A few years ago, shiny new startups were using their marketing dollars to tell the world that chatbots were going to change everything.

Those marketing dollars have now been spent and most of those startups are no more. But for the past few years, one company has been quietly making chatbots useful, and they are now starting to make some noise.

Today we sit down with Akemi Tsunagawa, founder of Bespoke and creator of the Bebot chatbot.

In several important ways, Bespoke is one of the most successful chatbot companies in the world, and you’ll be hearing a lot about them in the years to come.

Today, however, Akemi explains how she and the team managed to succeed where so many better-funded companies failed, and she gives some great advice about how to get consumers to try out new technologies. We also talk about why you should absolutely never build your business around Facebook or WeChat.

It’s a great conversation, and I think you’ll enjoy it.

Show Notes

  • Why most travel websites are doomed to failure
  • Founding a technical startup without technical co-founders
  • How to get people to tell chatbot what they really think
  • Where chatbots excel and where they should not try
  • Things you should never use a chatbot for
  • Why you should not build a chatbot on Facebook or WeChat
  • Why Japanese don’t want to use chatbots
  • Bespoke’s plans to go global
  • How to speed up decision making inside Japanese companies

Links from the Founder

Leave a comment


Welcome to Disrupting Japan, straight talk from Japan’s most successful entrepreneurs.

I’m Tim Romero and thanks for joining me.

I think the peak of the chat bot hype cycle came in 2017. If we cast our minds back into the midst of that distant age, perhaps we can recall that chatbots were going to change the way we work, the way we shop, the way we bank, the way we talk to our customers, and even the way we find love and raise our children. Yeah, that didn’t happen, and startup founders, start of investors, and start of media all moved on to focus on some newer, shinier object – blockchain, probably.

But, you know something, sometimes, all that media type and investor attention can actually make it really hard to build something worthwhile. A lot of times, the best ideas and the best use of technology come from trying to solve a simple problem without investors telling you you need to be a unicorn or a journalist demanding to know exactly how you plan on changing the world by the end of the year, and so it is with chat bots.

Today, we’re going to sit down with Akemi Tsunagawa, the founder and CEO of Bespoke, the creator of the Bebot chatbot. Now, Akemi will tell you exactly how Bebot works in just a second, but to really appreciate what that important story Bespoke is, you need to understand that outside of marketing and some trivial customer support apps, you’ve got to realize, there is almost no chatbot success stories. Bebot is one of the very few chatbots in the entire world that provides enough genuine utility that people not only willingly interact with it but start to rely on it.

Bespoke’s business model does not rely on novelty or cost-cutting, no. Bespoke is solving an actual problem. This is a great example of how the needs of one industry can push technology forward for other sectors, and Akemi and I also talk about why she didn’t even realize they were running a chatbot company at first.

She gives some great advice on how to get consumers to try out, not just chatbots but any new technology, and we chat about why you should never – and I mean never — build your business around Facebook or WeChat.

But you know, Akemi tells that story much better than I can, so let’s get right into the interview.


Tim: So, I’m sitting here with Akemi Tsunagawa, the founder of Bespoke, and we’re going to be chatting about chatbots today, so thanks for sitting down with me.

Akemi: Well, thanks for inviting me here.

Tim: You know, there are so many chatbot startups out there right now, but this is really an interesting application for it, so you can you just briefly explain about the service, about what Bebot is?

Akemi: Sure. We make chatbots for enterprise client in the hospitality space like hotels, airports, train stations to help them automate interaction with their visitors and guests, to provide better experiences.

Tim: Oh, so you are focused on tourism and travel, and I’ve noticed you managed to get some major deals – Bebot is being used not only at the airport, in the New Otani Hotel, Tokyo Station. Before we dive into Bebot and the business in general, I want to back up a little bit and talk about you.

Akemi: Sure.

Tim: So, you started this in 2015, so why chatbots?

Akemi: Okay, so to be honest, we never started this company as a chatbot company. We were doing something completely different three and half years ago. What I always wanted to do was to provide authentic experiences to people. The type of experiences you get through a local friend is very different; it’s very unique. You get to eat something special, something unique to the country, or you get to visit places that you can’t visit elsewhere. That’s the kind of experience that I was looking for, but I could not find any sort of services out there.

