How to Build a Powerful Chatbot in 5 Minutes with SAP Conversational AI [+Demos], SAP TechEd Lecture

good morning my name is Anil Kumar's and I'm a development manager for conversation hi and today we're talking about si peace solution conversation AI and how you can build powerful conversation interfaces with the solution so as ap conversation is si peace platform to build conversational interfaces and chat BOTS why is the city investing in a space and I should you consider conversation and use cases and how you can engage through chat pods with your customers and employees well let's start with a short survey just think about what do you think how many or what percentage of consumers use conversation interfaces just last year you can think of a number just keep it in mind and you can compare how close you are in a second in fact it's 67% and I'm certain that it's going to be much larger this year so there's no question the world is turning conversational and I guess many of you have used conversational interfaces in the past many of us have digital assistants at home Alexa Google assistant Siri use it almost every day to have easy access to information or even automate certain tasks like switching on the lights at the same time they're using communication tools like facebook Messenger Apple Messenger Apple messages whatsapp to communicate with our friends and colleagues and family everyday and as these tools get one more get a part of our life we're expecting the same ease of access to information and ease of access to help and collaboration from enterprises so now it's a great time to start looking into conversational use cases in your business and then doing that we suggest that you look at your two most important stakeholders which is our CSI as I mentioned your customers and your employees so today you're going to have a look at both employee experience and customer experience and how this can be amended and enhanced using conversation interfaces I'm going to show a demo very actually going to build a bot together on the conversation high platform which is pretty cool and I'm going to give a quick outlook into what features are to come so let's start with the employee experience let's do a short show hands here whoever thinks that enterprise applications are always easy and straightforward to use no not really so Jerry does shows events here in the room that's right business processes are inherently complex and enterprise applications need these forms in order to enter information you have these different systems in order to cover your whole business processes and to end you might start out as in the process with your Esfahani system but then want to create lead and your CRM system you're integrating with your success factors vehicle as arriba concur every day and that obviously doesn't make well that requires a certain level of training you have these system boundaries you have to get used to these different systems so that obviously doesn't make for the best employee experience what's the result well for one your employees I can easily get frustrated by the complexity that they have to manage every day they don't live up to their full potential so I see that their productivity equipment could be much higher if we could solve this challenge and it can also lead to an hasp an image of you as the employer we believe that chatbots are powerful way to overcome these challenges and that's why I say P is building chatbots as we speak into almost all of our big solutions so we're working with the s4 Hannity in the working with the CFO Hana team concrete success factors who will build these conversation interfaces into the applications to make it easier for users to interact 50 systems using natural language a chat pod can help the user to navigate to the right page that's a very simple use case you don't need to start with automation it can just be hey I got a question where do I find the right information can be information about the products can be information about elite when I'm a sales rep on the road and I'll quickly want to get information about a certain order I can be navigated to the right place where I can find this information at the same time buts are very good and answering simple questions or fa Q's so think of these policy documents that you might have or a use case that probably many of us have here is how much can i expense per day while I take it you you might already have an FAQ for that you can you could feed it into a chat pod and it can answer these questions without your employees needing to know where to exactly find this policy document and our Sea Shepherd's can execute and automate simple tasks so as mentioned as ap is building chat pods in their solutions and in the same way that you can expect your s4c for systems to be customizable to your specific business needs you can also expect in the future to be able to augment and customize the chat pods that we're building and then there's another dimension but just as a peace copilot or our digital assistant and this is actually very very powerful in what it can do for your employees so it can be the single point of access to a variety of conversational use cases so that you really just have this one entry point even your your launch pad or on an iOS application they can access these skills or use cases that have been built on top of conversation of our conversation high platform and I wanna show quick demo it's a vision video how this could look like and how our digital assistant can span boundaries of different systems and really bring entry and coverage into business processes for this we're going to look at the total Berg first minute I talked about first management it's not important that you can actually read what's here on the screen if you can that's great but I'm also going to talk talk over did the scenario that we're looking into here is Java which he's a marketing manager and he wants obviously to make sure that the marketing campaigns are running good and that the revenue forecast is