GemPages Shopify Page Builder App
내용
[subtitle-meta] lang=en auto=False
...and joining us on stage, we have three speakers bringing different perspectives
on the personalization ecosystem. I would like to invite to the stage Mr.
Ethan Nguyen from Pengu, on how design and creative teams leverage
AI to supercharge ideation. And we would also invite Ms.
Trang Vu from BettaMax, breaking down operational frameworks and systemic scaling
to help POD businesses. And we also want to invite Mr.
Tung Mark from Easify, walking us through how AI powers hyper-personalization,
connecting product creation all the way to final checkout. And lastly,
please welcome the beautiful lady, our last speaker, Miss Tracy from Printbelle,
breaking down Vietnam's POD landscape — the real GMV data, market volume, and brand strategies
to capture global demand. So ladies and gentlemen, now the time has come.
Let's explore how POD, personalization, and AI are coming together to
create new opportunities for ecommerce brands. So now, the stage is yours.
Thank you, MC. This session is probably the first time since this morning that I'm speaking in Vietnamese.
Hello everyone, I'm Sammy Ha, a POD seller. I'm very happy that ShopX invited
me to attend today, and I'm honored to be the moderator for today's panel discussion.
As a POD seller myself, I also have a lot of what you might
call burning questions. I'm quite curious about them, and I'd like to invite all
of today's speakers on this panel to answer my questions, as well as those
of everyone attending today's event. But first, I'd like to take a moment to let
our guests introduce themselves and their companies. May I invite Ethan to go first?
Hello.
Thank you, Sammy, for the introduction. Hello everyone. I'm Ethan; currently I'm the CTO of Pengu.
I'm very happy to be joining the ShopX event this year.
I think...
Hello Sammy Ha, hello everyone, and hello to everyone sitting below. I'm Trang from BettaMax,
and BettaMax is an e-commerce platform dedicated specifically to dropshipping and POD. BettaMax is aiming for
a solution where every seller can run all of their operations on a single platform.
Hello everyone. Very happy to be at ShopX today. I'm Tung —
in English you can call me Mark — currently CEO at Easify.
I have over 10 years of experience providing e-commerce solutions,
and currently we provide Shopify apps focused on custom products and personalization.
Hello everyone. I'm Tracy.
Currently I handle the Vietnam market for Printbelle. Printbelle is a POD
fulfillment company that supports POD sellers to develop new products and ship worldwide.
So, my first question is for all four speakers today. I can see there
are quite a few familiar POD-seller faces down here. This panel will discuss the POD
industry in some depth. Q4 is very close now — it's a peak season.
So in your view, what should sellers prepare right now?
Feel free — whoever wants to go first, please jump in.
Hi everyone. From Pengu's perspective, sellers should prepare a very solid idea-research process and
workflow — and that's actually the problem Pengu is aiming to solve.
Our goal in releasing Pengu is to help sellers find winning ideas faster in the market.
A winning idea here has two elements.
First, relying on historical data to know the patterns — for example, what motifs a niche has, what wordplay works.
Second, detecting very quickly the new trends appearing in the market. That's my point of view on what sellers
need to prepare, and I think it's also the strength — the unique selling point — of each team.
Let me answer your question...
This feels like a beauty pageant.
I also agree with what Ethan just shared — that we need to prepare research as
well as samples. From BettaMax's side, I'd add that we also need a good audit process
so we can test products faster, make quicker assessments of what we've just tested,
speed up our progress, and find what works so we can scale it during this year's BFCM.
To answer this question from my side: to prepare for BFCM
— based on experience working with POD brands on Shopify stores — first,
you should prepare a detailed BFCM plan, then set it
up and have it ready to run the moment BFCM launches.
And from the production side, I think every seller needs to optimize their operations first, then find fulfillment
partners that are stable and accurate, so they can meet the demand to scale in the year-end season.
So, based on what our four speakers shared, in Q4 you can't win
with a single lever alone — it comes from preparing a whole plan,
including product, operations, ads, and audit. First, let's follow the journey step by step.
Product is a very important factor — product is king.
So Ethan, could you share how to create what we call a "winning" product?
How do we make customers want to buy — create a product
customers are ready and eager to buy? Could you share your view, Ethan?
From my perspective, we shouldn't talk about a winning product in vague terms.
Especially when working with large teams, we need what I call a formula to find a winning product.
POD, although a new industry, already has historical data in the market.
So the way our team analyzes a winning product is based on past data. For example,
at Pengu we crawl several million listings in the market, then analyze, for each niche,
what motifs the audience tends to like, what wordplay they like, and which copy
has sold the most. From there we come up with the product idea. Second,
detecting new trends in the market. When we mix those two factors, we find a
product with a higher probability of winning. That's my point of view on winning products.
I'd like to go deeper into finding a winning SKU.
OK, let me give more detail about Pengu. Pengu is a product focused
on helping sellers find winning ideas and winning products in the market.
Every day, Pengu crawls POD from popular platforms — for example Etsy, Shopify,
Facebook Ads. Then for Etsy we rely on sales and favorites;
for Facebook we rely on reactions, comments, and shares. From there we extract different patterns.
Actually, recently, as people use AI, you'll notice the way AI looks at a winning product is very different.
Sometimes we, as humans, look at a product and get the feeling it will win.
But for AI, everything has to be put into a formula. That's what Pengu is aiming for.
When analyzing a product, we aim for AI to be able to evaluate what wordplay it has,
what typography, what emotional hook, what color, what layout. You can imagine we try to turn
every listing into a full formula and analyze it over a large volume of data.
Then we start to know, for a given niche family, which copy and which motifs tend to go together.
The second step, after having a knowledge base about the market and historical data,
is to work out how to catch the newest things. Because relying only on historical data would
be quite outdated, so we need to combine past experience and patterns with the trends
and new ideas appearing in the market. That's the real-time detection module, and by combining those two,
we help sellers detect which winning idea is starting to emerge in the market.
Pengu's goal — and my goal for the Pengu team — is
not to replace humans, but to help a POD team.
Instead of a 100-person team where an operator can handle about 4 to 5 brands,
now you only need a few people combined with AI to
detect winning ideas very fast and operate a full POD brand. OK.
I have a question for Ethan.
Right now everyone knows how to use AI to make products, right? I'm sure no one here doesn't use it.
So with the same workflow and the same way of thinking, is there a chance
the outputs will be products that look very similar to each other, and all products in the market
end up looking the same? So what solutions does Pengu have to create product differentiation so we
can win customers? OK. Actually, this was a problem we had to think about when we started building the product:
if we crawl market data and everyone uses that same data, everyone will generate similar
ideas and products. So to solve that, we worked with many POD teams and noticed the following pattern.
Although the market data is the same, each POD team has strengths in different niches.
For example, some teams are very strong at personalized products for best friends, and others are very strong
at selling farmer niches or gardening niches. So Pengu's goal is, first, to provide market signals
and insights based on data, AI, and big data; and second, to combine that with each
seller's own data to generate a new idea that fits what I call the DNA of
that POD team. That helps make that team's or seller's idea unique, instead of being very common
in the market and based only on generic data. That's how Pengu is solving it currently.
OK, one more question. If there are new sellers here just starting out in
POD who don't have any database to rely on yet, where should they start?
I think not having data is only the case when you're just starting, when you have nothing yet.
But at Pengu, for example, we solve that by having people start a store,
start listing, and run ads — then you'll definitely have traffic metrics.
You'll know total views, CR, add-to-cart. Based on that data, Pengu will collect and optimize the generation of new ideas.
So essentially it's a loop: from idea to listing to store, get feedback metrics
from customers on that listing, then loop back to optimize the idea and the product.
OK, thank you, Ethan. Ethan shared a lot about how to build a product with the potential to win.
Now, there are many teams in the market and the volume of products they
create daily is very high, because thanks to AI, anyone can make that many products now.
But whether they sell or not — will the number of buyers grow 10x or 100x?