Tim: Oh, I can understand it because pretty much 100% of the travel-oriented startups has that same pitch, right? That it is like having a local friend, but it is something that is really hard to do at scale. I’m curious, you mentioned at first you are not doing chatbots. What were you doing at first?

Akemi: So, it was a website to connect to other people, like you’ve seen that many ways at many places, but in addition to just connecting locals to travelers, you are also providing like, a little bit similar to like a trip advisor, but like a local version, like little local bars that you would not find in guidebooks types of places, so we launched the service first. We did lots of user interviews, and as we met more users, like tourists, they kept telling us like, the same kind of problem which is like in Japan, it’s always about the language, it’s always about communications, like I call the restaurant, they don’t take my reservations because I’m not Japanese or they don’t understand me, or the website is all in Japanese and I can’t understand. So, we decided to add a concert featured on top of the existing website, then as we did more user interviews, people kept telling us, concierge is great, but I don’t need your help like, next week, I need your help like, right now because I’m lost at a train station or I’m having a problem at the restaurant, so it needed to be instant. So, we decided to turn our concert service into chat.

Tim: You know, that’s interesting because I think that the vast majority of tourism startups, and even tourism technology companies approach it as a problem of lack of information that these travelers just don’t know about the local spots where the interesting things to do, but it sounds like you discovered it was really more a lack of communication ability.

Akemi: I think so, especially in Japan. I mean, there is not enough information out there. I actually agree 100%, right? But even if you can find information, there is not enough support afterwards. Especially now, the number of tourists coming to Japan literally tripled over the last five years and not enough bilingual staff to support those foreign tourists, so we decided to turn our service into chat, but the next thing we realized was there’s too many messages coming through that we can’t really reply back on time, and I don’t speak Chinese – I can’t chat in Chinese, right? So, we decided to automate it and that is how we became a chatbot company, so it was never like, we started a chatbot company.

Tim: I mean, sometimes, you have the pivot based on the information you are getting. So, right now, you support English, Chinese, and Japanese?

Akemi: So, we don’t do Japanese.

Tim: You don’t do Japanese?

Akemi: We don’t do Japanese.

Tim: Oh, okay, why is that?

Akemi: Because we built this product so that we can use it when we travel. When we travel, those places are not like Japanese-speaking countries, right?

Tim: Okay. Right, right.

Akemi: Yeah, and then also, I think in Japan, not many hotels are having problems dealing with the Japanese guests.

Tim: Right, right. Okay, you don’t have a technical background, right?

Akemi: No, I don’t.

Tim: So, did you have a technical co-founder? How do you build the original team to kind of pursue that vision?

Akemi: I don’t have any cofounders. It’s just me. When I first started, we were doing a website, so we didn’t need anyone for this, like PhD in computer science. We just needed like a very simple solid engineer-software engineer. The first person I found was a referral through a friend, so basically, what I did was I messaged literally everyone on my Facebook asking, do you have any developers? Like, do you know anyone who might be looking for a project? It doesn’t even need to be a full-time person, just a project, then I found a few and I picked the best one, so he was the first person to join, then I found a few more designers for the exact same thing, then I recruited a couple students entirely through craigslist.

Tim: Oh, really? Okay.

Akemi: Yes, so that’s how we started, so I didn’t have any cofounder, but then is our business kind of evolved, requirements also changed, right? In the beginning, we only needed like, a very simple developer, but halfway through, we decided to switch our business to make a chatbot, artificial intelligence.

Tim: Right, that requires significantly more technical ability than making a website.

Akemi: Right. So, what I did was I started asking around one more time, and I found someone who is suitable for the role, then we decided to shuffle around the team, and he recruited five or six more developers, and then that’s kind of how we started.

Tim: Okay. So, your team building has all been pretty much just word-of-mouth and personal contacts?

Akemi: In the beginning. Now, we also use a recruiting firm because we do need to speed up the recruiting process, but in the beginning, mostly, it was Craigslist or friends’  friends.

