Brainstorm AI Singapore 2024: The Future Of E-Commerce
Ewa SZYMANSKA, Global Head, Rakuten Institute of Technology (RIT); Executive Officer, Rakuten Group Moderator: Clay CHANDLER, FORTUNE
Category
🤖
TechTranscript
00:00So Yua, thank you so much for joining us today.
00:02We've had a lot of discussion about Southeast Asia,
00:06Southeast Asian firms and policies.
00:08So it's refreshing to be able to get a chance to have a little bit
00:11of a discussion about a Japanese company.
00:14Show of hands, how many know who, what Rakuten is?
00:19Everyone?
00:20Pretty good.
00:21All right.
00:23So why don't we just start there and you can tell us a little bit
00:24about Rakuten and maybe about the Rakuten Institute
00:29of Technology, but this is a company that's
00:32into all kinds of businesses, right?
00:33Right, right, right, right.
00:34So starting from the beginning, Rakuten was founded in 1997.
00:39So it's a time of economic downturn in Japan.
00:44Many businesses are struggling, shutting down.
00:48Small businesses need help and this is
00:51when Rakuten Online Marketplace opens up new markets to small
00:57and medium merchants, to rice sellers and sake brewers
01:01that before could only sell in their cities, in their regions.
01:06Now the whole Japan opens up to them.
01:09Now fast forward to today, we have 57,000 merchants
01:15on our platform and e-commerce still at the very heart of Rakuten,
01:22but since then we expanded and we are now in fintech,
01:26in digital content, in telco.
01:30So we have over 70 businesses overall.
01:33You can think of if you want to do travel booking or hair salon
01:40or horseback racing, there's just a wide variety of payments,
01:48car, credit card, insurance, you know, I can go on.
01:52There's 70, right?
01:53Sports also, yeah.
01:55Overall we have over 100 million users in Japan.
02:00So now most of people in Japan are involved with at least one
02:06of our businesses with a cross usage of over 76%.
02:11And 40 million users a month engaging with Rakuten
02:18which is basically a third of Japan's population.
02:22So it is very much a household name in Japan.
02:26So you're based here in Singapore.
02:28Tell us about the Rakuten Technology Institute and what that does.
02:31Right, right, right.
02:33So Rakuten Institute
02:34of Technology is Rakuten's research and development center.
02:38And our mission is to bridge the science with business applications
02:42and ensure that Rakuten benefits
02:44from this tremendous advance in AI.
02:49So we work with computer vision, language, our customer behavior
02:54from our 70 businesses.
02:57And we do both work traditional AI as well as now generative AI.
03:03Fascinating.
03:04And your background is that you're an expert in the science
03:07of human decision making, right?
03:09Right, right, right.
03:10So before artificial intelligence I was studying human intelligence.
03:13That's a good combination.
03:16You make an interesting point about Japan which is that although,
03:21you know, I mean we rank its giant companies in the Fortune 500 list
03:26and they're probably globally well known that Japanese,
03:30Japan's economy overall is very heavily dependent on small
03:33and medium sized enterprises who are often not as technically savvy
03:37or sophisticated as Sony or Toyota or something like that.
03:43And these are kind of the bread and butter customers for Rakuten.
03:48Tell us a little bit about their level of kind
03:51of technological sophistication and what, how we should think
03:55about how tech savvy they are.
03:58Sure. So we're committed to serving all our partners so regardless
04:06of their technological proficiency.
04:08So we make sure that they have the know-how and skills
04:13to make the most out of our platform.
04:16You know, when we started most
04:19of these sellers actually didn't know much about internet at all.
04:23I mean we're talking about late 90s.
04:26So very early on we established this function
04:29of the e-commerce consultant.
04:32So that's a person that works very closely with the merchant
04:39and assists them with, you know, whatever it is that they need.
04:43It might be help with setting up a storefront,
04:47running a marketing campaign, educating them about, you know,
04:50search engine optimization, what are the social media out there,
04:53what is Facebook, what's Instagram
04:56and how to take advantage of those.
04:58So very much this partnership between the consultants
05:04and the merchants is at the heart of Rakuten Marketplace.
05:09So a lot of these folks would maybe be, you know, small merchants,
05:13maybe live in a kind of rural area, might be older,
05:18kind of mom and pop type businesses.
05:20So that's very intense to reach out directly to those people.
05:27So how big is the kind of force of these consultants?