I'd guess buyers, or our customers, stay about the same, because you can't just create more people.
So we hit a bottleneck: many products, but how do we convert, how do we increase conversion?
For this part, may I invite Tung Mark to share his experience on how to improve conversion?
This question is for Easify, right?
That's right.
Yes. Let me reintroduce myself — I'm Tung, Tung Mark, from...
Sorry, Tung Mark
Could you give me the Easify slides, please?
To answer this question, I'd like to continue a point Ethan just made.
Ethan shared that POD teams are currently using AI to create products more easily and faster.
However, from experience working with brands that have personalized products,
I think we should also ask: when AI helps merchants create more products,
does that really help us sell more, and does it make buying easier for customers?
So, the advantage of AI is creating products faster. This advantage isn't just for one store;
right now many stores can also create products and expand their catalog that fast. However,
more choices don't necessarily make it easier for buyers, because with many choices they have to weigh more,
compare more, and sometimes during the buying process they feel stressed and overwhelmed when making a decision.
Often the purchase gets interrupted, and in many cases the buying journey ends entirely.
So, to make it easier for customers to buy,
there are two things to note. First, let customers personalize the product themselves
according to their preferences and purpose. Second, we can apply AI
to that process to make personalization more intuitive, so customers can clearly
see their final product and feel more confident when deciding to buy.
Let me give two concrete examples of personalization when a customer buys a product.
When a customer wants to buy a product — for example, a T-shirt like this
— the usual way is: the customer orders first and then sends the photo,
or many stores have a photo-upload field so the customer can upload before add-to-cart. However,
the problem is that when the customer uploads the photo, they can't yet picture
whether that photo suits the shirt or what the final shirt with their image will look like.
When it's still vague to them, it's hard to decide to buy right away.
To solve this, there's a solution: live preview. AI-powered live preview helps here.
When the customer uploads a photo, AI can remove the background or convert the
photo's style to suit the shirt — for example, grayscale — then show the
result directly in the live preview. Now the customer sees it's no longer a generic sample product;
it's their product with their image, almost exactly what they'll receive after purchase.
That moment of seeing your choice become a real product makes customers more excited
and makes it easier for them to add to cart and complete the purchase.
Another example: a customer wants to buy a hat and embroider their pet on it.
After uploading — they can upload just an ordinary photo from their phone — AI can remove
the background and convert that photo into an embroidery effect, then display it directly on
the hat like this. And you can see it has the effect of a real product.
So in this personalization part, AI doesn't just help the
merchant create more products; AI also helps customers feel more
confident and clearly see their product before deciding to buy.
That's how AI helps sell a personalized product better. However, there's another issue:
for teams and merchants selling gifting — gift boxes —
how do we make the buying journey simpler? Take the Father's Day campaign:
when a customer comes to your store wanting to pick a
gift box for dad, they already have an idea but still
don't know exactly what to choose. Normally they'd have to go into one collection,
pick a shirt, go back, pick a jacket, go back,
pick other accessories. In this process they visit many pages and sometimes
can't remember what they've chosen, whether the items go together,
or whether they've picked enough. It's very hard. Our catalog doesn't lack products.
However, these choices are scattered across many pages. The solution — you
may already know it — is Build-A-Box. This Build-A-Box solution brings all those
scattered choices into one journey, making it easier for customers to buy
and complete their gift box. For example, in this Build step: first,
the customer is on a single page. As the first step,
the customer can choose a T-shirt, polo, or hoodie; then choose outerwear or accessories.
And the final step can add a message or note in the card.
During this process, the customer can see the products they've
chosen in the summary box, and see the total and prices updated live,
instantly. So Build-A-Box gives customers an easier shopping journey and clearer visibility.
As for what AI does here: to let the merchant
create an experience like this, AI can help the merchant start faster.
AI can scan the store's catalog and suggest which product
sets can be combined. It can also suggest how to set up
discounts or offers for which product sets. And once set up,
AI can help set up a Build-A-Box page instantly. Beyond that,
we can use AI to scan whether our Build-A-Box page is good enough,
or needs improvement to increase CVR or AOV, before launching it.
So you can see AI can plan a box and build it right away.
So in this Build-A-Box part, AI can help the merchant set
up a shopping journey, make the journey easier, make the customer experience smoother,
and the merchant can also increase AOV more naturally.
Those are two solutions — one for a personalized product and one for increasing AOV via Build-A-Box — where merchants can apply AI.
Let me ask a bit. For this Build-Your-Box solution, you said its
function is to suggest to the seller which set to combine with which,
and what discount to use. Could you tell me the philosophy behind
what to combine with what, and what discount to use,
so I get a set that sells the most and gives the most profit?
What if the discount is too big and I lose money — maybe I don't need any discount and customers still buy?
And is combining these the right call? How do I get data feedback so I know whether combining set
1 with set 2 is right, or set 1 with set 3 is right? What options does Easify have for this?
Thank you. Great question.
In the Build-A-Box that Easify is building, the final decision is still the merchant's.
As for the suggestion mechanism, AI asks the merchant the most basic questions.
First, what's the product category? Second, allow it to scan the store's entire catalog. Then,
based on training from the user's side, it can suggest which collections can be combined and
give the merchant options. The merchant can then decide on option one, two, or three.
From the merchant's answer, it continues to the offer part. For offers, based on the
customer's choices and the catalog, the merchant can rely on the options AI lays out —
and on their own customers and profit margin — to decide which offer is suitable.
So AI provides options for the merchant to choose from. Instead of having no option in mind, AI provides one.
So AI won't give the exact answer, but provides options, and finally the merchant decides,
right? Now my second point: how do I measure whether the choice I made earlier was correct?
Because Easify is only providing options — AI is helping the seller with options,
not the decision. OK, so now I pick among four options; I close my eyes
and pick option A. How do I know if it's good, so next time I
keep choosing A, or if it's not good, so I stop and pick B, C,
or D? This part isn't in Build-A-Box; it's in A/B testing. On A/B testing,
Easify is also developing a feature to support merchants doing A/B testing, including on the personalization part.
For personalization, which I mentioned at the start — besides supporting live preview, AI can now also support merchants in
A/B testing options. That A/B testing also applies to this box part. The customer can choose option one,
then test it via A/B testing against option two, compare the two, see which is more profit-optimal, and pick the final option.
OK. So let me sum up a bit. AI gives us options,
and sellers can have more choices or decide more easily.
However, to verify the final result, we still need to run A/B tests to get the real numbers from the market —
what customers really want. AI really can't replace the customer at the testing stage — what they like, what they choose, right? Exactly.
I have one more question about the first solution you presented,
if we can go back to it.
The preview — let's call this the AI preview solution, where the customer personalizes and we
let them see their image in advance. Do you have a number measuring how much
conversion increased before versus after applying preview? As for those measurement figures, in today's session
or in this slide, Easify can't yet compile and present them, because first it touches on
analyzing data from customers. If you're interested in the live preview feature and want to see
its real effectiveness, you can contact Easify at the booth or contact me directly —
I can send an email or chat directly to show it. As you can see,
on this slide Easify can't yet compile those figures, and it also involves needing customers' permission.
Thank you very much, Tung Mark. So, we've covered having a product,
then how to have a product customers want to buy — one that looks good so they
decide and buy the most and fastest. Now I have a question for Trang:
I have a product and I can make it convert. As a seller,
I have a pain point: the same product, the same ad creative, at the same time,
put on two stores — one has sales and the other doesn't sell anything.
So from Trang's point of view, could you share
what causes a phenomenon like this,
and what solutions we have to convert as much as possible?
Give me the BettaMax slides, please.
To answer Sammy's question, let me give BettaMax's perspective. Sammy's question is: the same design,
the same product, put on two stores, gives different results, right? First, I think you can't look
at total sales alone, or ROAS alone; you have to go deeper. When running a campaign,
or doing POD, you need to understand each layer of the POD funnel.
I'll lay out basically five layers of the POD funnel. First is ad click. Second is product view.