Tim: Well, let’s talk a bit about Bebot, the chatbot and the main offering. So, can you walk me through the experience from sort of the user’s point of view?

Akemi: Sure. You land at Narita Airport. The first thing you’re going to do is to connect to airport Wi-Fi, right? You were on the plane for 10 hours, of course, you want to check your messages. As soon as you finish connecting to airport Wi-Fi, our chat window, Bebot window just pops up.

Tim: Okay, so it’s integrated with the free Wi-Fi itself?

Akemi: Exactly. So, there is no additional download, and then there, you can ask things like how do I get to Tokyo, flight status, then you can ask things like best souvenirs, best Japanese restaurant, Raymond, coffee shop, Starbucks, ATM – all sorts of things, and everything is automated in English and Chinese.

Tim: Okay, and hotel systems work pretty much the same way when you log into the hotel Wi-Fi, Bebot is there?

Akemi: It’s similar. Depending on the plans, then let’s say, Hotel New Otani is one of the three most iconic hotels in Japan. You check into New Otani Hotel and you receive a little flyer with a QR code printed. You scan it, and then that takes you straight to the page to chat, or another way to try is in Japan, most hotels now do offer free smart phones as part of their amenities, and then there, you see a little banner, you tap on the banner, and you start chatting.

Tim: Okay, all right, that’s great to have such high-level integration, but even with really useful technology or even with really good placement like that, it’s often really hard to get consumers to alter their behavior and actually try something new, so what percentage of guests are actually using Bebot and engaging with it?

Akemi: It’s more than 40%.

Tim: Okay, so it’s really high.

Akemi: We went literally almost from like, 1%.

Tim: Okay, what did you do to move from one percent up to 40% because that’s pretty impressive?

Akemi: A lot of testing. A lot of testing. For example, one of the things that work in the most was to reduce, to place a lot of buttons.

Tim: To place a lot of buttons?

Akemi: Buttons, so for example, when you first start chatting, if you just get like, “Welcome to Narita Airport!” you don’t know what to do, you’re not going to start typing, right? Because you don’t know what you can ask, so we realized that by providing more options like thumbs up, buttons, persistent menu, click replies, people tend to engage wave more. Then, once we learned that the buttons actually do work, we started placing different kinds of buttons and we are placing icons on the buttons, like font on the buttons, all sort of testing. It’s just a lot of A/B testing, but then we changed things every day, and then now, we have access to over, I think it’s about like, between 35,000 to 40,000 users every day, so it is much easier for us to test.

Tim: But, the most important change was letting users know what kind of things they could ask by putting those buttons in.

Akemi: I think that was one of the most important things. Another thing that we did was of that typing indicator and a reminder message explaining what they are, so typing indicator is when you are waiting for someone’s reply, usually, you will see dot-dot-dot, so you know that the reply is coming soon.

Tim: So, typing indicator, meaning Bebot is typing back?

Akemi: Yes, yes, dot-dot-dot, and then we also learned that like, many people left of the window, left of the chat after receiving welcome message, taking no action, so we decided to add another message. We call it reminder message. If you go silent for let’s say more than 5 to 7 seconds, we send you another message and say like, “Start typing below,” or something like that.

Tim: Okay. So, what type of questions do people ask at say, hotels? What are the most common sort of questions that they ask Bebot?

Akemi: “Can I do late checkout?” Wi-Fi password.

Tim: Wait a minute, how can you ask Bebot the Wi-Fi password if you need to be connected to Wi-Fi?

Akemi: Because they use free smartphones from hotels out of room amenities.

Tim: Oh, I got it. Alright, that made sense. So, if someone is requesting a late checkout, does that mean you have integrated Bebot within the hotel’s IT systems?

Akemi: We don’t integrate our system into their property management system. What we do is just providing information. For example, if the hotel offers a late checkout or not, and if it’s a yes, then how much? Usually, when we send out the answer like yes, we can extend your stay until 1 PM for additional fee of X thousand Yen, people are like, “oh, never mind, I’ll just check out on time.”

Tim: Okay. What other kinds of things? Is it like restaurant recommendations?

