 Apa yang kita akan lakukan sekarang adalah, jika sesiapa-siapa ada pertanyaan, boleh rasa bersyukur untuk melakukannya. Bersyukur untuk memperkenalkan diri sendiri. Beritahu kami, siapa anda, apakah anda dari mana, dan beritahu kami siapa-siapa yang menerima pertanyaan. Bagaimana untuk sesiapa-siapa yang diperkenalkan? Atau seperti sebuah perkara yang biasa. Sebelum kita mulai, kenapa tidak, saya akan menghubungi. Kita mulai dengan cara tidak serius. Bagaimana kita dapat tahu? Dari kamu. Apa pilihan kamu yang terkenal? Kenapa pilihan kamu yang terkenal? Kenapa dan bagaimana? Kamu tidak sepatutnya menguruskan diri sendiri. Sebenarnya, ia cukup mudah. Bagaimana dengan satu pilihan yang berlaku dan satu pilihan yang terkenal? Bersyukur, ia lebih mudah. Bersyukur, ia lebih mudah. Bersyukur, ia lebih mudah. Sebab itu semua orang yang terkenal. Bagaimana dengan satu pilihan yang berlaku? Yang terkenal adalah pilihan yang berlaku. Kenapa? Sebab saya rasa Greg dan Uber sangat berlaku. Saya sangat mengenai pilihan yang berlaku. Saya tidak menggunakan pilihan yang berlaku, saya menggunakan pilihan yang berlaku. Jadi saya rasa pilihan yang berlaku sangat bagus. Saya sangat membuat pilihan yang berlaku. Jika kamu ingin bercakap dengan saya. Bersyukur. Bersyukur. Bersyukur. Bersyukur. Bersyukur. Bersyukur. Bersyukur. Jadi itu pilihan yang berlaku. Pilihan yang berlaku. Bersyukur. Bersyukur. Bersyukur. Maksud anda bagus. I8. I8. Hapalapad I8. Hapalapad I8. 300k orang yang suka harga. Jadi sebabnya teman Instagram. Tapi mereka hanya menghubungkan permainan. Jadi itu sangat bagus. Jadi apa yang anda akan buat dengan harga? Bagaimana anda berdiscover? Itu yang bagus. Bagaimana kita boleh menjelaskan harga? media. BOTLIST.COE, jika anda ingin tahu BOTLIST.COE BOTLIST.COE, jadi itu dibuat oleh Venture B Kai yang benar-benar menguruskan segalanya di dunia Jadi ia bagus untuk kami juga Okey, baiklah Untuk teman-teman yang terkenal Sebenarnya, saya sangat suka Gov.SG Gov.SG SQ menguruskan yang mereka menguruskan untuk pemeriksaan sekarang Kenapa? Sebenarnya, apabila saya mengambil aplikasi kawasan atau mengambil aplikasi kawasan Saya berharap sesuatu yang akan sangat menakjubkan dan ia sesuatu yang tidak akan menunggu kamu tetapi apabila saya bercakap dengan Gov.SG saya merasa lebih bergantung daripada yang saya menunggu kerana ada banyak fungsi yang berbeza dan saya rasa mereka menghargai dengan sangat baik dan mereka membuatnya sangat menakjubkan dan anda dapat tahu segalanya yang anda akan tahu tentang kawasan Jadi, Gov.SG sangat bagus Untuk teman-teman yang terkenal saya, saya sangat suka Poncho Jadi Poncho adalah kata-kata dan Poncho akan beritahu anda tentang bagaimana kata-kata berada di tempat sekarang Saya selalu menggunakan Poncho dan beritahu saya apabila saya mengambil aplikasi kawasan dan Poncho selalu beritahu saya bagaimana kata-kata yang saya beritahu dan beritahu saya apabila saya mengambil aplikasi kawasan Saya sangat suka Baiklah Biar saya mulakan dengan teman-teman yang bergantung Saya rasa teman-teman yang bergantung Saya tidak boleh berkata-kata bahasa Gov.SG kerana saya akan bergantung Saya perlu berkata-kata bahasa Gov.SG Saya juga boleh berkata bahasa saya sendiri Saya rasa sekarang, kerana saya tidak pernah cuba bahasa Astro saya akan berkata apa yang saya suka tentang Astro dan saya akan berkata, tidak ada banyak bahasa di tempat yang bergantung Jadi, tolong lakukannya, lebih banyak bahasa Tetapi, saya rasa Astro sebenarnya sangat menarik untuk saya kerana apa yang saya suka tentang hubungan itu adalah bahawa mereka sangat mudah untuk anda untuk mengambil gambar dan tiba-tiba kategori yang sangat jelas dan bahawa mereka menggunakan sebuah pelajar-pelajar untuk kategori dan sebagainya dan ia mengingatkan saya sebuah konsep perniagaan perniagaan dengan bahasa Poncho jika anda melihatnya ia sangat menakjubkan untuk contoh perniagaan untuk dapat menggunakan itu dan saya rasa itu adalah perkara yang sangat menarik dan saya rasa saya suka adalah visualisasi bagaimana mereka menggunakan visualisasi atau perniagaan dan sebagainya dalam gambar tersebut saya rasa ia sangat mengingatkan dan itu yang saya suka dan kemudian segala keadaan tersebut saya akan menerimu sedikit saya tidak akan mengatakan bahasa yang bergantung kerana saya rasa ada banyak bahasa yang bergantung jadi saya tidak dapat memilih satu tetapi saya rasa satu yang saya benar-benar saya menggunakan setiap hari sekarang adalah Google Assistant di telefon saya dan juga di rumah saya jadi saya sangat suka itu dan saya