Third is add-to-cart. Fourth is checkout. And finally, purchase. When a seller runs a campaign,
grasping each layer and reading the metrics of each layer is very important,
because each metric has a different meaning. For example: if a lot of
traffic comes to the site and customers view products a lot, but add-to-cart isn't high,
it means your creative pulled customers into the product, but perhaps how
you tell the story, how you persuade customers, the design or mockups in the store,
or even the pricing, aren't reasonable or good enough — so it won't
convert into add-to-cart. Another example: the customer reached the product page and add-to-cart is high,
but the final conversion into orders is low. This is read differently
to find different problems — for example, shipping price, delivery time,
or not enough payment options for customers to convert into a purchase. A seller understanding such
details while running campaigns is very important. I also want to add some AI-related info.
The information we compile from reading each metric layer used to be quite time-consuming,
and some sellers even had to spend a lot of budget testing
which variable is the right one. Now that we have AI, we can read
these metric layers. AI can intervene and read these data and metric layers more deeply,
giving a more multi-dimensional view to reason about where things are working
and where they aren't. AI can even give assessments and make decisions to help us.
Previously, working as humans, we might have made choices that were a bit emotional.
But BettaMax thinks that with AI, the emotional element decreases a lot.
It may even have little emotion when making decisions to optimize the next steps.
And that's BettaMax's answer: AI helps us, first, save time; second, save effort; and third, save cost —
to know which layer here is broken and not converting well. Trang, could you share
a bit — for each layer, if it breaks, what symptoms usually appear, and how to fix it?
Is there a checklist a seller should fix when each of the five layers breaks?
So, take the first layer, from ad click converting to product view. If the ad-click
metric is low, you can read that the persuasiveness getting customers to click through to
the site is low. Maybe when they watch the ad they have no urge to click,
and your app and the content outside the ad aren't persuasive enough for them to click.
Going deeper into the next layer, product view converting to add-to-cart: here you can, for example, measure a variable — whether
customers stay on the site long enough. For example, if a customer is only on the site for 1 to 2 seconds,
in those 2 seconds they can only scroll to the design, maybe to the title and price of the product page.
Then we know they're breaking off at the point where they look at the design and don't like it,
or the price is too high or unreasonable. Those are metrics I can give as examples to see where customers are breaking off.
Going a bit deeper: if a customer has been on the product page about 10 seconds, or 15 to 20 seconds —
meaning they've read all the way to the last parts of the page — and they still don't add to cart,
that shows the story you tell on the page and your content aren't good or reasonable enough for them to buy.
OK, thank you, Trang. So far we've heard the speakers share a
lot about making the product, making a product that converts,
and optimizing the customer's journey once they're on the landing page and they decide and buy.
All of that is the before-sale stage. What about post-purchase?
How do we deliver a product that matches the customer's expectations so they
increasingly love our brand? May I invite Tracy to share about this?
Thank you for the question, Sammy.
On the production side, Printbelle thinks sellers should optimize their
fulfillment process by using AI to code their own fulfillment system.
Customized to the business's needs.
Then you can connect via API directly with fulfillment providers' systems like Printbelle, so order creation
and operations happen faster, more accurately, and optimize human resources. Next, on applying AI — let
me ask right here: for teams that don't have coding resources, how can AI help them now?
Can they build an operations system that can connect via API directly with suppliers?
Yes. For example, Printbelle's system already has an API ready to connect directly with
a Shopify store, so pushing orders from Shopify to Printbelle is automatic — almost
any order on Shopify can be pushed directly to Printbelle, and from Printbelle's
system it can produce directly based on the volume of orders pushed from Shopify.
Second, on how AI is applied in print-file production: currently the market
trend is shifting gradually from 2D to 3D products, so making 3D print
files is more complex — especially, for example, a very special Q4 product,
ornaments, which have very good metrics. So how is print-file work done?
Right now — please play the video for me.
Currently Printbelle also provides a lot of support in helping sellers make
print files that show the raised/embossed part of the product.
Shifting from 2D to 3D. For example, 3D ornament products currently
have better conversion metrics and a higher selling price than 2D
products with the same design. From Printbelle's customer tests, the trend of
3D products has shown very good signals for Q4,
not only for ornaments. And in the coming Q4, Printbelle plans to help
sellers develop more new products with our proprietary 3D printing technology.
And finally, Printbelle will soon launch an AI support application, 24/7. In the year-end season,
order volumes rise a lot — 5x, 10x versus now — so support
needs will be higher. Printbelle will soon launch this 24/7 support bot; for example,
when a night team places orders overtime and there's no human staff to answer sellers directly,
we'll have this AI support that can check in real time
and answer questions like which production step an order is at, where the tracking is,
or other info about the product or order. That's what Printbelle is trying
to support in the AI era, and we hope sellers will have more time
— once you've optimized time in operations, you can spend more time optimizing sales.
That's my sharing. One question: Q4 is peak season, so many things happen that
could lead to sellers in January having to handle a pile of disputes and complaints.
So Tracy, what should we prepare carefully to avoid the factors that
— after a happy sale season getting money in — could lead to
many complaints afterward? Do you have advice on what everyone should prepare right now?
Yes, I think you should choose a manufacturer with good production capability.
First, the capability to scale in the year-end season. Second, producing products that avoid issues
like breakage or products not matching the mockup, to minimize claims. And third, the shipping capability
of the logistics/fulfillment provider, to meet the timing to arrive in time for the holidays.
Those are the areas where the fulfillment provider can support sellers.
OK. Thank you very much to our four speakers today. From what you shared,
to prepare for Q4 we have to prepare very carefully — starting with how to apply AI in Q4.
For example, starting from product: don't make products by gut feeling, make them with data.
Then we let customers apply AI by letting them preview the product in advance
to help increase the CVR of our products. Then, when tracking ads, we shouldn't just pour
money into ads; we should stop and break down each piece of data — which
funnel stage we drop off at — and go deep into each funnel stage to optimize.
And finally, we must prepare a very good operations plan so customers receive the right
order with the right quality, exactly as they expect. So AI isn't actually about doing well
at any single factor; from my view as a seller, whoever does well must build
an entire AI process linking the start to the end of their product value chain.
You can't expect that the first time, someone or some AI does it better;
in my personal view, AI or automation is just a tool that magnifies formulas you've already done well.
Once you've rolled it out at a very small internal scale and it worked,
then thanks to AI you can scale it up. This lies in the ability to understand your own domain knowledge,
understand the market, and know how to apply AI to your business. So our panel
discussion only has time for today, up to now. Thank you very much to the four speakers,
and now we'll move on to the next Buildathon segment. Thank you all very much.
Thank you so much. And once again, I'd like to send sincere thanks to our speakers as
well as our moderator for this afternoon's panel discussion. Once again, thank you all very much,
and right at this moment let's capture this moment together before moving on to the next part — from
learning to practice. And we'll wrap up those exercises through the part coming up next. Once again,
thank you so much to our moderator and all the guest speakers who have joined us in the last,
final session today. Now we're moving from learning to building, and throughout the Buildathon,
teams have been working with real sellers to identify practical e-commerce challenges and explore how AI can solve them.