Akemi: We do get that a lot. Restaurant recommendations, restaurant booking requests, transportation-related queries, we also get just like a small talk, just conversations like, “Oh, this is my first time in Tokyo. I’m so excited to having you recommendations,” sort of question. We also have people sending us lots of pictures, like here is where I am today and another user goes like, “Oh, I met the restaurant you booked for me and here is what I’m eating.”

Tim: Really? So, people are interacting with Bebot as if it is a living person?

Akemi: It seems that way. We didn’t think it was going to happen and it’s definitely not happening to all users, it’s only just a percentage of users, right? Especially after we take more restaurant booking requests, then it seems like people – those users tend to chat until they check out.

Tim: That’s really interesting.

Akemi: Yeah, like, “Oh, the restaurant is actually booked,” and then they are impressed, and they keep coming back and asking for additional requests: can I do this? Can I do that? Like, can you find me information about this and that?

Tim: Do you customize request based on like, countries, so for example, Chinese tourists looking for souvenirs might want something very different than American tourists?

Akemi: Right. So, we don’t actually do those. We customize your property. What we found was there is a higher correlation of usage behavior with kind of like the average price. So, people for staying at more expensive hotel with may be preferred to dine at these places. Let’s say, restaurant, other’s price is more than $200 compared to users who are staying at $100 per night kind of hotel, and then they would prefer to dine at different kinds of restaurants.

Tim: That makes sense, and kind of keep things in the appropriate budgets. Do you do different recommendations for someone that might be staying in the two-room suite versus a guest that has booked a single from a discount travel website?

Akemi: We don’t right now because we don’t know which room they are staying at. There is no property management system integration, and we intentionally keep it that way. We don’t want to take too much personal information.

Tim: Okay. A few days ago, I tried out Bebot for the Tokyo station Bebot, and I think I understand what you mean about the experience, but first, you are not sure what you can ask, so I started asking things like where I could get money exchanged and where I can catch the train to Osaka and was really good, and then I asked it where I could get good apple pie, and it got confused and said you need to ask of the station managers about that, and I asked a couple of other questions, but then 20 seconds later, Bebot came back and said, “Oh, there’s a famous apple pie place to stops away. Here’s the link.” I was like, oh, okay, that’s pretty cool. What type of databases are you using on the back and to get this information?

Akemi: So, we have four different layers. Top layer is our own recommendations, then second layer is we receive information, recommendations from our clients too, then the third layer is we have partnerships with a bunch of different restaurant booking companies, activity booking companies. So, we can take information from their database as well, then the last layer is we also use external APIs such as Yelp or Google Place, and depending on how you ask and what you ask, we take information from different places, and if it’s the first time that we are receiving, let’s say Apple pie curry, it would take is a bit of time to find information.

Tim: Okay. You mentioned at an earlier interview that Bebot is actually a combination of machine learning and human operators. When in the flow, points this a jump from the machine learning to the human operators?

Akemi: It depends on the contract. So, let’s say high end hotels tend to pay us more for those human assistants, and in airports and train stations, it all depends. So, for example, if people say there is a terrorist or I found a bomb at the gate or something, and that sends us notifications to our operators so they can jump in.

Tim: oh, okay, so that humans become involved with specific keywords or is it just went Bebot can’t answer a question?

Akemi: Specific keywords.

Tim: Specific keywords?

Akemi: And then also, if people go like, “I’m having a fever, flu, need to see a doctor,” those are also keywords that would send us notifications.

Tim: Interesting. You know, you mentioned before that people were sending Bebot pictures of their food and saying thank you for making the reservation and treating it like a person in some way. So, do most people interact with Bebot as if it were a computer, or do get people that are on one extreme may be very friendly or on the other extreme, a lot of tired, frustrated travelers kind of taking out their frustration on Bebot?

Akemi: It all depends on the facility. Usually, guests from five-star hotels tend to treat people more like human assistance, then airport, it all depends. If it is a departure terminal, people tend to have more time, so they are like, “Yo, what’s up?” and “Are you married?” Like, “Where can I meet girls?” all sorts of different questions, and it’s like chatting with a friend, but in the arrival terminal, it’s more like an interaction with a robot because they just arrived and they just want to pick up their bags and go, right?