menggunakan sebagai alam sebagai pelajar musik saya menggunakan sebagai cara untuk mengingatkan apa yang berlaku esok dan sebagainya jadi saya rasa jika anda melihat bot ia bukan hanya tentang chat ia juga tentang voz jadi saya rasa itu adalah yang menarik yang berlaku dan saya cuba menjaga saya dalam hubungan dengan itu okey bagus jadi sekarang kita melihat sedikit lebih luar sesiapa daripada permainan ada apa-apa pertanyaan untuk panel kita saya akan mengandalkan miket lagi, beritahu kamu dan tiba-tiba untuk tanya siapa yang beritahu kamu tiada sesiapa yang berkongsi dan beritahu kamu ok, tidak ada saya akan terus ia sebagainya seperti satu atau dua kali boleh saya tanya pertanyaan? boleh saya tanya pertanyaan? tanya saya pertanyaan tidak, saya sangat percaya ok, boleh saya tanya pertanyaan? mungkin kita akan pergi ke list saya dulu saya ada beberapa tahun yang telah dimakasih beberapa pertanyaan saya ada pertanyaan ini untuk kami pendapatan orang yang tidak dapat menikmati kami hari ini okey jadi Mungkin saya akan menyebabkan sebuah pertanyaan untuk Arby, itu okey. Jadi pertanyaan ini untuk seseorang yang mencari untuk membangunkan konsumersinya sendiri. Ia menyebabkan, saya tidak yakin bahawa ia menyebabkan sekarang. Dia ingin tahu, untuk pertanyaan ini menjadi sukses, kan? Bolehkah anda mempunyai mungkin tiga metrik sukses yang dia patut mencari dari hari ini? Okey, jadi tiga metrik sukses. Saya akan berkata, tentu-tentu... ...berkata-kata sebuah pertanyaan adalah sebuah metrik sukses. Jadi anda boleh sebenarnya membangunkan bot anda... ...semasa anda membangunkan user, semasa anda minta mereka berkata banyak perkara. Tapi anda mungkin juga membangunkan bot anda... ...yang tidak menggantikan banyak mesej dari user. Jika anda ingin membangunkan bot yang menganggap, jika user tanya satu pertanyaan kepada bot... ...dan bot beri jawatan unik, user beri jawatan 4 atau 5 pertanyaan yang unik... ...dan bot beri jawatan 4 atau 5 pertanyaan yang unik. Itu memberi anda idea bahawa user sebenarnya menganggap dalam hubungan. Untuk membangunkan bot yang menganggap, ia akan menjadi sebuah metrik sukses. Sebenarnya, sebuah perkara yang anda mungkin membangunkan adalah... ...jika bot anda membangunkan... ...di mana anda boleh membangun hanya dua atau tiga pertanyaan... ...tapi anda melihat bahawa user tanya 10 atau 20 pertanyaan. Ia juga bermaksud bahawa bot anda tidak membangun... ...sebab user membangunkan... ...sampai membangunkan bot yang sama. Itu tidak mungkin sebuah metrik sukses yang benar. Sebenarnya, saya akan kata jika user tanya lebih banyak pertanyaan... ...tidak menganggap metrik sukses. Apabila mereka menganggap bot, ia perlu menjadi cara yang berlainan. Sebenarnya, sebuah metrik sukses untuk bot... ...saya akan kata ratenya. Saya mencari ratenya sebagai... ...berapa kebaikan untuk user untuk bercakap dengan bot anda... ...selepas pertama kali mereka bercakap dengan bot anda. Jika anda menghubungi banyak platform analytics... ...atau platform bot analytics yang berlainan... ...setiap contoh adalah DashBot.io... ...setiap contoh adalah bot analytics... ...Bot Sanko menggunakan analytics Facebook. Mereka sebenarnya beri anda mesyuarat... ...berapa kebaikan untuk user bercakap dengan bot... ...selepas tiba-tiba. Anda perlu menjuruhkan... ...berapa kebaikan yang berlainan. Kemudian, ada satu lagi perjalanan. Jika bot anda membangun user... ...sepanjang hari... ...saya akan membangun metric ini... ...dan ia akan membuat anda rasa baik... ...tapi sebenarnya, bot ini hanya menggunakan pakaian. Jadi, anda harus melihat perhatian itu... ...untuk ratenya. Akhirnya... ...mungkin sebarang kebaikan... ...selepas bot itu juga adalah metric berjaya. Kerana bot ini adalah platform baru... ...jika anda melihat... ...berapa user berkongsi dengan bot... ...saya menunjukkan... ...mereka berminat... ...dan mereka tidak akan beritahu teman tentang bot. Mereka menguatkan teman atau teman... ...tapi... ...bila berkongsi dengan bot anda... ...mereka mengunakan pakaian dari bot anda. Pakaian berkata, ya, kami menggelakkan dengan anda. Bots adalah pakaian yang baik... ...dan kami harus mengumpul pakaian. Mungkin saya akan menonjolkan satu lagi... ...part 2. Anda melihat pakaian... ...bukan mengikuti user terlalu banyak. Tapi di masa yang sama, jika anda tidak melihat mereka... ...pengalaman anda, ia akan menyerahkan mereka. So, how do you balance between not over spamming, number one, and number two, making sure that whoever is using a bot is still engaged. So, I think what we normally call as spam is a message