원문 자막 펼치기
WEBVTT Kind: captions Language: en 00:00:02.000 --> 00:00:07.499 ...and joining us on stage, we have three speakers bringing different perspectives 00:00:07.500 --> 00:00:12.999 on the personalization ecosystem. I would like to invite to the stage Mr. 00:00:14.000 --> 00:00:20.249 Ethan Nguyen from Pengu, on how design and creative teams leverage 00:00:20.250 --> 00:00:26.499 AI to supercharge ideation. And we would also invite Ms. 00:00:26.500 --> 00:00:32.749 Trang Vu from BettaMax, breaking down operational frameworks and systemic scaling 00:00:32.750 --> 00:00:38.999 to help POD businesses. And we also want to invite Mr. 00:00:40.000 --> 00:00:46.570 Tung Mark from Easify, walking us through how AI powers hyper-personalization, 00:00:46.571 --> 00:00:53.142 connecting product creation all the way to final checkout. And lastly, 00:00:53.143 --> 00:00:59.713 please welcome the beautiful lady, our last speaker, Miss Tracy from Printbelle, 00:00:59.714 --> 00:01:06.285 breaking down Vietnam's POD landscape — the real GMV data, market volume, and brand strategies 00:01:06.286 --> 00:01:12.856 to capture global demand. So ladies and gentlemen, now the time has come. 00:01:12.857 --> 00:01:19.428 Let's explore how POD, personalization, and AI are coming together to 00:01:19.429 --> 00:01:25.999 create new opportunities for ecommerce brands. So now, the stage is yours. 00:01:27.000 --> 00:01:32.000 Thank you, MC. This session is probably the first time since this morning that I'm speaking in Vietnamese. 00:01:33.000 --> 00:01:39.570 Hello everyone, I'm Sammy Ha, a POD seller. I'm very happy that ShopX invited 00:01:39.571 --> 00:01:46.142 me to attend today, and I'm honored to be the moderator for today's panel discussion. 00:01:46.143 --> 00:01:52.713 As a POD seller myself, I also have a lot of what you might 00:01:52.714 --> 00:01:59.285 call burning questions. I'm quite curious about them, and I'd like to invite all 00:01:59.286 --> 00:02:05.856 of today's speakers on this panel to answer my questions, as well as those 00:02:05.857 --> 00:02:12.428 of everyone attending today's event. But first, I'd like to take a moment to let 00:02:12.429 --> 00:02:18.999 our guests introduce themselves and their companies. May I invite Ethan to go first? 00:02:20.000 --> 00:02:21.000 Hello. 00:02:22.000 --> 00:02:27.999 Thank you, Sammy, for the introduction. Hello everyone. I'm Ethan; currently I'm the CTO of Pengu. 00:02:28.000 --> 00:02:33.999 I'm very happy to be joining the ShopX event this year. 00:02:35.000 --> 00:02:39.000 I think... 00:02:40.000 --> 00:02:46.999 Hello Sammy Ha, hello everyone, and hello to everyone sitting below. I'm Trang from BettaMax, 00:02:47.000 --> 00:02:53.999 and BettaMax is an e-commerce platform dedicated specifically to dropshipping and POD. BettaMax is aiming for 00:02:54.000 --> 00:03:00.999 a solution where every seller can run all of their operations on a single platform. 00:03:02.000 --> 00:03:08.749 Hello everyone. Very happy to be at ShopX today. I'm Tung — 00:03:08.750 --> 00:03:15.499 in English you can call me Mark — currently CEO at Easify. 00:03:15.500 --> 00:03:22.249 I have over 10 years of experience providing e-commerce solutions, 00:03:22.250 --> 00:03:28.999 and currently we provide Shopify apps focused on custom products and personalization. 00:03:30.000 --> 00:03:32.000 Hello everyone. I'm Tracy. 00:03:33.000 --> 00:03:39.499 Currently I handle the Vietnam market for Printbelle. Printbelle is a POD 00:03:39.500 --> 00:03:45.999 fulfillment company that supports POD sellers to develop new products and ship worldwide. 00:03:47.000 --> 00:03:52.999 So, my first question is for all four speakers today. I can see there 00:03:53.000 --> 00:03:58.999 are quite a few familiar POD-seller faces down here. This panel will discuss the POD 00:03:59.000 --> 00:04:04.999 industry in some depth. Q4 is very close now — it's a peak season. 00:04:06.000 --> 00:04:12.499 So in your view, what should sellers prepare right now? 00:04:12.500 --> 00:04:18.999 Feel free — whoever wants to go first, please jump in. 00:04:20.000 --> 00:04:25.999 Hi everyone. From Pengu's perspective, sellers should prepare a very solid idea-research process and 00:04:26.000 --> 00:04:31.999 workflow — and that's actually the problem Pengu is aiming to solve. 00:04:32.000 --> 00:04:37.999 Our goal in releasing Pengu is to help sellers find winning ideas faster in the market. 00:04:39.000 --> 00:04:41.000 A winning idea here has two elements. 00:04:42.000 --> 00:04:49.666 First, relying on historical data to know the patterns — for example, what motifs a niche has, what wordplay works. 00:04:49.667 --> 00:04:57.332 Second, detecting very quickly the new trends appearing in the market. That's my point of view on what sellers 00:04:57.333 --> 00:05:04.999 need to prepare, and I think it's also the strength — the unique selling point — of each team. 00:05:06.000 --> 00:05:08.000 Let me answer your question... 00:05:09.000 --> 00:05:10.000 This feels like a beauty pageant. 00:05:11.000 --> 00:05:18.499 I also agree with what Ethan just shared — that we need to prepare research as 00:05:18.500 --> 00:05:25.999 well as samples. From BettaMax's side, I'd add that we also need a good audit process 00:05:26.000 --> 00:05:33.499 so we can test products faster, make quicker assessments of what we've just tested, 00:05:33.500 --> 00:05:40.999 speed up our progress, and find what works so we can scale it during this year's BFCM. 00:05:42.000 --> 00:05:48.249 To answer this question from my side: to prepare for BFCM 00:05:48.250 --> 00:05:54.499 — based on experience working with POD brands on Shopify stores — first, 00:05:54.500 --> 00:06:00.749 you should prepare a detailed BFCM plan, then set it 00:06:00.750 --> 00:06:06.999 up and have it ready to run the moment BFCM launches. 00:06:08.000 --> 00:06:16.499 And from the production side, I think every seller needs to optimize their operations first, then find fulfillment 00:06:16.500 --> 00:06:24.999 partners that are stable and accurate, so they can meet the demand to scale in the year-end season. 00:06:26.000 --> 00:06:32.713 So, based on what our four speakers shared, in Q4 you can't win 00:06:32.714 --> 00:06:39.428 with a single lever alone — it comes from preparing a whole plan, 00:06:39.429 --> 00:06:46.142 including product, operations, ads, and audit. First, let's follow the journey step by step. 00:06:46.143 --> 00:06:52.856 Product is a very important factor — product is king. 00:06:52.857 --> 00:06:59.570 So Ethan, could you share how to create what we call a "winning" product? 00:06:59.571 --> 00:07:06.285 How do we make customers want to buy — create a product 00:07:06.286 --> 00:07:12.999 customers are ready and eager to buy? Could you share your view, Ethan? 00:07:14.000 --> 00:07:20.777 From my perspective, we shouldn't talk about a winning product in vague terms. 00:07:20.778 --> 00:07:27.555 Especially when working with large teams, we need what I call a formula to find a winning product. 00:07:27.556 --> 00:07:34.332 POD, although a new industry, already has historical data in the market. 00:07:34.333 --> 00:07:41.110 So the way our team analyzes a winning product is based on past data. For example, 00:07:41.111 --> 00:07:47.888 at Pengu we crawl several million listings in the market, then analyze, for each niche, 00:07:47.889 --> 00:07:54.666 what motifs the audience tends to like, what wordplay they like, and which copy 00:07:54.667 --> 00:08:01.443 has sold the most. From there we come up with the product idea. Second, 00:08:01.444 --> 00:08:08.221 detecting new trends in the market. When we mix those two factors, we find a 00:08:08.222 --> 00:08:14.999 product with a higher probability of winning. That's my point of view on winning products. 00:08:16.000 --> 00:08:22.000 I'd like to go deeper into finding a winning SKU. 00:08:23.000 --> 00:08:29.599 OK, let me give more detail about Pengu. Pengu is a product focused 00:08:29.600 --> 00:08:36.199 on helping sellers find winning ideas and winning products in the market. 