fitting and we're going to see tom who is a marketing coordinator who wants to get on board it in the company so let's roll the video yeah we have Giada who doesn't quick check for the revenue projects and projections for Canada unfortunately she sees okay so we see that the protections are off from from the actual goal and it reminds her hey didn't we have a marketing coordinator position that was to be filled and we're asking success factors for that position and we see that it's actually open already for two months so it seems like we're having challenges in filling this position but we do have an immediate need to to act so we create a contingent worker request in a few class to quickly get someone from a third party so right here we can create this contingent worker request in Fiat class and then after some time Giada comes back and as hey do I have pending approvals and we also see different improved coming from different systems here but we also have this contingent worker offer here which we could approve right in the application so now we are Tom we had our in initial conversation with Giada I'm one day out of joining the company and I want to just want to make sure that I have a good start so I can I can ask for information about the company right here from the digital system again and I can have another look at the welcome message that was sent through success factors you Giada is asking me to already go ahead and order my laptop and she's also asking me to book a flight to the project kickoff so what Tom can do is ask for the Mac box and the catalog and this will be pulled from the arriba buying catalog they can also see that the get a list of available Mac books and I can drill deeper into the details to make an informed buying decision so it meant in this natural language experience through user user interface elements here right here I can create a purchase order or a purchase requisition in the s4 system again and once this is done I can go ahead and put my flight tom is already providing information here he wants to go from San Francisco to Boston and he's already giving the dates so that concur can pull up the red flight information for Tom again we can go down into the detail make an informed decision and yeah the American flight looks good so we're gonna go love that one and can again we have a confirmation step here because yeah it's important that about reconfirms the step where you really want to do before making that transaction and then we can actually go ahead and put this and then to finish off the story rebec a Giada's and checking revenue projections and NBC that Tom actually made a good impact on a revenue projections so now that's a vision video how you can spend different systems and integrate them into one interface that you can interact with just using a natural voice without without going through complicated trainings but seeing as a new video is great does tickets so I actually want to go into the system and build a chatbot together a fuse so let's do just that right now so for that and going into the platform which is available to all of you right now so you can go to see I got tools totus AP and just start using the platform just like we're doing now in this demo you can sign up just using an email address and can get started experimenting with these use cases and building BOTS we also have excellent documentation on the platform tutorials that guide you through the process step-by-step so here I'm already logged on to the platform and I'm in my organization I see that I don't have a boat yet so what I want to do in this demo here is build a chat pod that can create a lead in our sea for a Hana system so I'm a sales rep and I want to get easy access to creating leads in the system maybe viola on the road and we gonna build the chat pod further so for that I'm going to select me bot MVC that we can already bootstrap the bot with the predefined skills that the bot can have so it would be very sad if the bot couldn't even answer to hi many users just start by saying hi to the bot and if we would then say oh sorry I don't understand that's obviously bad so we have built predefined skills into this that help you quickly get started fifties I'm gonna call this but see for leads I need to provide some information about the data policy and I'm gonna make that a private pot so that only we in our organization can see it so what is now doing in the background it's copying a variety of expressions that we have built into this greetings and small talks that be used to train our natural language processing models so we're not using rules or rule-based framework to understand the user and the past that was often the case there'd ever regular expressions that check for a certain word and then if you have this word you out to this intent or to this to this action if it's if it's another keyword in this expression then you're out something else so usually you were building huge decision trees which get amonges and very hard to manage and maintain over time so we are using machine learning based technologies here which means that we can train the bots by just training it with the expressions that users can say or could say base users would express a certain intent and here we already see such intense like ask feeling as joke these are all intents that we have just a bootstrap or but with so by just creating this bot I want to show you that I can already check with it so let's say hi what is answering let's ask for a joke actually and I'll give you a second to just reduce and yeah it's debatable if it's funny or not if you have different folk than you and I can also think about and do different kind of things so while this is not super helpful for your business yet it is something to get started and something to build build upon that's what we're actually gonna do here so I mentioned we want to create a lead in our c4 system so let's make the bot understand how users express that they want to create a lead so I create a new intent which we call create lead I'm