that's very generic. So, we get a lot of spam emails, but when you read the spam emails, the most personal that these emails can get is hey, first name, right? So, the thing is when you read a message from a bot like hey, first name, and a long paragraph, you know it's spam. You know it's something that they probably send it to everyone out there. But if you want to actually send your messages to your users, you should be able to customize the message to them in such a way that you tell them, I know things about you based on what you told me before and I'm sending you this message because it might be of use to you. I know that you like A and I know that you told me B before, why don't you go try this? So, that would actually make the user feel more like comforted with a bot. They won't feel like it's spam anymore. So, that's one way you can actually spam your user and it's safe. Say spamming? Why? Okay, thanks Abhi. Anyone has a question? Alright. Okay, I will go to both of you. Okay, I'll pass it to Mike. Actually, so, when people use normal websites or apps, Actually, this question is directed at everyone actually. So, when people use normal websites or apps, if there's a particular bug on a screen, we would get like a crash log or a crash report where we get to know where the problem is. But what sort of strategies do you guys use for detecting bugs in our bot conversations that you guys are having? What if a person goes in an endless loop where the same question keeps coming up again and again? Or if a person drops off a conversation because they didn't quite get what they expected? So, how do you guys go about detecting this and taking action on it? Your name? Yes, my name is Suresh. I'm an engineer here at 99.co, the property portal. Do you all need a bot? Do you all need a chat bot? Not at this point. There's a Spencer. Saya akan cuba. Okey. Untuk kita, kita mulakan dengan pilihan besar. Pilihan ux dengan perkara yang sangat penting. Dalam pilihan website, kita juga mempunyai pilihan yang tidak berlain. Kemudian kita mempunyai pilihan pilihan yang tidak berlain. Jadi, apa yang berlain dengan perkara yang tidak berlain? Pilihan chat bot kita juga cuba menggunakan pilihan yang masih baru untuk semua orang. Abi beritahu bahawa ada beberapa periode berlainan. Jadi kita perlu mengalami periode berlainan dengan baik. Untuk memastikan apabila ia berlainan, bagaimana kita boleh beritahu mereka. Jadi ia berlainan untuk menarik semua periode. Sebenarnya, metodologi adalah sedap untuk membangun web-side periode. Biasanya periode berlainan yang mencari masalah. Jadi ia berlainan untuk menerima bagaimana yang berlainan, meledik ux, lalu cuba periode periode periode yang berlainan. Anda menerima bagaimana jika itu berlainan, anda ada periode berlainan kembali ke periode periode. Tapi jika ia berlainan, anda perlu meminta periode periode yang berlainan. Semua orang yang membangun periode berlainan lebih lama. Tetapi saya rasa itu adalah tentang membangun periode pertama. Sebelum anda membangun periode periode periode, ux, lalu cuba periode periode periode yang berlainan. Untuk kita, saya tidak bersyukur dengan periode periode periode periode yang berlainan. Sebenarnya, banyak kerja kita telah dibuat. Tapi mungkin saya boleh bercakap lebih banyak tentang perkara yang berlainan. ağ scientifically, semacam itu apabila anda steak strip sole, mereng injected, pun juga menyebabkan bahawa yangkapK нашiran, awak perlu mengembang-mem computed di sekian bagi jad1 ticks. Salah satu perkara yang diperlukan contra eats thé comma sebuah pengguna behalf, ada kemah Babakур. Enggak säk��,eticFrankPengил memberikanErut. bagaimanapun mereka berhenti memperkenalkan bot. Kami perkenalkan bahawa banyak pengguna ini, saya boleh kata, 50%-60% pengguna memperkenalkan bot setelah setiap poin. Dan ia menjadi kelihatan ke kamu, kemudian anda mungkin mahu melihat apa itu? Mungkin anda boleh memperkenalkan mereka. Mungkin ia mungkin begitu. Atau selepas ia terlalu diperkenalkan, kemudian anda mungkin mahu memikirkan mungkin memperkenalkan sesuatu, kemudian kemudian kemudian perkenalkan, kemudian sesuatu seperti itu. Atau anda akan membuat konten lebih tinggi dan sebagainya. Jadi, saya rasa yang terakhir adalah perkara yang sangat menarik. Seorang kawan kami yang memperkenalkan bot setelah setiap poin, mereka sebenarnya memperkenalkan bot setelah setiap poin sebagainya. Dan perkara yang anda boleh lakukan adalah perkenalkan bot dan sebagainya. Saya rasa anda boleh memperkenalkan beberapa skrips dan sebagainya untuk