00:08:36.200 --> 00:08:42.799 Every day, Pengu crawls POD from popular platforms — for example Etsy, Shopify, 00:08:42.800 --> 00:08:49.399 Facebook Ads. Then for Etsy we rely on sales and favorites; 00:08:49.400 --> 00:08:55.999 for Facebook we rely on reactions, comments, and shares. From there we extract different patterns. 00:08:57.000 --> 00:09:03.856 Actually, recently, as people use AI, you'll notice the way AI looks at a winning product is very different. 00:09:03.857 --> 00:09:10.713 Sometimes we, as humans, look at a product and get the feeling it will win. 00:09:10.714 --> 00:09:17.570 But for AI, everything has to be put into a formula. That's what Pengu is aiming for. 00:09:17.571 --> 00:09:24.428 When analyzing a product, we aim for AI to be able to evaluate what wordplay it has, 00:09:24.429 --> 00:09:31.285 what typography, what emotional hook, what color, what layout. You can imagine we try to turn 00:09:31.286 --> 00:09:38.142 every listing into a full formula and analyze it over a large volume of data. 00:09:38.143 --> 00:09:44.999 Then we start to know, for a given niche family, which copy and which motifs tend to go together. 00:09:46.000 --> 00:09:53.399 The second step, after having a knowledge base about the market and historical data, 00:09:53.400 --> 00:10:00.799 is to work out how to catch the newest things. Because relying only on historical data would 00:10:00.800 --> 00:10:08.199 be quite outdated, so we need to combine past experience and patterns with the trends 00:10:08.200 --> 00:10:15.599 and new ideas appearing in the market. That's the real-time detection module, and by combining those two, 00:10:15.600 --> 00:10:22.999 we help sellers detect which winning idea is starting to emerge in the market. 00:10:24.000 --> 00:10:30.999 Pengu's goal — and my goal for the Pengu team — is 00:10:31.000 --> 00:10:37.999 not to replace humans, but to help a POD team. 00:10:38.000 --> 00:10:44.999 Instead of a 100-person team where an operator can handle about 4 to 5 brands, 00:10:45.000 --> 00:10:51.999 now you only need a few people combined with AI to 00:10:52.000 --> 00:10:58.999 detect winning ideas very fast and operate a full POD brand. OK. 00:11:00.000 --> 00:11:01.000 I have a question for Ethan. 00:11:02.000 --> 00:11:09.285 Right now everyone knows how to use AI to make products, right? I'm sure no one here doesn't use it. 00:11:09.286 --> 00:11:16.570 So with the same workflow and the same way of thinking, is there a chance 00:11:16.571 --> 00:11:23.856 the outputs will be products that look very similar to each other, and all products in the market 00:11:23.857 --> 00:11:31.142 end up looking the same? So what solutions does Pengu have to create product differentiation so we 00:11:31.143 --> 00:11:38.428 can win customers? OK. Actually, this was a problem we had to think about when we started building the product: 00:11:38.429 --> 00:11:45.713 if we crawl market data and everyone uses that same data, everyone will generate similar 00:11:45.714 --> 00:11:52.999 ideas and products. So to solve that, we worked with many POD teams and noticed the following pattern. 00:11:54.000 --> 00:12:01.428 Although the market data is the same, each POD team has strengths in different niches. 00:12:01.429 --> 00:12:08.856 For example, some teams are very strong at personalized products for best friends, and others are very strong 00:12:08.857 --> 00:12:16.285 at selling farmer niches or gardening niches. So Pengu's goal is, first, to provide market signals 00:12:16.286 --> 00:12:23.713 and insights based on data, AI, and big data; and second, to combine that with each 00:12:23.714 --> 00:12:31.142 seller's own data to generate a new idea that fits what I call the DNA of 00:12:31.143 --> 00:12:38.570 that POD team. That helps make that team's or seller's idea unique, instead of being very common 00:12:38.571 --> 00:12:45.999 in the market and based only on generic data. That's how Pengu is solving it currently. 00:12:47.000 --> 00:12:52.999 OK, one more question. If there are new sellers here just starting out in 00:12:53.000 --> 00:12:58.999 POD who don't have any database to rely on yet, where should they start? 00:13:00.000 --> 00:13:06.832 I think not having data is only the case when you're just starting, when you have nothing yet. 00:13:06.833 --> 00:13:13.666 But at Pengu, for example, we solve that by having people start a store, 00:13:13.667 --> 00:13:20.499 start listing, and run ads — then you'll definitely have traffic metrics. 00:13:20.500 --> 00:13:27.332 You'll know total views, CR, add-to-cart. Based on that data, Pengu will collect and optimize the generation of new ideas. 00:13:27.333 --> 00:13:34.166 So essentially it's a loop: from idea to listing to store, get feedback metrics 00:13:34.167 --> 00:13:40.999 from customers on that listing, then loop back to optimize the idea and the product. 00:13:42.000 --> 00:13:48.999 OK, thank you, Ethan. Ethan shared a lot about how to build a product with the potential to win. 00:13:49.000 --> 00:13:55.999 Now, there are many teams in the market and the volume of products they 00:13:56.000 --> 00:14:02.999 create daily is very high, because thanks to AI, anyone can make that many products now. 00:14:03.000 --> 00:14:09.999 But whether they sell or not — will the number of buyers grow 10x or 100x? 00:14:10.000 --> 00:14:16.999 I'd guess buyers, or our customers, stay about the same, because you can't just create more people. 00:14:17.000 --> 00:14:23.999 So we hit a bottleneck: many products, but how do we convert, how do we increase conversion? 00:14:24.000 --> 00:14:30.999 For this part, may I invite Tung Mark to share his experience on how to improve conversion? 00:14:32.000 --> 00:14:36.000 This question is for Easify, right? 00:14:37.000 --> 00:14:37.999 That's right. 00:14:38.000 --> 00:14:45.000 Yes. Let me reintroduce myself — I'm Tung, Tung Mark, from... 00:14:46.000 --> 00:14:47.000 Sorry, Tung Mark 00:14:48.000 --> 00:14:54.000 Could you give me the Easify slides, please? 00:15:01.000 --> 00:15:07.999 To answer this question, I'd like to continue a point Ethan just made. 00:15:08.000 --> 00:15:14.999 Ethan shared that POD teams are currently using AI to create products more easily and faster. 00:15:15.000 --> 00:15:21.999 However, from experience working with brands that have personalized products, 00:15:22.000 --> 00:15:28.999 I think we should also ask: when AI helps merchants create more products, 00:15:29.000 --> 00:15:35.999 does that really help us sell more, and does it make buying easier for customers? 00:15:37.000 --> 00:15:44.249 So, the advantage of AI is creating products faster. This advantage isn't just for one store; 00:15:44.250 --> 00:15:51.499 right now many stores can also create products and expand their catalog that fast. However, 00:15:51.500 --> 00:15:58.749 more choices don't necessarily make it easier for buyers, because with many choices they have to weigh more, 00:15:58.750 --> 00:16:05.999 compare more, and sometimes during the buying process they feel stressed and overwhelmed when making a decision. 00:16:07.000 --> 00:16:14.166 Often the purchase gets interrupted, and in many cases the buying journey ends entirely. 00:16:14.167 --> 00:16:21.332 So, to make it easier for customers to buy, 00:16:21.333 --> 00:16:28.499 there are two things to note. First, let customers personalize the product themselves 00:16:28.500 --> 00:16:35.666 according to their preferences and purpose. Second, we can apply AI 00:16:35.667 --> 00:16:42.832 to that process to make personalization more intuitive, so customers can clearly 00:16:42.833 --> 00:16:49.999 see their final product and feel more confident when deciding to buy. 00:16:51.000 --> 00:16:57.999 Let me give two concrete examples of personalization when a customer buys a product. 