Exploring Bot Frameworks and Modules for Enterprises with Stephane Nguyen: Chatbot Conference 2018
Transcript Summery:
All right, how are you guys? Good. How's the digestion coming along? Okay, fine asleep yet. No, okay, so we are assist we're in Enterprise Commerce platform on messaging and voice. Right and I'm Stefan had a product design there. I wanted to share a little bit with your use case that is really dear to us. Also, I'm from France. So if you don't understand my English accent, please give me a hard time. I'll be happy to switch in French for that presentation.
That way it's going to be easier for everyone. So I wanted to share something on our decision-making for how we build conversation. And also how do we select use cases and customers and how we work with customers around this industry and for us is really around three things. And like the panel earlier with the ecosystem panel with Lauren from Pain robots and offer for from Google the use case. The first one is really much the use case. We're not trying to boil the ocean here. We're trying to do something simple that people can relate to and if you're trying to say that you can do everything you are full of shit and we're trying to be less full of shit than you guys and we like to think about the use case and how we select use cases like the ATM to the bank in 1961 the ATM got released and everybody's was scared because it would kill some jobs in the bank and kill the teller and the reason why the ATM's is still around is because it can do something simple and only something simple but very well, right. So at the beginning you could withdraw money for $20 that therefore the teller cannot does it doesn't have to do that and nobody wants to do that anyway, right and as the ATM got better.
You can now withdraw money. You can deposit money. You can check balance due transfer, etc. Etc. So we like to think how we select use cases the same way right the second pillar to selecting a customer and selecting it a success story is the platform. We believe strongly in selecting the platform where your customers are already, right? Nobody wants to invest in something. Nobody wants to use right and so we are all on our phones already using either an iOS device when Android device right? And so this is really important for us. And the third one is the point of entry.
We want to remove friction as much as possible when it comes to our success stories. I'm going to let you sit and look at this. Nobody wants to download an app and sign up and then enter your credit card until your email then open your email confirm that email and then go back to the app and start using it. And so we're really pleased and excited to announce that we created the first. Automated fully automated experience on iMessage using Apple Business Jet. And guess what the use case is ordering beer and just ordering beer not ordering water not ordering fries not ordering nachos, but it's ordering beer. So we partnered with apple and our Mark and our Mark is the biggest supplier for food and beverage in stadiums. So they have 94 stadiums across the US.
Across four major leagues, right NFL NHL NBA and I'm French. I don't know the other one. But so we wanted to Pilot it around one specific use case, which is be ordering. And of obviously reordering so the first thing is removing. So for the reducing friction over there. The first thing that we wanted to do is make sure that we can reduce friction. So people can try it. Right and so the first thing is as you know on iOS since iOS 11 your camera is a QR code reader.
So what we decided is to put a QR code and when we said put a QR code, we literally said put a QR code everywhere. So we started on chairs. On diners and even on Our Friend Martin over there and you can see Martin in the back had a QR code at you it was walking around like with a duck tail like with people like doing this and trying to follow him all along just to try and so this was reducing friction. And so once you scan that QR code, what is it? What does it do? Well, it's literally Jumping in your iMessage. And this is the actual experience so you could pick a beer once you pick a beer it will ask you for the section. And nobody wants to go and get up and go in line toward a beer as $19 beer when you can actually pay $19 with saying in your seat, right so you can pick a tip and then you have Apple pay which is already in your phone already configured and your card is already there. So by removing friction here, we increase the cell the cells so and tada A guy would literally walk up to your seat and deliver that beer and on top of this because you already ordered once you already gave you're sitting number. It would be just a one top reordering and funny enough.
We had an immense amount of ordering and I sorry I cannot share the numbers. However, he was really high and it was really surprising even to us and because also we were literally behind the scenes with the our Mark team and our team we realized that because we're using a medium where people ask and type We realized that blue moon was not a beer that we would sell but everyone was asking for it. So we quickly changed to Blue Moon and he turned out to be the most selling beer in the stadium that day. And so when he comes to this, I want to mention something about iteration, right? We didn't plan to sell blue moon. But because we change right away on the spot. It turned out to be a great pilot. So it's really important for us to take a look at errors because we think the errors are the intelligence. So everything is behind her errors, especially when you have an open tight medium such as messaging.
So in here the seaside number was so high. I got mentioned in the apollonian calls with the CFO Luca maestri and also learning calls for our Mark. You can go and check it out if you if you want and so we really thought that by having a simple use case on a platform people love and reduce reducing friction. It will be easier and this is what I wanted to share. So now I wanted to give you a sneak peek about how it works and behind the scene right? When you think about receiving that particular message A lot of people are saying well we are going to build a tree right like a lot of people here are building trees in a nonlinear way and it turns out that if you want to change your mind. Or change your beer. You cannot do that without having to go back and nobody likes ivr systems. I don't think I maybe you guys do but I don't like to type one to go back to the menu, etc.
Etc is just like as a conversation to a friend. I just want to continue the conversation and say hey, by the way that other topic that we talked about. I'm going to change that and then go back to what I was doing. Right and so understanding how we build that conversational layer for us is really important. And so when we started five years ago we decided that at Reflow was the way to go. How hard can this be? Right if you go, yes you go here. If you go know you can hear but when you look at the conversation with your friend, this is not how the conversation goes and especially with your significant other you're having a fight with right. It's all available place.