memperkenalkan beberapa perkara ini. Jadi, perkara yang lebih otomatis, perkara yang lebih otomatis, ia akan sangat menarik. Kemudian, saya rasa, sebelum itu, saya mungkin menggunakan perkara sekarang, dan perkara lain yang saya telah lakukan. Tapi tentu saja, penggunaan penggunaan penggunaan menggunakan, menggunakan penggunaan yang sangat agresif untuk melakukan secara indah. Maksud saya, membuat penggunaan lebih banyak untuk mencoba dan kira-kira, bahawa kadang-kadang, memperkenalkan penggunaan awal, saya rasa itu akan sangat membantu untuk anda mencoba mencoba mereka. Tapi, pada hari ini, apabila seseorang bercakap dengan seseorang, jika anda berada di dalam komposisi, dalam peribadi dan anda mendengar mereka, anda tidak tahu dari siapa yang salah. Dan kadang-kadang, bagaimanapun anda boleh mencoba skrips dan segala-galanya, memperkenalkan semua ini. Tapi perkara terbaik, kadang-kadang, ia hanya mencari berbicara. Ya, dan itu sesuatu yang, pada hari awal, apabila anda memperkenalkan skrips dan segala-galanya, ia sangat penting. Jadi, jika anda melihat bahawa bahagian bahagian bahagian itu, ialah, maaf, saya tidak faham. Jadi, sepanjang bahagian bahagian itu, apabila ia tidak faham, bahagian bahagian anda menulis beberapa cara berlaku dengan berlaku menghubungi meskipun di tepuk. Anda akan selalu memiliki sesuatu yang adalah kelembapan, dengan sebarang-barang kelembapan. Kemudian, jika anda memotong pengetahuan oleh teman-teman, anda akan melihat bahawa banyak kelembapan untuk kelembapan, lebih daripada kelembapan yang penting. Anda tahu bahawa sesuatu yang salah. Anda tahu bahawa ia adalah tentang, atau bahawa seorang berlaku memperkenalkan ia tidak memperkenalkan, dan ia hanya memperkenalkan kelembapan. So what I try my best to do is reduce the fallback rate. Like make sure the button, I mean minimize the rate at which the bot says sorry I don't understand. So the next person is Chris Wright, you got a question? I don't remember all the names in front so I apologize. I came in pretty late and I wasn't really paying attention to repeat the names again. But it's directed to all three of you. I work in the VR virtual space. I just invested in a virtual business. So it's not really quite related what you guys do. But my concern is security related. One is security and reliability concerns that keep you guys up at night. And I want three from each of you. Because one is a cop out, two is an excuse. I want three. You guys want me to start, right? Security concern. First concern I'm afraid of is that any of one of us get hit by a bus. Or our laptops are hijacked. Or we access an internet that is stooped on by someone. I mean, these are all practical things, right? Because at the end of the day when we work, if you are a small team and you work with enterprises, you are the biggest security miss to the enterprise when you work with them. And which is why they ask you this question, how many people do you have? And if you have less than, for example, 10 people they will look at you and you're like, are you sure you can support our needs? And they use this qualification question to essentially ask you to go with it. It's easy for a company to do that, which is why I think the easier part is, the easier part, not the easy part, it's very hard. But the easier part is let's not do B2B let's do consumer focused stuff. I think for us, we like hammering our heads against these kinds of problems, so we kind of overcome that. So that's what keeps me on the internet, actually. But if customers come to me, if every customer today comes to me and tell me, you guys are the security risk, we're not going to work with you, then I need to solve that. The second thing is actually, I think maybe on the infrastructure side, if we work with companies that are very prominent, the problem is that that itself invites security risk. Because people all over the world are being paid, handsome as money to crack open these enterprises and essentially try to school over with their business. And if they do that, it's a batch of honour. So we are also a little bit careful about announcing who we work with sometimes because I think that's something that might be a bit sensitive and we don't want to become a target of ourselves. So recently, for example, one of our previous products, we actually became a target of a guy who just started doing lots of, he wrote a couple of scripts to attack us for no reason and essentially that overloaded our email servers quite a bit. So we have to essentially do a lot of countermeasures