Tim: So, do you think that’s because the arrival is a stressful busy time, or do you think you’d like, by the time they are going to departure, they’ve had experience with Bebot and gotten used to it?

Akemi: I think – no, I think it’s just like where the location timing.

Tim: Okay. So, how do we incorporate feedback from the users to find out, for example, if a particular café or restaurant is good, if the users enjoy it?

Akemi: So – but then in case of Narita Airport, we don’t take feedback on individual shops because our contract is with the airport. So, we take feedback on let’s say, the Wi-Fi speed was not good, or there are not enough chairs, not enough hotel rooms inside the airport. Those feedbacks are very valuable for our client.

Tim: Is the feedback just information users voluntarily type into Bebot or do you send out like, does Bebot ask questions about the satisfaction?

Akemi: Oh, we also do ask about satisfaction. So, for example, if you say thank you to Bebot, you get a reply like, “Glad you think so and how would you rate me for the service?” You see a button like thumbs up and thumbs down.

Tim: Let’s talk a bit about chatbots in general because I think that the technology is going through this interesting phase right now. So, a year or two ago, maybe about one year ago, we were sort of at a peak chatbot fight and where chatbots were going to be used for everything, and what he is saying chatbots are really good at and what type of applications do you think that they are not quite ready for just yet?

Akemi: In my opinion, chatbots in general are quite useful when it’s used properly, but what I see now in many cases, it’s not actually done properly.

Tim: So, what does it mean to do it properly?

Akemi: Sure. For example, it needs to be very, very vertical. It can’t be just like, travel in general – I booked you a hotel, I book you this and that. That does not work, so for example, we also try booking rooms and booking activities, and never, never really worked. So, what we learned it was people don’t really do high-value transactions in a bot, and then also, matching vacancies and requests, those are very difficult too.

Tim: That’s really interesting because that is one of the things I desperately want a chatbot to do for me because it is so complicated. What was the problem? Why didn’t it work?

Akemi: So, people go things like, “I’m traveling with 10 other friends. Can you also give us discount?” It does not happen – it is not in the system, right? You can’t really match it unless –

Tim: So, it’s too much variation.

Akemi: Too much variation, and then also like, a lot of things are very, very irregular, and I can give you another example: so, one of the chatbots that I do enjoy using on a weekly basis is Kuroneko Yamato, the chatbot delivery service. You just enter your pin number – not the pin number but the delivery number, then you choose when, like today, tomorrow, day after tomorrow, and after that, we choose up morning, afternoon, evening, and then you just hit done. It’s very simple, but it works very, very well.

Tim: So, would it be fair to say that chatbots are really used in domains where there are very clearly defined rules?

Akemi: I think so, yes, exactly, yes.

Tim: So, the artificial intelligence part of chatbots is really more on the understanding the input side and not above making decisions inside?

Akemi: Right.

Tim: Okay.

Akemi: So, I think a lot of companies tend to mix everything together. In our case, we had the same problem in the beginning. The very first year, we built everything based on our assumption thinking oh, users probably want this recommendation, that recommendation, and then maybe they want to book a taxi and they want to book a room, and they want to book activities, and all these things that we built, not many people used it. So, we decided to kind of like, through all of them away, and we started chatting manually one more time just to collect information, collect the actual queries.

Tim: That makes a lot of sense. It’s tricky to know from – because from the outside of a startup, all chatbot startups are talking about AI because everyone wants to hear about AI – the VCs want to hear about it, the users want to hear about it, but often like, inside the company, it’s a much more practical approach where you are going down a decision tree.

Akemi: In our case, I mean, we also to make our own natural language processing and we do heavily invest in that, but it doesn’t mean that we would only do like, AI part because we also do know buttons work perfectly fine, and in that increases engagement much, much higher, and then also, Smalltalk part is very important and that can be hardcoded, that can be dynamic – it all depends, right? And, we also integrated lots of APIs, external APIs, for example, when you enter flight code at Narita Airport, you get an instant reply like, “Here’s your gate,” or like, “Status: You need to go to the gate right now. Final call.” All these messages. A lot of different factors –

Tim: Right, it’s not simply answering a question. It’s being very proactive.