going into this intent and now I train it with expressions in the English language I could also do it in other languages so let's provide some samples of how you could say that you want to create elite and obviously when you create a lead you need certain information like the organization name and users might already be providing that information in their expression so I also want to cover these cases so let's say that's include an organization name here I could also say create lead Samson for dollar I might also want to provide a deal size and an expected closure date for the for the deal so let's say 25 october 2019 and let's do another one provide another example here and then it's a April 1st so while in an actual pot that you deploy in production you would provide more examples here because this is increasing the performance there should be good enough our demo here as our board doesn't have so many intense there's one thing that I want to show you here which is which is really important so there was a question your audience is these are potential ways of saying that you want to create a lead so what I want to show you here is in these expressions our natural language processing already understands certain things and I just see here that it also misunderstood a thing which we can correct in a second but going into this for example we see that our models already expected that Samsung is an organization it also extracted the deal size as a money amount and it's fact that the daytime as or the day that I've given as a day time so this is very powerful because I didn't even need to tell the platform what these things are it could automatically extract these remediating some work for me and actually I could fix this by just doing this manually but I'm going to do it a little differently over so I just changed the expression and then add this should also have the algorithm to differentially it is so fine so this is very important because in the end what we want to do we want to call an API that creates our lead in the system and for this we need to extract this information we need to give the organization name we need to give the deal size so we need to make these information easily available and this entity extraction that we did this he just here is doing exactly that for us and we see in a second how that comes together so with this my but should be able to understand the user already so when I say create new lead for example then our bot understands that the user wants to create a lead so does referring to the intentive is just created so the first step in our entrant pot building process is done now you come to the next step which is once the bot understands that a user wants to create a new lead you need to tell about what to do and what information that actually needs for that so in the end we want to call an API so we go into the build tab to do just that and we create a new skill it's also called as create lead because that's what the school eventually does load up here and then we guide it through free step process for building this skill the first thing that the school needs to know is bench with this skill get active when do we want to create a lead well that's simple we just created an intent for that so when the user says that they want to create a lead I want to call the action for creating the lead it's very simple the next step is to requirements and I just mentioned that we need certain information in order to create a lead so let's tell the platform that you notice that we have an entity organization so let's use the organization and let's provide a variable name that we can then use in the backend to access this information so organization has lead name we're also gonna do the date time as read date and we're doing the money amount as each amount it's another book knows that it needs this information but we also want to make the bot ask for the information in case it is missing so let's just reply with a simple text message here what's the organization if it's missing be gonna do the same for the day time please tell me the closure date or projected date however and they're gonna do the same for the money amount I'm keeping it short here so now the bot knows how to ask for this information and now the only step missing is actually calling an action and here we have already prepared an API that is actually taking this information from the bot and then creating the lead in the c-4 system so we won't see that part but so this is where our back-end lives there is some creating delete which we are calling here we also see that it can be any open API so it doesn't have to be an SI p platform service that you're providing it can also run on another hyperscale for example and one thing that I want you to finish this off is at the end of after creating the lead I want to reset the memory of the bot just to make sure that we can test about multiple times in a row so that it doesn't remember what the organisation was so if that quick check our training is finished so let's just check with the bot here and see if it works so let's start again with create new lead and now the bot is identifying yes we want to create a new leap so it's asking 14 for making the V that we just told it to so let's say we want to create a new lead for Nike what sees that we still need a closure date so let's say the Creek and we have a good feeling so that's tomorrow because it's been using natural language we wouldn't always give like this valley fine way of giving a date but we're using our natural language right so that should say tomorrow and these sighs let's say is something doesn't really matter so now the bot has all the information it needs so it can take its action and that's what it's doing right now and it's communicating with the c-4 system to create delete and I should get a response very soon that tells me the number or okay well it's a live demo sometimes the middle of air gets times out or something but it the good thing is it should have worked so even though we didn't get