and stuff. So these kind of things do happen and we are wary of that. I think the third thing is that for POS, right now Facebook and all these platforms do a pretty good job already. So for example, people cannot really abuse the bot too much. If someone tried to, for example, bomb the bot maybe like 10 million conversations every second, Facebook will greatly limit the user and immediately do that for us. So thanks to Facebook, that's good. But when we launch like for example, companies websites for application and stuff, then we need to make sure that we handle some of these things ourselves or we'll give the party to do that. So that's something that keeps my team up at night when we team up with this. So Spencer has answered like security risks on B2B point of view where it builds bots for other people but I don't have any of those answers because I'm not do bot spot companies. So for security with risks for building your own bot or at least other developers need to look at. So you could repeat that? Or at least what do other developers need to look at and think about anything else. If that's not the kind of experience that I've got. So in my presentation, I had shown like the three main parts of what I think it takes to build a bot. The first part is the messenger platform. The first part is the messaging platform. That's the part that's the only interface to the user. And the other two are the web server and machine learning platform but that happens in the background. That's in the back end and it's not interface to the user. The biggest security risk is any part of your app or any part of your bot that is interface with a user because that's a point of entry for the user. Luckily if you use a very popular messaging platform like Facebook, Facebook shields you. They shield you from any kind of attacks that users may try to do in your bot. Previous, I mean previously if you didn't have Facebook and you just accepted plain text from users, they could always do some kind of SQL injections or these kinds of things that will trigger this malicious stress to your bot but Facebook shields you. I'm pretty confident that even Slack as a platform will also shield you in that way and so do all the other messaging platforms. You just want to ensure that the points of entry to your bot are just limited and only Facebook knows that single point of entry. Okay, maybe I can share a bit about ethical hacking since that's what I've been doing for quite some time before we do the business. So in the website world I think it works similar as well. So sometimes our website gets hit by some random hacker in Morocco or Pakistan and our web pages just change. So it's a real problem that happens too. I think it will happen to bot as well. So last time when we did it for those of us in the IRC generation if you remember MRIC there was this group called ethical hackers and then the idol there was like Sean Parker the founder of Napster so he dropped in once in a while to answer a message but you know this game called Neopets Neopets, you're so old. It sucks out a lot of our time but we wrote a lot of script to get more Neopoints but how we do that is we go to other people's account and get their Neopoints because we can't hack their algorithm to increase our own Neopoints on a passive basis so we're gonna get it. So we found it more fun than playing the game Neopets itself. So if Abbie is my friend I'll just take all his Neopoints and then we'll get more fun doing that and technically at that period of time web security was really low it's easy to write scripts to do that so it came upon our time whereby people start to people start to hire people to prevent people like us so what happens is we have to write a script to prevent people to prevent us so it stacks up and that's how technology gets better so the guy that's trying to prevent me from he has to write a script that directs me away to another IP so I see like a fake Neopets site even if I put Neopets.com so he directed me away and then the real user goes to this account so they track by like ID tracking so it comes to upon whereby there's 3-4 layers of ethical hacking such that I can't be bothered anymore because it's too deep to crack after that but I believe there's still people doing it so that's one way that we can counter it so all the ethical hackers they tend to write scripts that are like 3-4 layers deep meaning if you can hack what you're gonna do so it's like when you plan the it's a UX for hackers so you plan if this word is gonna