Akemi: Yes. The most important thing is just to get users questions answered, right? They don’t care if it is a human or a robot. They just want to get their questions answered.

Tim: I think that certainly seems why someone would start using it, but as you were talking about some of the most useful functions are things like those reminders where Bebot is being proactive and when you were talking about importance of small talk, so small talk, like for example, when a user sends in a picture saying, “Oh, here is my dinner. Thanks for the recommendation,” Bebot would say something like, “Oh, it looks delicious. I’m glad you enjoyed it,” that sort of thing?

Akemi: Mm-hmm.

Tim: Yeah, on a technical level, I noticed that you used to run Bebot on both Facebook and WeChat, but you stopped doing that. What was the reason for that decision?

Akemi: The biggest reason was not everyone has Facebook or WeChat. We needed to make it consistent, so like our users have problems like remembering their password, so in our case, we did not make Bebot public, meaning you needed to receive a token from a hotel and you needed to enter the token, then after that, you needed to enter your Facebook Messenger login credentials and that was a long step, then we thought okay, maybe what if like, we just removed the token, but our clients didn’t want to make their chatbot public to all these people were staying at their hotels. So, authentication method needed to be in place, so we decided what if we may be turned into just a web chat so anyone can chat as long as they have access to the link?

Tim: That makes sense. Facebook and WeChat aren’t part of your customer acquisition, so they end up being a barrier to use.

Akemi: So, that was the biggest reason. Second reason was, Facebook, they do a good job updating their APIs too often, and every time they make changes, we also have to make changes on our side, and there are so many limitations around it, like we needed to use their template.

Tim: And, I guess even long-term, a lot of companies have found out that placing their business on top of Facebook is a dangerous thing to do.

Akemi: Agreed. WeChat was the same problem.

Tim: Yeah. Because of the 2020 Olympics, right now, Japan is going through this real – well, there has been a tourism boom going on for the last few years, but it’s really gearing up for this year and next. How does that factor into your business plans?

Akemi: We are gradually shifting our client base more towards cities and governments instead of individual hotels to cover broader area. So, for example, this year, we started servicing one of the most popular tourist destinations, Kyoto city. So now, when you connect to Kyoto city Wi-Fi, you see Bebot there.

Tim: And do you plan to introduce Japanese language support? Because it sounds like as you expand to cover more general city information, that will be something a lot of Japanese travelers would want access to too.

Akemi: We avoid Japanese for now for several reasons. Biggest reason for us is there is not enough market, we believe. The declining population, we think our time is better spent making let’s say, Spanish or Thai.

Tim: Really?

Akemi: So, that is the biggest one, and the second is technically, it’s also more challenging making Bebot available in Japanese language, and then another reason that I don’t normally mention is expectation from Japanese users are much, much higher than known Japanese customers. I think especially let’s say, you go to information counters or concierge desk at the hotel, usually, people who are complaining are locals because they expect much, much higher standard of service, so you can’t just send emoji and be like, “Sorry, I’m not smart enough!” when you don’t have the answer.

Tim: Right, right, that can make Japanese consumers even more angry.

Akemi: So, you know, it’s like a combination, but primarily for those three reasons, we avoid it for now.

Tim: Alright. I’ve also noticed that NLP, natural language processing in general is far more advanced in Chinese and English than any other language.

Akemi: I think more data available.

Tim: Yeah, in bigger companies working on it too.

Akemi: Agreed.

Tim: Is that part of the reason behind the technical roadmap as well or has it been strictly business reasons?

Akemi: Both.

Tim: Are you thinking of going global with this? Because it seems like this is a system that is needed I mean, everywhere.

Akemi: We are already running some of our services outside of Japan since last year – wait, 2017.

Tim: Can you talk about where?

Akemi: So, Philippines, Singapore, and then we are starting very soon in the UK as well.