00:16:58.000 --> 00:17:04.999 When a customer wants to buy a product — for example, a T-shirt like this 00:17:05.000 --> 00:17:11.999 — the usual way is: the customer orders first and then sends the photo, 00:17:12.000 --> 00:17:18.999 or many stores have a photo-upload field so the customer can upload before add-to-cart. However, 00:17:19.000 --> 00:17:25.999 the problem is that when the customer uploads the photo, they can't yet picture 00:17:26.000 --> 00:17:32.999 whether that photo suits the shirt or what the final shirt with their image will look like. 00:17:33.000 --> 00:17:39.999 When it's still vague to them, it's hard to decide to buy right away. 00:17:40.000 --> 00:17:46.999 To solve this, there's a solution: live preview. AI-powered live preview helps here. 00:17:47.000 --> 00:17:53.999 When the customer uploads a photo, AI can remove the background or convert the 00:17:54.000 --> 00:18:00.999 photo's style to suit the shirt — for example, grayscale — then show the 00:18:01.000 --> 00:18:07.999 result directly in the live preview. Now the customer sees it's no longer a generic sample product; 00:18:08.000 --> 00:18:14.999 it's their product with their image, almost exactly what they'll receive after purchase. 00:18:15.000 --> 00:18:21.999 That moment of seeing your choice become a real product makes customers more excited 00:18:22.000 --> 00:18:28.999 and makes it easier for them to add to cart and complete the purchase. 00:18:30.000 --> 00:18:36.499 Another example: a customer wants to buy a hat and embroider their pet on it. 00:18:36.500 --> 00:18:42.999 After uploading — they can upload just an ordinary photo from their phone — AI can remove 00:18:43.000 --> 00:18:49.499 the background and convert that photo into an embroidery effect, then display it directly on 00:18:49.500 --> 00:18:55.999 the hat like this. And you can see it has the effect of a real product. 00:18:57.000 --> 00:19:03.332 So in this personalization part, AI doesn't just help the 00:19:03.333 --> 00:19:09.666 merchant create more products; AI also helps customers feel more 00:19:09.667 --> 00:19:15.999 confident and clearly see their product before deciding to buy. 00:19:17.000 --> 00:19:24.099 That's how AI helps sell a personalized product better. However, there's another issue: 00:19:24.100 --> 00:19:31.199 for teams and merchants selling gifting — gift boxes — 00:19:31.200 --> 00:19:38.299 how do we make the buying journey simpler? Take the Father's Day campaign: 00:19:38.300 --> 00:19:45.399 when a customer comes to your store wanting to pick a 00:19:45.400 --> 00:19:52.499 gift box for dad, they already have an idea but still 00:19:52.500 --> 00:19:59.599 don't know exactly what to choose. Normally they'd have to go into one collection, 00:19:59.600 --> 00:20:06.699 pick a shirt, go back, pick a jacket, go back, 00:20:06.700 --> 00:20:13.799 pick other accessories. In this process they visit many pages and sometimes 00:20:13.800 --> 00:20:20.899 can't remember what they've chosen, whether the items go together, 00:20:20.900 --> 00:20:27.999 or whether they've picked enough. It's very hard. Our catalog doesn't lack products. 00:20:29.000 --> 00:20:35.666 However, these choices are scattered across many pages. The solution — you 00:20:35.667 --> 00:20:42.332 may already know it — is Build-A-Box. This Build-A-Box solution brings all those 00:20:42.333 --> 00:20:48.999 scattered choices into one journey, making it easier for customers to buy 00:20:49.000 --> 00:20:55.666 and complete their gift box. For example, in this Build step: first, 00:20:55.667 --> 00:21:02.332 the customer is on a single page. As the first step, 00:21:02.333 --> 00:21:08.999 the customer can choose a T-shirt, polo, or hoodie; then choose outerwear or accessories. 00:21:10.000 --> 00:21:16.916 And the final step can add a message or note in the card. 00:21:16.917 --> 00:21:23.832 During this process, the customer can see the products they've 00:21:23.833 --> 00:21:30.749 chosen in the summary box, and see the total and prices updated live, 00:21:30.750 --> 00:21:37.666 instantly. So Build-A-Box gives customers an easier shopping journey and clearer visibility. 00:21:37.667 --> 00:21:44.582 As for what AI does here: to let the merchant 00:21:44.583 --> 00:21:51.499 create an experience like this, AI can help the merchant start faster. 00:21:51.500 --> 00:21:58.416 AI can scan the store's catalog and suggest which product 00:21:58.417 --> 00:22:05.332 sets can be combined. It can also suggest how to set up 00:22:05.333 --> 00:22:12.249 discounts or offers for which product sets. And once set up, 00:22:12.250 --> 00:22:19.166 AI can help set up a Build-A-Box page instantly. Beyond that, 00:22:19.167 --> 00:22:26.082 we can use AI to scan whether our Build-A-Box page is good enough, 00:22:26.083 --> 00:22:32.999 or needs improvement to increase CVR or AOV, before launching it. 00:22:34.000 --> 00:22:44.000 So you can see AI can plan a box and build it right away. 00:22:45.000 --> 00:22:51.666 So in this Build-A-Box part, AI can help the merchant set 00:22:51.667 --> 00:22:58.332 up a shopping journey, make the journey easier, make the customer experience smoother, 00:22:58.333 --> 00:23:04.999 and the merchant can also increase AOV more naturally. 00:23:06.000 --> 00:23:16.000 Those are two solutions — one for a personalized product and one for increasing AOV via Build-A-Box — where merchants can apply AI. 00:23:17.000 --> 00:23:23.399 Let me ask a bit. For this Build-Your-Box solution, you said its 00:23:23.400 --> 00:23:29.799 function is to suggest to the seller which set to combine with which, 00:23:29.800 --> 00:23:36.199 and what discount to use. Could you tell me the philosophy behind 00:23:36.200 --> 00:23:42.599 what to combine with what, and what discount to use, 00:23:42.600 --> 00:23:48.999 so I get a set that sells the most and gives the most profit? 00:23:50.000 --> 00:23:56.999 What if the discount is too big and I lose money — maybe I don't need any discount and customers still buy? 00:23:57.000 --> 00:24:03.999 And is combining these the right call? How do I get data feedback so I know whether combining set 00:24:04.000 --> 00:24:10.999 1 with set 2 is right, or set 1 with set 3 is right? What options does Easify have for this? 00:24:12.000 --> 00:24:14.000 Thank you. Great question. 00:24:15.000 --> 00:24:22.499 In the Build-A-Box that Easify is building, the final decision is still the merchant's. 00:24:22.500 --> 00:24:29.999 As for the suggestion mechanism, AI asks the merchant the most basic questions. 00:24:31.000 --> 00:24:37.666 First, what's the product category? Second, allow it to scan the store's entire catalog. Then, 00:24:37.667 --> 00:24:44.332 based on training from the user's side, it can suggest which collections can be combined and 00:24:44.333 --> 00:24:50.999 give the merchant options. The merchant can then decide on option one, two, or three. 00:24:51.000 --> 00:24:57.666 From the merchant's answer, it continues to the offer part. For offers, based on the 00:24:57.667 --> 00:25:04.332 customer's choices and the catalog, the merchant can rely on the options AI lays out — 00:25:04.333 --> 00:25:10.999 and on their own customers and profit margin — to decide which offer is suitable. 00:25:12.000 --> 00:25:21.000 So AI provides options for the merchant to choose from. Instead of having no option in mind, AI provides one. 00:25:22.000 --> 00:25:28.999 So AI won't give the exact answer, but provides options, and finally the merchant decides, 00:25:29.000 --> 00:25:35.999 right? Now my second point: how do I measure whether the choice I made earlier was correct? 00:25:36.000 --> 00:25:42.999 Because Easify is only providing options — AI is helping the seller with options, 00:25:43.000 --> 00:25:49.999 not the decision. OK, so now I pick among four options; I close my eyes 00:25:50.000 --> 00:25:56.999 and pick option A. How do I know if it's good, so next time I 00:25:57.000 --> 00:26:03.999 keep choosing A, or if it's not good, so I stop and pick B, C, 00:26:04.000 --> 00:26:10.999 or D? This part isn't in Build-A-Box; it's in A/B testing. On A/B testing, 00:26:11.000 --> 00:26:17.999 Easify is also developing a feature to support merchants doing A/B testing, including on the personalization part. 00:26:19.000 --> 00:26:26.332 For personalization, which I mentioned at the start — besides supporting live preview, AI can now also support merchants in 00:26:26.333 --> 00:26:33.666 A/B testing options. That A/B testing also applies to this box part. The customer can choose option one, 00:26:33.667 --> 00:26:40.999 then test it via A/B testing against option two, compare the two, see which is more profit-optimal, and pick the final option. 