hack what happens next but if it didn't break through okay good so you go back to your normal state then next one if it breaks through the hacking thing what happens after that so if I break through to its new points are you gonna stop me from taking the new points or are you gonna boot me out of the system such that I log out but if you still can crack that then it goes to the next layer so ethical hackers need at least 3 layers so on a context business in business not on giving anymore so we work with GIC so this part we can share no worries so how we counter this thing is we try to do it in such a way that even if they manage to hack our board they can't extract any important data so we always we advise people to come to us or like companies that want to buy services try to try to segregate as many parts as possible so some coders they write a really fast code like one file has everything like the whole file.js or the whole file.py so the entire the entire system runs on one screen is it so when they can extract anything from their screen it becomes a problem so I can change like SG Bus Uncle SG Bus Aunty and I can give you the wrong location to make you pissed off and I can do this but I think probably RB runs it in different layers so he separates the thing out there such that even if the first layer gets hacked there's still a second layer there's a third layer so that's what the way we do it and then we do backup servers for example like if information number one gets hacked but information number one is really important and it will bounce to the second database of information one so exactly the same thing it's like it's copied but such that if it reads a malicious threat it goes to the next one straight away and it reads another malicious threat it goes to the next one so you have to do a lot of stacking and the the functional thing is big companies pay for that so they are willing to pay money to counter this thing and same thing it's also three layers so any ethical hacker will also tell you anything three layers because the third layer yeah because we don't make any money we just hack for fun previously okay thank you any other questions guys? yes my name is Max I'm the CTO of Q Reply so I just want to add two more points to the previous discussion about security of bots I think there are two main points first is the API that you use to pass them the user's message so for us we couldn't use anything like with AI or API.to act because we just cannot send the user information to external parties because usually if you read clearly the WEAI contracts and their terms and conditions they actually say they have ownership of your any messages sent to them so they can really if they want to so that that alone just doesn't fly with any enterprise just doesn't work so that's why we have to develop our in-house NLP processing to understand all the messages and intents and entity that the third party are supposed to do and this allows us to put it into the private deploy private cloud so in case that the company just super sensitive with all this we can actually put it into their computers and their servers the second point is about Facebook Messenger platform so most of the libraries that you see on Github today doesn't actually protect you from the denial of service so a hacker can once they figure out that your server response there are ways to figure out once they figure it out they can spoof as a user and send like minutes of messages directly to your back-end server for example your Heroku server or Amazon server but at that point you can't just you have to do some verification to check that this actually came from Facebook otherwise you'd be wasting your API calls to or your other services and then your your users couldn't get any message through so the way to do it is there's a signature that Facebook sends called xhub xhub signature and it's a shaman algorithm so your Node.js code usually you have to implement it yourself or you have to be very sure to check which open source library you actually use and whether they implemented that authentication if you don't anyone can just print out your bot instead so if you want to know which how to exactly implement that you can kind of ask me later I can show you the code Thanks Max for anyone else any questions or a wrap Thanks My name is Kiha I'm working at the insurance as a data analytics manager so this year it's going to be digitalization of insurance companies like insurance company is a traditional quite traditional but they