Tim: Oh, that’s fantastic!  Let us talk a bit about your customer acquisition strategy, not so much the users, but bringing the hotels and the stations online. How did you approach those organizations to sell?

Akemi: In the beginning, so first year, second year, and third-year, now, it’s the fourth year, it’s very different, right? So, first year, I didn’t have any hotel contact, so what I did was very spammed everyone on Facebook one more time asking, do you have any friends who work at the hotels? And surprisingly, many of my friends did, so they introduced me to those people who work at the hotel and I started going to events, seminar – like hotel seminars, and we got the first customer in maybe two months.

Tim: That’s incredibly fast!

Akemi: But we still didn’t have the product yet. We sold it before we built a product.

Tim: No, that’s always the best way to do it. I don’t think I’ve ever built a product I haven’t sold already.

Akemi: Okay, okay, well, I’m glad you say so. So, we sold it first, then the clients paid for it. We used that money to build a product, then after that, we wanted to get big name the client, so we went to Holiday Inn and said, “Look, we can make an offer for you. Would you like to use it?” And they were having problem communicating with guests. 90% of their guests are foreigners, but they didn’t seem to have enough bilingual staff at the front desk, so they were like, yes.

Tim: And the buyers’ motivation was the international support – the Chinese and English support?

Akemi: Exactly.

Tim: Is that true to this day? Is that still the major driver behind Tokyo station?

Akemi: I think so, 70%, 70%, and then were hotels approached us and asked we got more press coverage, Narita Airport found us, so they came to use and said, “If you can do hotels, you should be able to do the airport.” We were thinking , that volume is going to be massively different and the type of queries are going to be different, but then we were like, sure, then after a few meetings, the airport agreed to sign a contract with us. So, we launched at the airport. It was supposed to be just a proof of concept, very small POC, but then we did well, so they kept renewing the contract and now, we have a permanent contract. After that, Tokyo station of JR Railway Company approached us and they said, “Look, if you can do the airport, why can’t you do the station?” We were thinking, you know, station, Tokyo station, it’s massive. There is no way we can support that kind of volume, but we didn’t want to say no to that opportunity, so we said, “If you give us three weeks, we can build it for you,” then we did, then after that, Kyoto city approached us, “If you can do station, why can’t you do the city?” We were like, why not?

Tim: Well, I mean, it’s a high-quality problem when you have new customers coming to you like that. What is the biggest difference though when moving from like a hotel to an airport, is it just different supporting databases you need to put together? Do have developed new models to understand the type of queries?

Akemi: it’s entirely different, it’s entirely different.

Tim: Really?

Akemi: Yeah. We share some of the platforms of course, internally, but a lot of things are different. For example, user behavior is completely different. With hotels, you want to make sure you spend a longer time interacting with users. For example, the reply does not go out like, as fast. The airport and the station, people are usually in a rush, so they want a reply instantly. They don’t want like a dot-dot-dot like, the indicator – things are very different, and then with airport sensations, conversations tend to end much quicker than with hotel guests. Hotel guests that are like staying for like, five days, six days, they have lots of time to chat, so they are like, “How are you? How was your day? I’m looking for may be breakfast, maybe lunch. Do you have any recommendation?” and the chat goes on forever.

Tim: All right, and then it makes sense because they are at a hotel, particularly a five-star hotel where you’re going to associate a good experience with a slower, more complete, more polite response, whereas if you have just gotten off the plane, all you want is that information. So, Bebot has to just completely different personalities?

Akemi: Mm-hmm.

Tim: Excellent. Well, listen, Akemi, before we wrap up, I want to ask you what I call my “Magic Wand” question, and that is if I gave you a magic wand and I told you that you could change one thing about Japan – anything at all – the education system, the way people think about risk, the legal system – anything at all to make it better for startups and innovation in Japan, what would you change?

Akemi: Decision-making process.

Tim: Really?

Akemi: For large corporates.

Tim: So, what would you change about it? Just the speed?

Akemi: The speed.

Tim: Why do you think it take so long here?

Akemi: Because 20 different people have to stamp the paper, and if one person is on holiday, your contract is like, on hold for like, three months.

Tim: The dreaded ringisho.