05:32Right. So we have our regional offices all over Japan
05:35and we have e-commerce consultants in all
05:38of those regional offices making visits to merchants,
05:41sitting at the table with them and working out the strategy,
05:45you know, for sales for next year, for, you know,
05:49Valentine's campaigns, et cetera.
05:52And so do you see that AI will make it easier
05:56or actually more complicated for those kinds of merchants to operate?
06:01Yeah, no, that's a great question.
06:02I mean, we're extremely positive about this technology, about AI.
06:08I mean, Rakuten was founded on this belief
06:11that technology brings people closer together.
06:15Right, right.
06:16You know, even internet, our online service people can think,
06:19you know, online means impersonal, it's convenience,
06:23it's like a vending machine.
06:25Whereas Rakuten have the very different approach, right?
06:28And from the very beginning, it was all about allowing merchants
06:31to express their personality of their business, tell their story
06:36of the product, of the family.
06:38So with generative AI, we see a lot of opportunity
06:43to help merchants, you know, do this even better.
06:48So, you know, a lot of them, as you pointed out, are small merchants.
06:53They may not have marketing teams or designers on their team
06:57and getting in their hands tools that allow them to, you know,
07:01create more beautiful storefronts,
07:04tell their story in a more visually compelling way.
07:07It's something that we're really excited about.
07:10Another area is just to give them ways to connect
07:13with their customers better, allow for personalization,
07:18help them to craft responses to the customers, which, of course,
07:23they have full, you know, control over the final output.
07:27We believe that this will help them be, you know,
07:30even closer to their end customer.
07:34Which kind of plays also on this idea of omotenashi,
07:39and we discussed this at Clay earlier, which is very central
07:43to Japanese culture, and it goes back to the tea ceremony,
07:50and this is this mindfulness and the principle of really care
07:57and hospitality, understanding the needs of your guest,
08:03of your customer, and then going above and beyond to, you know,
08:07meet and exceed those expectations.
08:10So we believe those tools will now allow our merchants
08:14to practice that omotenashi.
08:15Right, right.
08:17Very Japanese kind of idea.
08:19Let me ask you just a quick question about language and culture
08:23and how it combines with AI.
08:24I mean, obviously, you both develop in-house RockGen's own large
08:29language models in GenAI, and you're also working with companies
08:33like OpenAI and others to collaborate in rolling things out.
08:38So you kind of pick and choose as to when and where
08:41to make use of which one.
08:43Can you give us a Japanese perspective on this question
08:48that we discussed earlier this morning about whether
08:52or not large language models that are trained mostly in English
08:56can actually be very responsive and useful in different cultures,
09:00you know, like Japan, Japanese language,
09:02where the culture is so specific?
09:03Right. Very good question.
09:05A few months ago, I would give a different answer.
09:09Today, my answer is that these models are pretty good.
09:14Yeah, yeah.
09:15They are not as good as they are in English.
09:19There is still certain cultural differences,
09:23and I think the panel discussed this before as well
09:26for other Asian languages, the politeness levels.
09:29Yes.
09:30There is some kanji might not be correct, but it is close.
09:37Having said that, it's still a little bit more effort
09:40for the Japanese user to utilize that,
09:44and their reasoning skills are not as good as they are in English.
09:50But yeah, it's actually pretty impressive.
09:53So an interesting perspective.
09:54I think it's a question about which there's a lot
09:56of different opinions and points of view,
10:00especially the more advanced the technology.
10:03Right, right.
10:04But I would like to note something, however,
10:06and that there are certain.
10:08So tokenization for Japanese was not working so well.
10:15And basically, tokenization is that when you go from text
10:20to numeric encoding, you need to kind of parse these text input
10:27into tokens that can be as long as a word
10:30or a subword or a character.
10:32And basically, these commercial large language models
10:36that have been available, they were very inefficient in the way
10:40that they tokenized Japanese language.
10:44And they would use multiple tokens for a single kanji,
10:48for example, which was very inefficient, very costly.
10:51So what we put effort into when we developed
10:55and open sourced our Rakuten LLM 7 billion parameter model,
11:02we put effort into developing a tokenizer
11:04that is much more efficient and, you know, is able to take
11:10in more context into and be just more,
11:17less computationally intense.
11:21So that was kind of the contribution that we made.
11:24And that was something that had to,
11:26that was very language specific, right?
11:29So, you know, my view is that, you know, we are very close,
11:36but certain things have to be solved in order to close the gap.
11:42So.
11:43Yeah.
11:45That's all the time we've got.
11:46But fascinating Japanese perspective, Iwa.
11:47Thank you so much.
11:50Pleasure.