How Bespoke faked AI until it actually worked (w/ Akemi Tsunagawa) | E2320

30:58Yes. They need to appreciate this group of people coming to help and then also still maintaining the culture. It's a delicate balance. How is Japan doing? If you were to look at it, you think it's doing a pretty good job of managing the issue? We can certainly do better in many different areas. For instance, the multi-language information is also one thing city governments could do, right? So the number one problem oftentimes occurs in each area, like local area, is taking out the garbage. This creates problems. like non-Japanese workers won't follow the local rules because they can't read the instructions and the instructions are somewhat complex it is complex yes yes i literally at my airbnb it was this big with tons of pictures explaining it combustible not combustible plastic with food on it plastic without food on it and trying to explain it is like um yeah a little bit of work there's a resistance to putting it into a foreign language or it's just a burden for the local community to do that i guess um it's it's a bit both so like locals would say things like they don't belong here and they can't even follow simple rules right but then city governments need these foreign workers right so local factory can operate so there's all this tension yeah it's interesting tension now there's a third business you discovered there's a tension here and the immigration is not going to solve the entire problem.

32:41It may take them from six ships to maybe not going to five and not going smaller, but it might not bring them to nine. But there's some revolution going on right now with human robotics and even non-humanoid robotics. So this is another area you've identified for a future product. So tell me about the problem you discovered and then how you're thinking about your third business line? Sure. So now I spend half of my time inside chemical tanks at shipyard. Chemical tanks. Chemical tanks. Very glamorous. Yeah, thank you. Yeah. Well, my work has never been more excited than this, really. And then when we first started offering the training solution to shipbuilding companies, construction companies, many different companies, they all appreciated the service, but then after a year or two, some of them said, there's only so much we can do with non-Japanese workers.

33:42Certain tasks need to be done by humans, but not every task. And they asked, is there anything more that you can help automate? We're not a robotics company. We didn't think there was much we could do at that point. But then we actually spent more time speaking to these construction workers and shipbuilding experts. And then as we spent more time looking at how they actually do their jobs, we thought it might be a good idea to actually create a robot that can help automate part of the inspection that they do at shipyard. This is a single task for a robot. It's a repetitive task. It's a boring task.

34:33Very boring, I think. Perfectly suited for a robot. And then that robot can report into maybe one person. You could have 10 robots reporting into one person instead of having to fill those 10 positions. Exactly. So one of the things that we looked into first was scaffolds. So scaffolds are everywhere at construction site, shipyard. And scary thing is, even though the law requires you to inspect manually every day before you use it, no company has resources to do so. Right. So they may skip it. They skip it almost 100 % of the time. Okay. And then it does result in injuries and severe accidents.

35:18In extreme cases, people... Could die. Yeah. Very tragic. Yeah. And there's no reason for it because you could, with a very simple robot and cameras, do a better job, and you could inspect it 10 times a day for free. I mean, once you have the robot, an incremental inspection has got probably zero cost, yeah? Exactly. At first, we thought it was going to be very easy. But then we learned that at shipyard or plants, that's another segment we're trying to help, there are shiny metals. So the reflections make it very difficult. So we use V-DAM to solve that problem. What we are essentially trying to do is the robots climb up scaffolds by itself, grabs the joint, shakes it, and then record each joint is safe or not.

36:13And then send the inspection records to whoever is actually in charge of the inspection so that person can come in and fix it before accidents. Right. And they could even take a video, I'm sure, if they wanted to so they could pre-inspect it. Yeah? Exactly, but the next problem we encountered was because no one inspects scaffolds, it's not a sexy thing and it doesn't produce any profit for companies. We spent more time interviewing them and learned there are other type of inspections we could also automate with the same robots. So a paint thickness inspection, a shipyard, some companies that produce ship over like 200 meters, they would be spending more than 1 ,000 human hours to do the paint thickness inspection.

37:04They literally just scan the ship with a device like this. And then it's like, okay, this is enough. And then you scan, scan, scan. 1 ,000 hours. That can be easily automated with a robot. Yes. At$20 per hour, it's$20 ,000. And every ship, you have to do it over and over again. And probably the human could make mistakes. Exactly. So there are like three other type of inspections we're looking to automate. It would be all done by one robot. You know, it's very interesting when I hear your story and your approach. It's very similar to, you know, what we do in America, which is spend a lot of time doing customer interviews.