a reply we can try it in a second again I do have a new lead created here which is Nike with debate tomorrow actually to just show you that it actually works do it again I want to do something a little bit different now which is I already want to provide some of the information so the but shouldn't be stupid and ask for information that it already has right so let's say you lead for Walmart on and then let's say next year first of January 2020 or that's ready like this so now the bot is only asking for the deal size because that's the information that I didn't provide and now hopefully fingers crossed you get a response this time okay so lead 3:07 100 has been created and then I refresh here then we see that this lead has indeed been created a form art and then the first of January 2020 so hey we've pulled a bot but obviously we don't want on our sales reps to go into this platform and chef turbot so we gotta expose it somewhere where they live and something that works for them so it could be in the for your launch pad it could be on a web page when you're not looking at employee experience but customer experience for example you might want to expose these spots on a web page and that's the next step that I want to show you here so let's go back in the system and you see that we have a connect tab here and this connect tab shows you which integrations we deliver out of the box you just close this we click so we have our own web chat that you can integrate in your own web page and you're gonna use this in a second you can integrate your bot that you've just put into Amazon Alexa and facebook Messenger skypes like as a Peace Corps pilot you can even use Twitter or phileo which many of our customers use to build SMS gateway integrations so that you can also chat first about via SMS to keep it simple we're just gonna use the web chat here and I'm gonna I'm going to be guided through this process step by step with the web chat it's really simple because these are all optional but I can provide a color let's fill them let's say I'm feeling orange today and I can skip the other settings and just create this channel and the platform will provide me a snippet that I can integrate into my page so all I need to do is take this go into the website so this is a blank website just to keep it simple but I'm just putting the script tag here at the end of the body and now I can actually open this page that's the exact same file and you see that our orange button comes up here that lets me speak with the bot so I can click this see my awesome chat pod and I can say you lead in order to actually chat with the bot as we've seen earlier to finish this off I also want to give a quick glimpse into the monitor tab which shows you the different expressions that we've given the bot and we see what intent it has been made to and this is the functionality that I can use to continuously monitor my bot and make adjustments along the way which is actually a very very important step in your bot building process because you can't possibly assume all the different ways that users ask for a certain thing so it's very important to money toward this and make adjustments to the god - so everybody doesn't understand what a user was saying then use this expression to train your intense in order to improve the performance of the bots for this intent and in that regard this is this is so important because again this is machine learning based so you don't have a huge decision tree so it can happen that when you optimizing for one intend it can happen that you're introducing expressions which overlap with another intent so by having good intentions you might actually degrade the performance of another intent so it's very important to be able to recognize that and for this we have build training analytics which is not a very important or a very exciting feature but I just want to show you in this video very have another board prepared with this and we have this training Analytics tag here because it I can show you this in the bots that we've just build just because we've just built out the bots and I don't have to say the set available so let's just look into the demo here how this looks so it gives you some documentation how our metrics are calculated it gives you suggestions to improve the training data set just by analyzing how close the expressions of different intents are and if they like in a technical sense well encapsulated and don't overlap too much it can help you make your data set closer to reality because we've seen when you're training about the the sentences that have fir diverge but users just say send free words creating a lead but you have expected them to give much longer sentence then that could lead they're not the best performance so we're also giving you this disinformation here when we scroll a little further down we see classification benchmarks you see accuracy precision recall f1 score these are things that your data scientists will be very happy to see because it allows them to do very deep analysis and actually benchmark based on metrics we can see these different criteria for the intents that we have we can filter and sort by these and we also have a confusion matrix which shows you then the user asks for the result intent or I want to drink intent how often do they actually get the result in tank or what do they get instead here in this example you might also get one to drink for result in certain situations so this gives you a very good overview of where you need to improve the bottom so see that it's an end-to-end platform that guides you from our by the way this does noise coming from this channel here so yeah it's an end-to-end platform that allows you to build a bot connected to wherever your users are and then and then money toward the consumption in order to continuously improve the performance then let's go back to the presentation so I'm thinking about what building platforms why USS AP well it's native an output for you and we are building these BOTS that you can easily consume