It turns a one-way ad into a conversation, which is where interactive advertising and AI avatars meet.
  5. Explaining complex products. Loans, insurance, gold loans, electronics, cars, real estate: categories where people have questions before they buy. An avatar can walk each shopper through their own questions, step by step. (Regulated advice still needs proper compliance; an avatar doesn't make a regulated conversation unregulated.)
  6. After the sale. Setup guides, troubleshooting, warranty questions and reorders, delivered by the same familiar face that sold the product.
  7. Training and enablement. Onboarding sales teams, partners and dealers across markets, consistently and in their own language.

The playbook: human-led, AI-extended

The brands getting this right follow a simple model:

  • Humans own the relationship. Big launches, festive sales, creator collaborations and anything that needs genuine personality stay human.
  • Avatars own the coverage. Long hours, extra languages, repeat questions and product education go to AI.
  • Knowledge makes or breaks it. The best avatar performances come from deep product knowledge (ingredients, specs, comparisons, offers), not just a realistic face. A beautiful avatar with shallow answers loses trust faster than no host at all.
  • Disclose, always. Tell viewers they're talking to an AI host. Trust is the whole point of live commerce; don't spend it on a trick.

The honest caveats

  • Uncanny still loses. If the avatar looks or sounds slightly off, viewers leave.
  • Consent and likeness rights are non-negotiable. Digital twins of real people need clear contracts, approvals and limits.
  • Returns can rise. Live selling drives impulse purchases; scaling it with AI can scale returns too, if expectations aren't set honestly.
  • Platform rules are moving. Marketplaces and regulators are still deciding how AI hosts should be labelled and where they're allowed.

The bottom line

The future of live selling isn't humans versus AI. It's humans for moments that matter and AI for every other hour, in every language, for every question.

The brand that never sleeps, speaks every language and still feels human will win the next decade of commerce in Asia.

— Aditya Tibrewal


Sources