37:44Did you learn that technique from some startup book? or is it just you understand, hey, hanging out with the customer and asking and being curious is the secret to developing great products? How did you learn this technique? It was from the first business, the chatbot business. We didn't create a chatbot company first. When we first started a company, it was a simple, our product was just a simple website, similar to TripAdvisor, but local version of it. and then we well i have finance background but not like the product background i didn't know what to do so i just went to shibuya crossing or roponi station grabbed any tourists with a giant suitcase asked a bunch of questions and uh user interview cut us to the next business which was the chatbot and some founders have a hard time doing this because they're nervous in of other people right so do you get nervous stopping somebody at a train station and saying can i ask you questions they might think you're crazy i got kicked out of rapongi station many times i also got banned from major shrine oh wow we also when i got they put your picture up in the security office don't let her in the security guards all knew me so i stayed away strange she keeps asking them how they could have a better experience in tokyo it was a very strange thing And then we also ended up using a dating app to source users because at the time there were some non-dodgy dating apps we could find, all sorts of non-Japanese tourists.

39:21So that type of experience helped me learn that actually speaking to the end users is the quickest way to build successful business. Yeah. Any other startup hacks that you've learned along the way or stories of frustration that led to opportunity? There are so many other things. But then my favorite one is when our chatbot broke down during Christmas holidays. That was a high season for us. We had high-end hotels as customers and we had Narita Airport, Japanese government, all sorts of customers. And then after our engineers made update to our software, it just simply stopped working. So for the next one week, I was the chatbot.

40:11Wow. But then... No sleep. No sleep. I responded faster than anyone. And then during that time, we learned that human-like communication actually increases the user engagement. So we went back and then copied the whole user interaction between me and the users, and then tried to implement the same type of conversation into our chatbot. And then suddenly the engagement has gone up so much. And then our chatbots became very popular even beyond the segments that we were trying to go after. So you had to make the chatbot less efficient, more human, maybe a little more chit-chat? Yes. Interesting. Because I used to think chatbot is just to answer a question.

41:01You have a question and you want your answer as quickly as possible. But that wasn't really the case with the tourism. But then we also have emergency response project and that works very differently. For that purpose, for emergency response, you want the evacuation center address as soon as possible. So you don't want any chat chat, but a different chatbot for different use case. So just asking somebody, is it your first time in Japan? Or what are you looking forward to in Japan? Or something like that increased their utilization and their NPS score, their happiness with the product? They definitely did, yes.

41:37Yeah. Such a good insight. And what have you learned about raising money from venture capitalists who didn't understand the first business And then you have gone into, I'll tell you, what would seem to an investor as three completely different businesses. But I understand the thread. Each time spending more time with customers seeing the same problems around language and around work projects that are under-resourced. Each one is an under-resourced problem, right? They're trying to get a job done, but they don't have the resources, so you come up with a better resource. One better resource is an AI chatbot.

42:13Another good resource is video training, like YouTube-style video training and quizzes. Another technological source for not having enough employees is a robot. So you are trying to solve not having enough employees, yeah? That's the theme. Yes, and actually, even technology-wise, there are some consistencies. So we still use chatbot for the second video learning business. When people have questions, then we answer. And then also we generate quiz of the videos. It's still like the text, right? Yes. And then for the third business, the second business involves recording and video processing. And it's kind of similar for the third business as well because we still need to get people, workers put on the smart glasses and record how they do inspections.

43:08Right. Then we process the video and then create simulation. So it's kind of connected, but then it was difficult to convince investors. One of my friends built a rocket ship company, and then he learned a lot about materials, and he was working on a car company. And in the car company, he learned a lot about batteries. And then the battery technology and the stamping of metal and building a factory for the rocket ships then informed how they built the factory for the cars. And then they needed energy. And then the cars needed connectivity. And all the material science, he was able to apply to both companies.

43:49And eventually, when I asked him, or we were talking, and he said, really, my job is just to make a machine called a factory. And that is what Elon does, Elon Musk. His job is to make a factory. And so that's his speciality. And the new factory is to make robots. And the other new factory is to make chips. But he's already made a factory to build rocket ships, a factory to build satellites, and a factory to build cars. So you're just building another factory, right? And so the tools, they are going to be different, but the factory is a factory, yeah? Yeah. And so very, very interesting. What have you learned about raising money, though, in venture capital?