know there's time to change so we got thinking bringing the chatbots in two ways one is customer service the other is finding our prospects many base movie Korea is Kakakoto I think it's quite popular platform equivalent of WhatsApp and Singapore the 40 customer service I think if we bring the chatbot there must be a lot so different there is freedom of questions like how to claim what's the claim for sales or what's how I have a new phone number I like to change the phone number there must be a lot of freedom of questions I think it might be quite difficult to chat about the play because it's a lot so I don't understand but acquisitions I think it's the customer asking pretty much there's expected you know I like to know the premium amount my age is blah blah and what's the best product I think it's more expected closed questions for the customer service and cost saving it's great you don't understand how many calls in a day and dealing with those simple inquiries and however I'm worried about the you know lots of loves regarding many different types of questions and which aspect do you think the first thing we should consider to show tablets or a team for insurance prospecting acquiring prospect customers or dealing with customer service complaints and then asking as inquiries Do you have a FAQ page? I believe so but I believe so Okay Your company has developers, right? Has developers? Actually I went to Korea to review some vendors we don't have I mean when you said anyone who can do no JS just JS or even PHP because the solution there must be yeah, there must be IT If there's an FAQ page is I mean we are at Microsoft so I've got to promote you on Microsoft right but the Q&A maker.ai is really good Q&A maker.ai so how it works is it scripts the page the FAQ thing then it just you can train the bot inside the system as well so it scripts the page for example how much is your your policy or like how to make claims so you have a certain like standard protocol you can copy the whole thing to the bot just like that then you can start training bot such that you can respond to different questions that mean the same thing like how much is my premium or how much how much am I expected to pay so you can start training and putting as many questions as you want so what you're saying is customer service it can work for customer service the lowest barrier of entry for us in sales I mean like where we sell to the companies usually people look for FAQ bots because they really got a certain infrastructure in place like FAQ questions that they want to just plug and play to reduce the time the thing is the insurance in a conservative people are really conservative you know they really worried about those fall out I don't know if there's a lot then by detrimental to the company reputation it's going to be also detrimental to my reputation as well if I bring in that sort of you can trust Spencer Norris okay sudden answer right on behalf of Spencer okay and another thing we can do right okay IBM so there's something good you can try to retrieve and rank retrieve and rank so what we did is you know carousel rank so carousel they have FAQ page but it's quite complicated it's all the languages all the because they are big company already so people start to ask a lot of things so retrieve and rank can work such that it will read the whole FAQ then it ranks for example I ask like how to sell stuff on carousel so it give you the most important most relevant question it reads on the database then it issues back the correct thing so that actually increases or decreases the chance of screwing up in that sense so if you're worried about the bot now responding to give you the weird answers then they actually can help so but I think it's about exploring across the board because every big company now they are the way they view API so fast the updates come so very fast so if you have a developer and your end is even better to try to explore yeah because I think the retrieve and rank just updated like yesterday when I just checked it as well Tuan Tuan Tuan Tuan so maybe I can start by saying that we looked into Korean language a little bit we have a couple of friends in Korea we focus on NLP API we work to use one of their pretty good companies co-no the founder the founder of ours so I understand that in Korean language the challenge what you talked about for insurance company implementation for chatbots even I think we won't bendo for you guys it's because we really have zero experience with Korean Korean language