Live commerce in Thailand is no longer an experiment. It is a channel with its own operating rhythm, its own production grammar, and its own set of expensive mistakes. This is what we have learned running it.
#The channel is a habit, not an event
The most common failure we see is treating a stream like a campaign launch. A brand books a big set, flies in a celebrity, spends heavily on a single three-hour broadcast, sees a respectable spike, and then goes quiet for six weeks.
That spike does not compound. Live commerce audiences are built on appointment viewing — the same slot, the same host, the same week-to-week rhythm. Viewers who know that you are live every Wednesday at 8pm will show up on Wednesday at 8pm. Viewers who saw one impressive stream in March will not remember you in April.
A modest weekly stream, run for twelve weeks, will almost always outperform three large streams run quarterly, at a fraction of the cost.
#Host structure decides the back half
Energy decay is the single biggest driver of drop-off on long streams. A solo host is fine for sixty minutes. Past ninety, their pace slows, the pauses lengthen, and the concurrent viewer count follows them down.
Two hosts fix this — not because two people are twice as interesting, but because conversation regenerates energy in a way monologue cannot. The practical structure:
- A lead host who owns the product knowledge and the calls to action
- A support host who reacts, asks the questions the audience is typing, and takes over during product swaps
- A moderator off-camera feeding real comments to the support host, so the on-screen conversation tracks the actual chat
That third role is the one brands cut first and regret fastest. Unmoderated chat drifts, and a host reading their own comments cannot maintain pace.
#Platform mix follows the product
| Platform | Best for | Viewer mindset |
|---|---|---|
| TikTok Shop | Beauty, fashion, snacks, impulse categories under ฿1,000 | Discovery — was not looking for you |
| Shopee Live | Home, electronics, bundles, mid-basket | Shopping — already had the app open |
| Lazada Live | Considered purchases, brand-official stores | Comparison — checking before committing |
| Facebook Live | Established community, older demographics, regional reach | Loyalty — already follows you |
Most brands should simulcast rather than choose. The production cost of adding a destination is near zero once the control room exists; the audience overlap between platforms is smaller than people assume.
#Redundancy is the product
A live stream has exactly one hard requirement: it must not stop. Everything else is a preference.
We run every broadcast on a dual path — primary fibre with a bonded 5G secondary, and two encoders running hot in parallel. If one leg fails, the other is already pushing the same keyframes and the audience sees nothing.
A ten-second dropout during a pinned-product window costs more in lost carts than the backup connection costs for a year. This is the cheapest insurance in the entire production.
Every camera channel is also recorded in isolation at source. Even in a total transmission failure, the client leaves the building with the full multi-cam masters.
#The stream is raw material
The most under-exploited asset in live commerce is the footage itself. A two-hour show contains, typically:
- Six to ten clean product demonstrations, each usable as standalone paid social
- Two or three genuine host reactions worth cutting as organic content
- A full transcript that can be mined for the questions customers actually asked
- Sales-velocity data mapped against the exact minute each product was pinned
We render the first cutdowns within two hours of wrap. By the next morning the client has a folder of vertical assets, ranked by comment velocity during the moment they were captured — which is a far better performance signal than anything you could guess in a planning meeting.
#What we would tell a brand starting now
Start smaller than you think and more often than is comfortable. One camera, two hosts, a moderator, a fixed weekly slot, and a redundant connection. Run it for a quarter before you spend a baht on set construction.
Then look at the data and build the studio around what actually sold.