44:31and I'm assuming, I know you've raised a little bit of money, but are you able to build these businesses without a lot of capital or do you still need to raise a bunch of capital? So for the initial business, for the chatbot business, we raised about 450 million Japanese yen or so. But the last round was back in 2020, so March 2020, like right in the middle of COVID. and then we've been self-sufficient ever since. Got it. So you've raised low millions of dollars, I think. I'm not sure what 450 million is in U.S. Ah, so that's like 3 million U.S. dollars. And we also secured a lot more bank loans because we have a lot of government contracts.

45:18I learned along the way that we can use the government contract as the collaterals, which is much faster than speaking to investors. Ah, so this is very unique. In America, very hard to government or banks don't give loans to startups. They only get in America, you know who they give loans to? Rich people. When I had no money and I was starting my businesses, the only loans I could get were American Express and Visa. And they made you give the money back 30 days, 60 days later. No banks would give you a loan. And then when I made some money, the banks would call over and over again and say, would you like lots of loans?

46:02We'll give you as many loans as you like, and we'll give you 2 % interest, 3 % interest. If you want to buy a home, we'll give you only interest payments for five years. It would be 10 years, just interest. Then you pay principal. And anytime I wanted to buy a house or anything, they just give you a loan because you had some assets, and they just made it so easy. It's kind of backwards. The rich people don't need the loans. They have assets, and it makes it easier for them to increase their assets because they don't ever have to sell an asset. They can keep their stock. They can keep their homes, and they can keep appreciating, and they get loans against it.

46:41It's a very big problem in America where we really do need to give small businesses loans, but the problem is they can default on it. Right. Here, as long as you're profitable, you can secure loans. So it's a bit difficult for startups, but as soon as we turn profitable, it was quite easy. And what do they base it on? They look at like, oh, if you're making$1 million a year in profits, we'll give you$2 million? Or is there some ratio? Did they actually look at the contracts and make sure they're a strong contract? In our case, they demanded the government contracts, and then the loan they gave up was equivalent to the amount of the collateral we provided.

47:22Ah, so if the contract was for$1 million, they'll give you$1 million. Exactly. Would they give it to you for the life of the contract or just in advance or do they give it to you for a little longer? For the short-term loan, it's only until you get paid. But for the long-term loan, the longest we have is 10 years. Amazing. Yeah, and then it's like 1 % interest rate or something. So beautiful. Oh, wow. So, you know, this is interesting. There should be more startups, but I understand the goal of getting to profitability now. Getting to profitability means you don't have to sell your equity, so you get to keep more equity as the founder.

48:00Correct. Well, this is good. And then, so what is the outcome for your business? Do you aspire to just keep growing here in Japan, or do you think, hey, maybe I should go to other countries and apply this there? Because in the Middle East, in UAE, in Qatar, in Saudi, they have the same exact issues. The society is growing. It's very, the society needs workers. And there, you know, you might have 500 ,000 Qataris and a million immigrants. So it's like two to one immigration. But they have the same issues around training, around integration into life, and just not causing, you know, friction in society.

48:43So have you thought about expansion plans globally? Or you have so much work here to do? We have thought about it. and then right now we think at least for the next two years or so, Japan is the hottest market. So we want to get to the point where the robots are working and we buy a couple of factories and then try to kaizen those factories. Yes. And then we take that model overseas. Just to give you a little bit of credit here, you want to go from building a chatbot to building robots and having your own robot factory. So we'll buy, let's say, like a small shipyard and then deploy our own robots to show them how efficient it could be, there are successor problems in Japan.

49:25So there are many companies that go bankrupt here are actually profitable. They simply don't have successors. That's amazing. So I just want to make sure I understand. There are profitable businesses that go out of business because they don't have somebody to give the business to. Yeah, not bankrupt. Exactly. Just to go out of business because they don't have anyone taking over the business. They have no kids to take over the business. Oh, my God. It's such a big opportunity. Younger generations don't want to be factory owners. We want to be factory owners. I want to own a factory. This is a great idea.