So building true for in the first place. If you open the tool that you're using right now and you see this you probably not on the right path you are on the right path for something that is very static. There is very straightforward. But as soon as you get into changing your decision, this is not how we think they could work. So, how do we think about it? Well when you start ordering a beer and when you interested in ordering a beer, basically we look at not steps and flows, but we're looking at information you need to order beer. What are those information? Well, you need a beer. You’re seating chart and then maybe your tip. Right and here you have an arrow here that says well if you if you answering one by one, I'm going to answer the beer first you're sitting and then add the tip.
But if I say, hey, I want to order a beer and I want to put 10% tip. They should work to even though I'm asking for the beer first, right? And so as you collecting those information one by one, right and then the last one you can finally say, okay. I have all the information I need in order to confirm and pay with Apple pay. It's not about treefles questions, question 1 then answer a question to answer a question three answer it it's about what are the all the information you need to complete what you're here to do. Does that make sense? All right, so in here the seating though sitting might be a little bit complicated because a seating is also another set of information when you are doing an address. You have a street named a street number you have a zip code you have a city same thing for seating when you have a seating you need a section a row and a c and the reason why we create that decision on its own as a decision cluster or decision-making framework here is because maybe we can reuse it somewhere else in this case. When we have a help intent here and we say hey, by the way, you can also order but we can get help as well. Well to get help you need to State the problem.
And also give me your seating chart and so here we can reuse this because you already actually you already ordered your beer. So we already know you already know your information. Is that all make sense? So once again to us how we select customer and how we think about use cases revolves around decisions and making sure that we don't build tree flows. Right. So use case platform and also easy point of entry and so I This slide for what I'm talking about now, but I really I mentioned it a little bit earlier about errors. We are in the medium. We using a medium where people actually tell us what they want. It's not about positioning this button that goes to the next page because you think it was the best position for that button.
It doesn't work like that in this medium in messaging errors are the intelligence errors are telling you what you should build next if you want blue moon and 90 people 90% of people wants Blue Moon. It will not work the way we build it, but we looked at errors and we turned out that blue moon was the most popular beer that people ask same thing. If you want to say I want to cancel my order. I want to cancel my beer order and we look at errors. And this is where what he tells us to do. So it's a sort of double edged sword because a lot of people are using old Paradigm to build a conversation meaning that they're measuring a funnel. They're measuring a conversion rate in that particular flow, but it doesn't work like that when it comes to errors are where you understand what to build next and if people are saying I want a refund more than once then this is what you should build in terms of use case. Right, so that's it for me.
I have if you have any questions, I'll be happy to answer it. Who has whoa? Not my fault. Hold on. Who has the first question? I don't but he does. All right. Now is your time to think of the questions? All right. Oh, yeah. I almost got one here.
You've got one. Thank you for the great content. So trees and flowcharts are bad and states are good. That's what I I'm not saying they're bad necessarily. I think if you want to have flexibility in your experiences, you should not explore tree flows. What you should do is looking and all the information that you actually need to complete what you're trying to do and then regardless of the order all you need to do is to collect them regardless of the order. It's not about question one level one of that tree flow. I'm asking for the city and if you say anything but a city it won't work.
All right. So let's say in your space model right where you have space model and your knowledge model and your eyes space model. You have all the states that you have to react to you're saying that you're suggesting that it's better not to use trees. But instead being flexible to hop from one state to another what is the information architecture to comply with this idea? Great, so, it's good cash question the way we see we the way we classify. Our information is we start with a simple use case. This use case needs a certain type of data set right understanding what a section means what all the abbreviation of that section all those taxonomy or what we call branded learning. So everything without that use case that you want to build. Everything else is considered an error, right? It's consider something that you cannot understand or you cannot handle or you understood but cannot handle yet this data set now we have to find a way to look at this data set and slowly adding it.
To the data set that we actually are you are using and so by doing this. Buy one and a not in a let's boil the ocean and trying to do everything. We're trying to be very conscious about. What is it take to add that particular error to that flow? So be very nice. All right. Thank you. Thanks.
Who wants a baton? Who wants to make a decision to ask a question? Alright, here's another question. How many decisions do you think a toddler who seven-year-old who seven years old makes? Lower I would say twelve hundred thousand. Sorry. She beat you to it. Yeah, I know crazy. These people are fast. All right, any other questions about decision making anybody have a decision in their life that they're not sure what to make and you'd like to talk about it here in front of everybody. I think we can crowdsource the botanist personal struggles.
Yes. Alright surgeons talk. Tell me about the question upstairs just you know, but he has a question or are you going to talk with a struggle or question? Okay, good. Here we go. Let's just go with it. Okay, let's go with it. We're live on TV. Here we go.
Where should I draw the line? When I ordered beers three four, I'll tell you from personal experience. I it only takes me one beer to get drunk. It's just I'm not sure if it's the fifth or sixth so go with your gut when your friends drag you towards the exit. That's the time to stop drinking. Okay, I've done and I'm from friends. So I drink wine anyway, so it doesn't matter for me.