you can customize them and you can build your own boards on top of that and integrate them into the same channel that we're using SSA P to provide these chat pods so we seen how you can leverage conversation interfaces from each SSP solution these chat pods are to come in the future we simplify the experience having the single interface for employees as a piece digital assistant and you seen that we can build check pots on our own that concludes the employee experience chapter and I want to go into customer experience experience next so thing of recent interactions that you might have had with customer service if fear things go wrong let's face it we can't always account for everything that could come up we have problems and we rely on companies to get them solved and in fact it it has even proven by service that are shown that if you have a problem and it gets solved by the company and a quick and easy way then your custody your attention your empathy for the brand is actually much higher than if you didn't have to problem in the first place so I guess many of you can think back of a situation that you had to call a customer support and you you might have have to wait in line for 20-30 minutes which which like has felt like an hour or two only to get connected with an agent but you explained the problem they couldn't directly help so they needed to redirect you so you had to tell the same story again so you see how that actually leads to an experience that's that's not up to the standards of our customers who increasingly again coming back to what I said at the beginning increasingly expect customized personal support that's easy that's easy to access so again chatbots can help there because or actually let's let's do quick survey again can you guess the amount that is accounted to annual losses from bad customer experience well yoga mood are already gave one away in his in his keynote so I guess by now you're all aware that this experience gap accounts to one point six trillion revenue lost annually account according to Accenture so it's clear that we have to act again chatbots a great way to do this because they're easily accessible on a page that for our customers expect us to be or even inside the messaging application by offering a facebook Messenger bot for example or what's that and these chat bots are always available you can provide them in different languages languages that your users speak and they can help automate certain tasks for the user or at least guide them to the right agent which already makes for much much better experience and I actually want to show you one example of one of our reference customers SFR who is using chat bots to do exactly that meant the their customer experience so um I left a video role if you see the chat again it's not someone that you can read everything but maybe on a safari web page or a mock of that the happy user asking for invoices fairly common in the telly can telecommunication industry that you have fees or additional charges for different services so we can compare different invoices and see and the chat board will respond with if the if the amounts of the difference of the invoice amount was different and why I can also check if I have extra any extra charges for this month the here the bought says no badly we don't and then I remember that I have to go from Europe to the US so I might need an international data plan and I can also asked about that and here we see an example how the bot doesn't directly answer that question because we see hey the users asking for assistance here that potentially requires more than more consulting than its head but could provide so in the sales scenario we can guide the user or basically guide the conversation towards an agent who can then pick up the conversation either here in the chat pod or in this case they're going to call the user back in order to really understand the needs of the customer and I provide the best experience so we see it as a mock here because what is actually available in French on a website but you can go to as far as website to actually checked it out I want to stay with this case because it I think it's a great example of how such Chatwood projects can evolve in a company because we do have a certain set of features here that the bot now provides but it didn't start out with this rich feature set that as a farm has built into the bot by now so they thought about what our simple problems that we can solve that have a that have a measurable return on investment that we can show to our company and let's just get started as simple the really simple use cases so the first thing that they looked at is not automating anything but just using the bot to ask for certain information to be able to classify if the user does have a support question in a certain field or if they actually need sales assistance so the first part the first part that they built was able to do just that take on taking some information and then route the user to the right agent so again coming back to this example from from the beginning this already takes out the pain of having to explain the same problem again and again they've been so satisfied with the results of this part that they thought about okay what can we do to now automate some of these tasks so they looked at the logs they looked at what are the expectations of users when they chat with the bot and then you have prioritize the use cases for which of these use cases are most simple to integrate like it's type of tend to be much better with simple tasks that only use a certain number of interactions rather than big interactions they have to provide a huge complex context so they focus on these with a high volume again to be able to show the return on investment and they've built a couple of use cases into this pot so that by now the bot automates 22% of all customer interactions so these can be handled completely autonomously by the bot without touching a human agent ever like these invoice requests downloading invoices explaining the