00:26:42.000 --> 00:26:47.499 OK. So let me sum up a bit. AI gives us options, 00:26:47.500 --> 00:26:52.999 and sellers can have more choices or decide more easily. 00:26:54.000 --> 00:27:01.999 However, to verify the final result, we still need to run A/B tests to get the real numbers from the market — 00:27:02.000 --> 00:27:09.999 what customers really want. AI really can't replace the customer at the testing stage — what they like, what they choose, right? Exactly. 00:27:11.000 --> 00:27:16.999 I have one more question about the first solution you presented, 00:27:17.000 --> 00:27:22.999 if we can go back to it. 00:27:24.000 --> 00:27:30.874 The preview — let's call this the AI preview solution, where the customer personalizes and we 00:27:30.875 --> 00:27:37.749 let them see their image in advance. Do you have a number measuring how much 00:27:37.750 --> 00:27:44.624 conversion increased before versus after applying preview? As for those measurement figures, in today's session 00:27:44.625 --> 00:27:51.499 or in this slide, Easify can't yet compile and present them, because first it touches on 00:27:51.500 --> 00:27:58.374 analyzing data from customers. If you're interested in the live preview feature and want to see 00:27:58.375 --> 00:28:05.249 its real effectiveness, you can contact Easify at the booth or contact me directly — 00:28:05.250 --> 00:28:12.124 I can send an email or chat directly to show it. As you can see, 00:28:12.125 --> 00:28:18.999 on this slide Easify can't yet compile those figures, and it also involves needing customers' permission. 00:28:20.000 --> 00:28:26.999 Thank you very much, Tung Mark. So, we've covered having a product, 00:28:27.000 --> 00:28:33.999 then how to have a product customers want to buy — one that looks good so they 00:28:34.000 --> 00:28:40.999 decide and buy the most and fastest. Now I have a question for Trang: 00:28:41.000 --> 00:28:47.999 I have a product and I can make it convert. As a seller, 00:28:48.000 --> 00:28:54.999 I have a pain point: the same product, the same ad creative, at the same time, 00:28:55.000 --> 00:29:01.999 put on two stores — one has sales and the other doesn't sell anything. 00:29:03.000 --> 00:29:10.332 So from Trang's point of view, could you share 00:29:10.333 --> 00:29:17.666 what causes a phenomenon like this, 00:29:17.667 --> 00:29:24.999 and what solutions we have to convert as much as possible? 00:29:26.000 --> 00:29:34.000 Give me the BettaMax slides, please. 00:29:35.000 --> 00:29:42.399 To answer Sammy's question, let me give BettaMax's perspective. Sammy's question is: the same design, 00:29:42.400 --> 00:29:49.799 the same product, put on two stores, gives different results, right? First, I think you can't look 00:29:49.800 --> 00:29:57.199 at total sales alone, or ROAS alone; you have to go deeper. When running a campaign, 00:29:57.200 --> 00:30:04.599 or doing POD, you need to understand each layer of the POD funnel. 00:30:04.600 --> 00:30:11.999 I'll lay out basically five layers of the POD funnel. First is ad click. Second is product view. 00:30:13.000 --> 00:30:19.999 Third is add-to-cart. Fourth is checkout. And finally, purchase. When a seller runs a campaign, 00:30:20.000 --> 00:30:26.999 grasping each layer and reading the metrics of each layer is very important, 00:30:27.000 --> 00:30:33.999 because each metric has a different meaning. For example: if a lot of 00:30:34.000 --> 00:30:40.999 traffic comes to the site and customers view products a lot, but add-to-cart isn't high, 00:30:41.000 --> 00:30:47.999 it means your creative pulled customers into the product, but perhaps how 00:30:48.000 --> 00:30:54.999 you tell the story, how you persuade customers, the design or mockups in the store, 00:30:55.000 --> 00:31:01.999 or even the pricing, aren't reasonable or good enough — so it won't 00:31:02.000 --> 00:31:08.999 convert into add-to-cart. Another example: the customer reached the product page and add-to-cart is high, 00:31:09.000 --> 00:31:15.999 but the final conversion into orders is low. This is read differently 00:31:16.000 --> 00:31:22.999 to find different problems — for example, shipping price, delivery time, 00:31:23.000 --> 00:31:29.999 or not enough payment options for customers to convert into a purchase. A seller understanding such 00:31:30.000 --> 00:31:36.999 details while running campaigns is very important. I also want to add some AI-related info. 00:31:37.000 --> 00:31:43.999 The information we compile from reading each metric layer used to be quite time-consuming, 00:31:44.000 --> 00:31:50.999 and some sellers even had to spend a lot of budget testing 00:31:51.000 --> 00:31:57.999 which variable is the right one. Now that we have AI, we can read 00:31:58.000 --> 00:32:04.999 these metric layers. AI can intervene and read these data and metric layers more deeply, 00:32:05.000 --> 00:32:11.999 giving a more multi-dimensional view to reason about where things are working 00:32:12.000 --> 00:32:18.999 and where they aren't. AI can even give assessments and make decisions to help us. 00:32:19.000 --> 00:32:25.999 Previously, working as humans, we might have made choices that were a bit emotional. 00:32:26.000 --> 00:32:32.999 But BettaMax thinks that with AI, the emotional element decreases a lot. 00:32:34.000 --> 00:32:41.332 It may even have little emotion when making decisions to optimize the next steps. 00:32:41.333 --> 00:32:48.666 And that's BettaMax's answer: AI helps us, first, save time; second, save effort; and third, save cost — 00:32:48.667 --> 00:32:55.999 to know which layer here is broken and not converting well. Trang, could you share 00:32:56.000 --> 00:33:03.332 a bit — for each layer, if it breaks, what symptoms usually appear, and how to fix it? 00:33:03.333 --> 00:33:10.666 Is there a checklist a seller should fix when each of the five layers breaks? 00:33:10.667 --> 00:33:17.999 So, take the first layer, from ad click converting to product view. If the ad-click 00:33:18.000 --> 00:33:25.332 metric is low, you can read that the persuasiveness getting customers to click through to 00:33:25.333 --> 00:33:32.666 the site is low. Maybe when they watch the ad they have no urge to click, 00:33:32.667 --> 00:33:39.999 and your app and the content outside the ad aren't persuasive enough for them to click. 00:33:41.000 --> 00:33:48.124 Going deeper into the next layer, product view converting to add-to-cart: here you can, for example, measure a variable — whether 00:33:48.125 --> 00:33:55.249 customers stay on the site long enough. For example, if a customer is only on the site for 1 to 2 seconds, 00:33:55.250 --> 00:34:02.374 in those 2 seconds they can only scroll to the design, maybe to the title and price of the product page. 00:34:02.375 --> 00:34:09.499 Then we know they're breaking off at the point where they look at the design and don't like it, 00:34:09.500 --> 00:34:16.624 or the price is too high or unreasonable. Those are metrics I can give as examples to see where customers are breaking off. 00:34:16.625 --> 00:34:23.749 Going a bit deeper: if a customer has been on the product page about 10 seconds, or 15 to 20 seconds — 00:34:23.750 --> 00:34:30.874 meaning they've read all the way to the last parts of the page — and they still don't add to cart, 00:34:30.875 --> 00:34:37.999 that shows the story you tell on the page and your content aren't good or reasonable enough for them to buy. 00:34:39.000 --> 00:34:45.999 OK, thank you, Trang. So far we've heard the speakers share a 00:34:46.000 --> 00:34:52.999 lot about making the product, making a product that converts, 00:34:53.000 --> 00:34:59.999 and optimizing the customer's journey once they're on the landing page and they decide and buy. 00:35:00.000 --> 00:35:06.999 All of that is the before-sale stage. What about post-purchase? 