50:03Wow. I want to invest in this company. Go buy factories. And what would they charge you if you're the person retiring? What would they charge you to buy their factory? Like one times revenue or they're just happy to get anything? Could be lower. Could be lower. So you could even maybe they give you owner financing too. Would they give you owner financing where the owner sells it to you and then you give them money every year until it's paid off? Or do you have to go to a bank and get the money? I think it depends. Depends on the term. Yeah. This is an incredible opportunity. It reminds me when I go out to go skiing in Hokkaido, there's a bunch of Americans now on Instagram.

50:41And they were telling me six or seven years ago and I started skiing there, oh, you know, you can get a house here for free or for$1. It's true. Or they'll just give it to you if you pay your taxes and you keep the house from collapsing. Because all the houses outside of Naseco and in the countryside, they just keep collapsing because too much snow and nobody's lived there for 10 years or 15 years. And so these young kids from America are coming here and they go renovate the houses in the countryside and then they rent it during ski season and that's their revenue stream and then they talk about it on their instagram accounts and all these people in america now want to come and have a countryside house and and rebuild it it's very interesting phenomenon this incredible opportunity yeah it's everywhere in japan i think every city you would go to like outside of Tokyo, a metropolitan area, they have the exact same problem.

51:40And it's because young people can't find work there, or they just want to live in the city because the city is so much fun and convenient and so great? So when kids turn, let's say, high school age, maybe oftentimes there is no high school in that area, so they already leave their town to go to school in the city, or maybe university, right? Once they leave, they don't come back because there's no job. Wow. So all of those towns are getting hollowed out. They're turning into ghost towns. There's so many of them. Yeah. So this is what happened in, I made a friend out in Niseko in Hokkaido. He found an old ski resort.

52:22There's hundreds of ski resorts in Japan that are abandoned as well. And he convinced the local town to let him take over the ski resort. But all the lifts were broken except for one. so the one lift is available to the locals the remaining locals um or anybody japanese who wants to drive out or take the train um and then the other ski routes he has um like a cat you know like with the tracks and he drives people to the top to ski uh one run every hour so you ski seven or eight runs you know you can do it maybe 45 minutes to an hour depending on the length of the run, drive up for 15, 20 minutes, ski down for 10, 15 minutes, do it again, do it again, do it again.

53:05But he had to convince the local town to give it to him. So incredible opportunity, I think. AI is allowing you to do more with less employees. Are you playing with these tools, with agents, Claude Code or Claude Cowork? Has it infected your company yet to make you build products faster? So yes to all of those questions. Because we've been doing AI since 2015, We, of course, you know, adapt many different AI tools in-house. Half of our team is actually non-Japanese. We have 65 people, but more than half is actually non-Japanese. And on the engineering side, every single person, of course, uses, clothes, many different services.

53:50so in fact last year when our German Germany engineer had to move back to her country we didn't replace her because we just decided hey like everyone can use AI agent like the code review is now faster and then there are things we can still improve in the process so instead of like you know spending extra efforts to hire train and manage the person And why don't we just like, you know, get everyone learn how to code with AI? It's very interesting. We're now asking our startups to teach us as investors in their companies when they are going to consider a new hire. How do they make the decision to hire a human or automate with AI?

54:41Right. And so we're seeing many of them where, you know, at a certain point of scale, like 60 people, you might go, hmm, we need to have three accountants. We need to have three people in HR and, you know, this type of thing. And, you know, maybe three people to do meetings and scheduling and assistance. And now people look and go, huh, I think we could have one person who builds, you know, the HR department. And if you write a job description now, have you used AI to write a job description recently? All of it, yes. All of it. And it's mind-blowing. It's so much better than the ones we used to write.

55:22I've had this conversation now with like three or four founders where we all had the same experience. I've been writing job descriptions for over 20 years. I've had hired people and I used to pay them$1 ,000 or$2 ,000 to write a job description because the right job description could get 10 times more applications. And then I started asking Claude and it had access to our Slack, to our Notion, to our Gmail, and it would incorporate things from the company for that position. And then I would ask it in free flowing. I use a microphone, WhisperFlow, to just talk about what I want the person to do.

56:03And it structured it so perfectly. And I wonder to myself, how can it make such a perfect job description better than I can? And it's so obvious. There's so many message boards online with jobs. Yep. And they've all been indexed by the AI companies. Exactly. So they actually, there's never been a human who had all the job descriptions in the world in their brain. And now AI has every job description in every country, in every language. And it's crazy, like, what it can do. So amazing. So we do rely heavily on AI to do pretty much everything. Like every job in the company is now done by AI. And sometimes I even wonder if the company still needs me.