it's very different from English and it's arguably actually it's easier or more difficult English so we actually look into it so it's actually quite difficult for us to just say that we want to pop over the language so that's also the problem so for English if we train it to a sufficient level it's actually possible to exhaust most of the possibilities you know using methods and using training of like customer ticket and so on so I think what you talked about support and you know proactive engagement personally I think I like to focus on proactive engagement but the problem you're doing that is that when you do proactive engagement on bots everyone is going to expect support as well so for example if you launch an experience like oh you know it's a sales bot or something everyone's going to ask what's the weather or all this kind of weird things and of course they're not supposed to and the thing is that if you're not ready to do that then you're definitely going to get all these questions so even if you do proactive engagement you're still going to get all these reactive questions that you need to handle so that's why that's also the risk of launching so you rather not launch than to launch something that has that so that's why when we go have enterprises we really make sure that we understand who are the stakeholders that you're working with as well is this the CIO level thing is it the CMO level thing or is it like for example the internal team doing this and you are putting your company at risk then the conversation will not happen it's not going to happen so that's why from my perspective for enterprise sales it's my job to figure out with you how level of buy-win do we have how can we actually make this happen is even technologically feasible for what you're looking for and then what are the steps that we can use our platform to support you on that and then are you going to do RFP are you going to invite Watson are you going to invite Microsoft and all these guys or are you going to take one smaller than 1,000 to the play as well and are you thinking about in the 3 months 6 months or more sort of a pilot perspective and we don't do free pilot so companies some companies do free pilot some companies don't we don't do free pilot so because of that there's going to be some consideration down the road and having to chat offline if you want to have a meeting about this can I one more questions we are thinking in Q1 Q2 person in the chat board from risky business we're thinking in our hireings of chatting Asia real human person to deal with customer complaints to be going to accumulate some text histories to build better chat boards do you think it's the right way or I don't have to hiring the human person to chat to get some more knowledge in a conversation interaction between customers and agents so I can give you an answer based on user experience point of view so if you want to build a bot that does customer service of course in the start you have to give you have to train the bot with a list of questions and a list of all the answers and the thing is over here you have a fixed set of answers but questions but there are many different ways of asking the same questions so you have to train your bot in the start for at least a few different ways of asking the same question and then when you deploy it live I wouldn't say you actually need to hire someone to just monitor this instead I would say it would be best if you kept your customer your old customer service people like don't fire them yet but let them just monitor how the bots I mean the users are speaking to the bots and they might actually give you a lot more insight into the different things that users might mean when a bot answers them so the thing is customer service agent is a person and person and your user is also another person and when two people speak there's a lot of information that goes from person one to person two and a lot of miscommunication might happen but for a bot there's always fixed answers and if a bot gives your user a fixed answer and the user says oh now I understand the customer service agent was monitoring your bot actually would know that okay so that's what the person meant so definitely keep your customer service agent first like let them monitor it and then later you can optimize the flow okay thank you okay guys we running short of time I know there's a lot of body questions for everybody what we will do now is that please join me to say thank you to our panelists get that round of applause