different items on the invoice and from the remaining requests 90% are i routed to the right agent at the first try and by having this information already available that the customer entered in the chat pod the actual length of the conversation could be cut in half so only half of the time that the agent now takes to resolve the request for the user which obviously helps the user they're happier because their request could could get solved Iran were completely automatically or in a quicker time and as so far is happy because they can now cover a bigger load of requests and more targeted with the Service agent city so yeah again ycp you belong in the s AP portfolio your data stays in the ACP ecosystem you've seen that the platform is easy to use we have support for any language if you have a question around that I can I can answer that in more detail at the end and we made for enterprise and I'm going to go into a little bit more detail on this in a second when you see the slides you can download them from the Tekkit portal we have more cases in here that you can take the time to read these cases are also available on our CI 2 tools that as ap webpage I want to stop here for the customer experience because I think you've you've got an idear and can now think about how you could build similar use cases for your organization what's new again we do have a huge focus on enterprise I mentioned training and atler analytics there are other features that we've built in the past like versioning you can build a bot and with many other platforms you only have this one version of your bot which means if you see that what has a problem if you're making changes then it's then the changes that you make are immediately available to your customers without you having the chance to identify if they're actually performing if you're degrading the performance of our intents so we allowing you to build different versions for development testing production for example so you can experiment with changes and then deploy them when you feel confident visible what's new coming out of the recast ai acquisition as some of you might know we were very open to integrate with all platforms but we've also seen that integrating with as a piece in a security context wasn't always as easy as it could be so this traditionally you would have to build these integrations yourself we've taken this point very seriously and I'm very happy to say that with our next release throughout in the next month we have a much tighter integration into sa P systems and that's only the start so what we will offer lift the next release is seamless connections to s ap services via the business connector this will be using the s AP cloud platform and the destination services that you're already using to integrate your cloud solutions into your entry ASAP systems so it's a single point of access that's where you store the authentication information and that's also very store information about single sign-on for example so by hooking into that platform and using the cloud platform we can use the same authentication context that user already has been operating with an S AP system and use it in a chatbot so that you don't have to worry about the authentication process anymore and we have our powerful web client for chat pods so what we had until today is the web chat that I've just shown you but if this web client for Ted pods we also bring this into this channels you're offering this as a challenge connector and if allow you to very easily integrate a chat pod into your your launch pad as a man I know from SOP copilot already when I give a quick glimpse and what's to come farther down the line as well and you're gonna see demos of this in a second very excited about the sa pbot generator which will allow you to based on all data interfaces that you already have and by the way this is also what s IP i'll be using internally to enable many of the existing transactions that we provide like creating leaf request checking for approvals and there are all data services beneath that and we will be able to easily integrate into these our data services and generate a bot on top of them you can do this because these all data interfaces at the bottom layer they are just regular api's they do provide additional metadata they tell you in the metadata which information is actually mandatory or optional to call this certain API and we can use this information to generate a bot on top of that so if you know that for creating a leaf request we need a start date and an end date we can use this information from the metadata and just generate a boat on top of it once we realize that the user wants to create a leaf request we can ask what is the start date what is the end date and start at an end date how would we get from the metadata and once we have all the mandatory information we can actually ask the user for confirmation and then make the call to the so data API this has already been available in copilot if you're if you're using that in the past we're bringing that back to the ASAP conversation I platform in the near future I'm also very excited about the FAQ button order again we're gonna see demo in a second which allows you to take an FAQ document question-answer pairs so question answer question answer the question answer and that many of our customers already have and it will it will allow you to upload this document into the platform and generate a bot on top of that so that now users can use the natural language to just create for information how much can i expense a ticket and smooth party bot interactions what we've seen in the channel here earlier was integrating one bot that we've built into the channel that we've chosen but for example you might have different BOTS built by different people in different domains which is a fair use case and you might want to integrate them into one single interface which really yields the power that we've seen earlier and the total workforce management demo and we want to provide you with the same capabilities of