Akemi: Yeah.

Tim: And for our foreign listeners, a ringisho in Japan, it’s an authorization form that certain people in the company have to stamp it in order for a contract or an agreement to be approved. Some of your customers decided pretty quickly.

Akemi: I think so. We got lucky.

Tim: Well, why do you think they were able to make that decision so fast because it’s conservative industries?

Akemi: Yeah, like railway company, airports and cities. There are lots of paperwork, but they did fully understand how important it was to provide assistance to those people coming to Japan.

Tim: So, do you think that sense of urgency made them act faster, or do you think that sense of urgency made them make an exception to their normal processes?

Akemi: I don’t think they made exceptions, but they were like, “Okay, we have to close this suit would because it’s an AI solution. It learns over time, and the sooner we do, the better.”

Tim: Okay, so big companies, when it is important, they can move fast.

Akemi: I think so, yeah.

Tim: That’s encouraging.

Akemi: I never knew, I never knew. Some of our big clients, it took them maybe almost 2 years to close a contract. We have some of those as well.

Tim: Huh. So, what advice do you have for other startups who are trying to sell to big companies, and startups can’t wait around two years for a contract because they will run out of money?

Akemi: The quickest way is to find the decision-maker. If you can’t meet them, find someone else who knows them.

Tim: How do you figure that out? Because in a Japanese company, it can be really hard to know who the actual decision-maker is.

Akemi: You ask around until you get the answer.

Tim: Yeah? So, you have to do a lot of kind of word-of-mouth research?

Akemi: Yeah.

Tim: Yeah.

Akemi: You also need to be in the right community.

Tim: You know, I noticed that it seems like a lot of your strategy is based on word-of-mouth and personal connections from recruiting to initial sales, to even your current big-company sales. It’s a really personal direct involvement.

Akemi: I’m just like, very persistent, I have to say.

Tim: Well, that seems to be working.

Akemi: I would keep asking you if you don’t give me the answer.

Tim: That’s fantastic! Well, listen, Akemi, thanks so much for sitting down with me, I really appreciate it.

Akemi: Thank you for inviting me here.


And, we’re back.

I think it’s interesting that Akemi Bespoke is focused on English and Chinese and have stayed away from the Japanese language. I think that is a first. I mean, I talked with many, for example, AI companies from Germany or France where the English language accounts for most of their business, but I have never met one that does that to the exclusion of their own language.

But I have to admit, Akemi’s reasoning on this point is quite sound. The immediate market demand is very English and Chinese, the potential global market demand is overwhelmingly English and Chinese. The most advanced natural language processing research is being done in English and Chinese, and the high level of Japanese customer expectation sets a much higher bar for the kind of sophistication that would be needed in a Japanese chatbot, so yeah, it makes sense to forget about your own language and focus on what your customers need.

As always, it’s not about you. It’s about your customers.

The other thing that struck me in our conversation with Akemi was the importance of personal connections to her success, and I don’t mean personal connections in the cynical sense of knowing powerful people; Akemi’s connections were all from her Facebook friends and her school friends, and just regular people she knew – people who are willing to help her out because they liked her, and they liked what she was trying to do.

Even now, Bespoke is growing quickly but in a very capital-efficient and somewhat organic way, and what is most interesting here is that this is the exact opposite of what Silicon Valley investors and founders insist is the only way to grow a new technology company. The standard playbook is that you need to raise the most money to hire the best engineers, to crank the PR machine up to 11, to gather the most mind share and then prevent the competition from ever getting a foothold.

Now, sure, that can work, but sometimes, the best thing is just to solve an important problem and to do it really well.

If you want to chat more about chatbots, Akemi and I would love to hear from you. So, come by DisruptingJapan.com/show138, and if you leave a question in the comments section, I guarantee you, I will respond, and most of our guests are pretty good about answering listeners questions as well. Also, if you find yourself killing time on social media, be sure to check out our Facebook and LinkedIn sites sometime.

But most of all, thanks for listening and thank you for letting people interested in Japanese startups know about the show.

I’m Tim Romero and thanks for listening to Disrupting Japan.