00:35:07.000 --> 00:35:13.999 How do we deliver a product that matches the customer's expectations so they 00:35:14.000 --> 00:35:20.999 increasingly love our brand? May I invite Tracy to share about this? 00:35:22.000 --> 00:35:26.000 Thank you for the question, Sammy. 00:35:27.000 --> 00:35:33.999 On the production side, Printbelle thinks sellers should optimize their 00:35:34.000 --> 00:35:40.999 fulfillment process by using AI to code their own fulfillment system. 00:35:42.000 --> 00:35:43.000 Customized to the business's needs. 00:35:44.000 --> 00:35:51.749 Then you can connect via API directly with fulfillment providers' systems like Printbelle, so order creation 00:35:51.750 --> 00:35:59.499 and operations happen faster, more accurately, and optimize human resources. Next, on applying AI — let 00:35:59.500 --> 00:36:07.249 me ask right here: for teams that don't have coding resources, how can AI help them now? 00:36:07.250 --> 00:36:14.999 Can they build an operations system that can connect via API directly with suppliers? 00:36:16.000 --> 00:36:22.499 Yes. For example, Printbelle's system already has an API ready to connect directly with 00:36:22.500 --> 00:36:28.999 a Shopify store, so pushing orders from Shopify to Printbelle is automatic — almost 00:36:29.000 --> 00:36:35.499 any order on Shopify can be pushed directly to Printbelle, and from Printbelle's 00:36:35.500 --> 00:36:41.999 system it can produce directly based on the volume of orders pushed from Shopify. 00:36:43.000 --> 00:36:49.999 Second, on how AI is applied in print-file production: currently the market 00:36:50.000 --> 00:36:56.999 trend is shifting gradually from 2D to 3D products, so making 3D print 00:36:57.000 --> 00:37:03.999 files is more complex — especially, for example, a very special Q4 product, 00:37:04.000 --> 00:37:10.999 ornaments, which have very good metrics. So how is print-file work done? 00:37:12.000 --> 00:37:17.999 Right now — please play the video for me. 00:37:18.000 --> 00:37:23.999 Currently Printbelle also provides a lot of support in helping sellers make 00:37:24.000 --> 00:37:29.999 print files that show the raised/embossed part of the product. 00:37:31.000 --> 00:37:37.499 Shifting from 2D to 3D. For example, 3D ornament products currently 00:37:37.500 --> 00:37:43.999 have better conversion metrics and a higher selling price than 2D 00:37:44.000 --> 00:37:50.499 products with the same design. From Printbelle's customer tests, the trend of 00:37:50.500 --> 00:37:56.999 3D products has shown very good signals for Q4, 00:37:57.000 --> 00:38:03.499 not only for ornaments. And in the coming Q4, Printbelle plans to help 00:38:03.500 --> 00:38:09.999 sellers develop more new products with our proprietary 3D printing technology. 00:38:11.000 --> 00:38:18.332 And finally, Printbelle will soon launch an AI support application, 24/7. In the year-end season, 00:38:18.333 --> 00:38:25.666 order volumes rise a lot — 5x, 10x versus now — so support 00:38:25.667 --> 00:38:32.999 needs will be higher. Printbelle will soon launch this 24/7 support bot; for example, 00:38:33.000 --> 00:38:40.332 when a night team places orders overtime and there's no human staff to answer sellers directly, 00:38:40.333 --> 00:38:47.666 we'll have this AI support that can check in real time 00:38:47.667 --> 00:38:54.999 and answer questions like which production step an order is at, where the tracking is, 00:38:55.000 --> 00:39:02.332 or other info about the product or order. That's what Printbelle is trying 00:39:02.333 --> 00:39:09.666 to support in the AI era, and we hope sellers will have more time 00:39:09.667 --> 00:39:16.999 — once you've optimized time in operations, you can spend more time optimizing sales. 00:39:18.000 --> 00:39:25.599 That's my sharing. One question: Q4 is peak season, so many things happen that 00:39:25.600 --> 00:39:33.199 could lead to sellers in January having to handle a pile of disputes and complaints. 00:39:33.200 --> 00:39:40.799 So Tracy, what should we prepare carefully to avoid the factors that 00:39:40.800 --> 00:39:48.399 — after a happy sale season getting money in — could lead to 00:39:48.400 --> 00:39:55.999 many complaints afterward? Do you have advice on what everyone should prepare right now? 00:39:57.000 --> 00:40:01.000 Yes, I think you should choose a manufacturer with good production capability. 00:40:02.000 --> 00:40:08.332 First, the capability to scale in the year-end season. Second, producing products that avoid issues 00:40:08.333 --> 00:40:14.666 like breakage or products not matching the mockup, to minimize claims. And third, the shipping capability 00:40:14.667 --> 00:40:20.999 of the logistics/fulfillment provider, to meet the timing to arrive in time for the holidays. 00:40:22.000 --> 00:40:27.000 Those are the areas where the fulfillment provider can support sellers. 00:40:28.000 --> 00:40:35.153 OK. Thank you very much to our four speakers today. From what you shared, 00:40:35.154 --> 00:40:42.307 to prepare for Q4 we have to prepare very carefully — starting with how to apply AI in Q4. 00:40:42.308 --> 00:40:49.461 For example, starting from product: don't make products by gut feeling, make them with data. 00:40:49.462 --> 00:40:56.614 Then we let customers apply AI by letting them preview the product in advance 00:40:56.615 --> 00:41:03.768 to help increase the CVR of our products. Then, when tracking ads, we shouldn't just pour 00:41:03.769 --> 00:41:10.922 money into ads; we should stop and break down each piece of data — which 00:41:10.923 --> 00:41:18.076 funnel stage we drop off at — and go deep into each funnel stage to optimize. 00:41:18.077 --> 00:41:25.230 And finally, we must prepare a very good operations plan so customers receive the right 00:41:25.231 --> 00:41:32.384 order with the right quality, exactly as they expect. So AI isn't actually about doing well 00:41:32.385 --> 00:41:39.537 at any single factor; from my view as a seller, whoever does well must build 00:41:39.538 --> 00:41:46.691 an entire AI process linking the start to the end of their product value chain. 00:41:46.692 --> 00:41:53.845 You can't expect that the first time, someone or some AI does it better; 00:41:53.846 --> 00:42:00.999 in my personal view, AI or automation is just a tool that magnifies formulas you've already done well. 00:42:02.000 --> 00:42:08.799 Once you've rolled it out at a very small internal scale and it worked, 00:42:08.800 --> 00:42:15.599 then thanks to AI you can scale it up. This lies in the ability to understand your own domain knowledge, 00:42:15.600 --> 00:42:22.399 understand the market, and know how to apply AI to your business. So our panel 00:42:22.400 --> 00:42:29.199 discussion only has time for today, up to now. Thank you very much to the four speakers, 00:42:29.200 --> 00:42:35.999 and now we'll move on to the next Buildathon segment. Thank you all very much. 00:42:37.000 --> 00:42:43.713 Thank you so much. And once again, I'd like to send sincere thanks to our speakers as 00:42:43.714 --> 00:42:50.428 well as our moderator for this afternoon's panel discussion. Once again, thank you all very much, 00:42:50.429 --> 00:42:57.142 and right at this moment let's capture this moment together before moving on to the next part — from 00:42:57.143 --> 00:43:03.856 learning to practice. And we'll wrap up those exercises through the part coming up next. Once again, 00:43:03.857 --> 00:43:10.570 thank you so much to our moderator and all the guest speakers who have joined us in the last, 00:43:10.571 --> 00:43:17.285 final session today. Now we're moving from learning to building, and throughout the Buildathon, 00:43:17.286 --> 00:43:23.999 teams have been working with real sellers to identify practical e-commerce challenges and explore how AI can solve them.