building different BOTS and integrating them into one single interface so let's have a look how this looks in the system they're going to start with the easy recipe integration so are we going to build as a bot that can create for how many vacation days do I have left and can also create leaf requests and we're going to use the business connector and the single sign-on for that so here we are in the asipi cloud platform cockpit and as I mentioned we use the destination service to create this connection into our s AP back-end system or it could also be a cloud system so here we do that for human capital management we provide the URL and we're using app to app SSL to make sure that the user or they'll be using the exact same authentication context that the user already has in your s4 system so now we go into conversation here and we go into the skill that we seen earlier see ok we'd leave as a trigger here and now in addition to the web hook we also have the option to consume API services and just provide the name of the destination that we've just created so since the authentication context is completely handled by the cloud platform destination service we don't need to enter anything here I'm just saying the response that comes back from this API call should be stored in a variable called HCM leaf and then in the response here we will be able to directly reference fields that come back from this API so the last step in order to make this available is using this new I said big conversation area iia web client for the connection here and we provide the system ID that we want to integrate us with that should be available in the free or Launchpad give the connection a name created and be done so now if you go into RS on a system that we just integrated we can open the digital assistant and ask for the leaf request one thing we've seen here which I was asked the past days and quite often is if you know the platform then you before before we had this feature you had to build a small piece of middleware that can take the information that our platform was providing and making this API call and then translated into the format that you need to make the actual API call so here is here is theme that you didn't need to provide an extra API for that we can directly call the API here and make this information available to you and allow you to use it right away in the response that you give them to the user it's can I can I come back to the question at a at the end thank you so let's have a look at our bot generator let me actually just see if I can turn off the sound here any you know the curstyn the video it does have sound but I actually want to talk over it so that's good second let's see if that works so this is a way I think you bought generator and just plan for the end of the year and what we're building here is and it's about that can answer questions about our SP conversation I platform for example what is an intent and see here's how to build it so okay okay there you so I'm gonna walk you through how to build this but so behalf our new document step in here where I can upload a new document and we're just using a CSV file that contains these question-answer pairs and upload it to the platform then we can see the questions and the answers that were included in this document and I also have the option to add new questions if I see that my knowledgebase didn't account for a certain user question and I can also add alternate what we call alternative questions to existing questions if they're different ways to ask for the same thing this is all that you need to do in order to have the bot and be able to chat with it what we're going to do here and while chatting with the bot we will also see that we provide out of the box three levels of different thresholds that about can understand what I mean with this is when I use what is an intent then our bot is fairly certain that it has an answer or like really really certain based on the confidence that the board has from the document uploaded so here we give the right answer right away and he can ask the user if he was happy with he or she was happy with the answer I can also ask another question like I need help for training analytics and here the bot might not have an answer right away or it might have an answer but it's not so not so sure about that and it can then provide different answers that it has and just ask the user what is the question that you were going to ask and then what we've also seen just a second ago is if the boat doesn't know the answer that what can also say hey sorry I don't know again in the monitoring tab we see the different expressions that we've just asked about and this little number at the back there is the confidence that about had in the answer and we saw that for I need help of training analytics the bot wasn't so sure about the answer so it used us disambiguation so what we did here is assign this expression to one of the existing answers and now if I ask this question again the bot is more confident and can give the right answer right away so this is how I improve my body that's all I wanted to show for today I think we've seen how you could use s AP Club s IP conversation I right away you could just go to CI the tools that I say P experiment with the platform yourself this team that s AP is using the same foundation for building what's in there or in our product so you can look forward to having these out of the box with with the s AP solutions very soon yeah and that is easy to get started so I can only encourage you to try it out do we have a roadmap session so or is it just a last session but if you look in the available slides you will find a in8 free one with the road map information you're also going to have it here in these slides very provide an outlook into what's to come because we do have despite the features that I've just shown a very nice features in store [Music]

Channel: SAP TechEd

Views: 5,754

Keywords:

Id: i92KYZP40K0

Length: 49min 25sec (2965 seconds)

Published: Tue Jul 07 2020