56:50I think it's a lot of people who are in the management positions who their job is to make everybody at the company better. Well, the AI understands everything they're doing. It has perfect insight. So I have been thinking about creating a second CEO, but that CEO has access to every email of every person in the company at all times and can know every Slack message and everything they've ever done so we can give better advice. so that's what I'm working on right now in my company is I had everybody connect and ask Claude Cowork give me a summary of my week on Saturday based on my Gmail my Notion edits my Slack messages and my calendar and I used to have the human do that tell me at the end of the week what you accomplished and then tell me what you're going to do next week the AI does it and then the next phase is to ask the AI, what should I be doing?

57:51What should I do next? And that's the question we always ask ourselves as CEOs or leaders. What am I supposed to do next? And the only way you figure that out is either by meditating on it. If I go skiing, I can maybe figure out what am I supposed to do next? And well, it's your choice, right? You're the main character. So you get to choose what do I do next? But do you have a coach, like a CEO coach? Do you have one? So my number two person at work is almost like the CEO coach. He's got a lot more experience than I do. He's American. We imported him from the U.S. Oh, good. And so he can give you coaching and this kind of thing.

58:31So this is what I'm thinking about is, hey, what did I spend my time on? Was that the highest use of my time? And what else can I do, right? And what should I do? And I'm trying. Do you guys use Slack? What do you use for? We use Slack, yes. This is, I think, and you've been using it for how many years? 11 years. Yeah, same with us. So we have the entire knowledge in Slack. But do you have the Enterprise Plus version or the regular version? Enterprise Plus. So, like, it's very expensive. So with Enterprise Plus, you can see every DM or every, like, group people chatting. You have full access. So now you can connect the API to it.

59:09So I just upgraded it this week. I've been working on it. and I'm going to ask it just to look at the history of the company and make a timeline and say, what could we have done and how should that inform what we do for the next 10 years? What did we do for the last 10 years? What could I have done and the whole team done better? And then do forward. Have you connected and created like an oracle, a god, a slack god yet? We haven't yet. But we're still carefully evaluating because part of me feels like we shouldn't be going through some of the messages, even though I am not going to personally go through them.

59:47Yeah, I think you have to let, I let everybody know at the company, even DMs, everything's recorded because we're a finance company. Of course. In a finance company that exists. So hopefully employees aren't complaining about the boss in DMs. Hopefully they're smart enough to do that. Your people are laughing. are you please don't i don't know are you guys uh no kabir is like nope i'm smart we do that on signal we complain about the boss on whatsapp disappearing messages but i think this will be the big unlock is to have the oracle doing it so um but i've been thinking maybe there should be a new slack like um a more open source slack or i'm thinking about maybe having cloud code or building our own version of open source Slack kind of thing that is made for this purpose.

1:00:36Like what should we do as an organization? The other thing that's pretty crazy is one person on my team said, you know, there was this person who was at the company for seven years, this person who's five years, this person retired, this person went to anything. Would you like me to recreate them? Oh, genius idea. I haven't thought about that approach. So. Genius idea. It's genius. Then you don't need to rehire them. Well, you can go, exactly. You don't have to rehire them. Or if you have a question, you can say, ghost of Jason. What happened in the first five years of the company with these three investments?

1:01:14And it goes, ah, yes, I can tell you the conversation I had with that person. So it's like Luke Skywalker being able to talk to Obi-Wan Kenobi's ghost. You have a ghost Jedi. You know the movie, yeah, where they come back as like a force ghost. so I thought oh this is incredible for continuity then I was like do I need the permission do you need permission I don't think you do if the person is no longer with the company you don't need permission right but to use their name and likeness yeah you kind of need a permission maybe for the name unless you change the name I think you have to change the name like if it was Jason we call it ghost Jason is that enough I don't know if it's enough or not but it's going to be a brave new world okay listen Akami it has been incredible to talk to you.

1:02:00Let's give her a big round of applause. Thank you. I'm very interested in your business, specifically buying factories. This sounds like a really, I think you should raise money to do a factory roll-up. Oh, so exciting. All right, we'll see you next time on This Week in Startups. Bye-bye.