Explore the benefits of our customer service management system in Mexico

How to create outlook signature

and this is a particularly exciting solution it is the The Graduate of predictive intelligence for customer service management so you're going to see very exciting updated model or Autobox capabilities and just a lot of value to deliver to your customers so at this time I'd like to turn it over to two of my student colleagues Andre and Alexandra to go ahead and kick it off so Andre do you want to go ahead and get us started absolutely uh good morning good afternoon good evening everyone um my name is Andre amstrad I'm an outbound PM for our customer service Management Group I oversee task intelligence wfo and some other product capabilities and Alexander you want to choose yourself yes my name is Alexandra is for customer service management all right um so we'll get rolling uh let's go to the next slide we'll just start with a brief fast oops go back to the safe harbor if we talk about any exist uh any forward-looking functionality we will uh call that out uh on either verbally or on the slides um so just a quick Safe Harbor statement on that next slide please so when we're talking about task intelligence for customer service from a product uh perspective this is supported in our Tokyo release and the assumption would be that you would have some type of pro level licensing next slide please so what is Task intelligence the way that I like to think about task intelligence is that it's going to help you automate the different aspects of your case life cycle and this will speed up resolution um in with respect to support workflows so when we think about uh the case life cycle there are kind of five steps in the case life cycle step number one is what we call creation so this is where you may be in taking work let's say from a portal from a chat from an email Etc and what we want to be able to assist with here is with optimizing the routing so perhaps we're predicting things like the language or a topic or sentiment or in the future we might be predicting duplicates or spam or things of of those nature the second area that we often impact is what we call self-service so this is where your goal is to drive deflection and in the future we're going to be offering functionality uh in this area so this is going to be to help automate response perhaps via kbia articles or to proactively recommend uh case resolution or Etc the next step step number three there's two areas here there's the what we call the triage and handoff so that would be the situation in which there's other teams that are involved in the support process and then of course there's resolution and we're task intelligence impacts here is we're going to improve age of productivity so by Auto categorizing let's say a case or by extracting values from an attachment and inserting them onto the case form uh or even recommending similar knowledge based articles or similar cases we could help drive and streamline that agent productivity and then the final player here is uh after the case has been resolved how do you optimize your operations in an ongoing Manner and this is areas where we might be able to identify let's say gaps in knowledge or perhaps using uh capabilities across this optimization look at root cause analysis so say seeing where there is breaches that are occurring or bottlenecks in the support process and just to overall identify uh opportunities for Downstream automation so when we think about how does task intelligence play in the overall customer service landscape you can think of it from this perspective from this case life cycle perspective so let's have a double click on this a little bit and look at the specific capabilities that we're going to talk about today the first capability is um sentiment analysis where is the ability to detect customer sentiment using AI so this is the ability to detect the initial customer sentiment as well as their trending sentiment as perhaps they they go through the case life cycle so we are leveraging pre-trained uh AI models to detect that additional sentiment and when we also detect their change in sentiment that may also trigger some Downstream action so perhaps it may trigger the escalation of a case or the reprioritarization of a cage or maybe even agent coaching right so if an agent is handling a case and it starts off positive but then it starts going uh very negative perhaps that is an opportunity for Asian coaching as well and from a higher level perspective service owners can prioritize rise which areas which product areas and service areas they may need to improve based on overall sentiment across their segment of customers next slide please document intelligence is another exciting capability in task intelligence so this is the ability to extract field values some attachments we won't go into this too much today because it is has been covered in other Academy sessions but it is a very exciting and new capability within task intelligence next slide please categorization so case categorization and language detection this is a capability that we're going to be looking at in depth today and this is the ability to interrogate an attachment let's say that comes in on a case and then use that to predict certain field values on the case form so perhaps you're interrogating an attachment let's say an invoice and you're saying that this invoice is is to go to uh the billing support group as an example or that this invoice is going to Spanish support as another example so having that ability to analyze an attachment and use that to predict field values can once again help automate routing so for a lot of organizations they have multiple inboxes and many routing rules the case categorization is going to really help reduce the complexity and the administrative overhead associated with that next slide please and admin console so admin console uh pulls it all together so in the past um it may have been more difficult for a admin or developer to kind of set up uh Ai and task intelligence because there was multiple places within the application that you had to go so we brought it all together in the admin console and what this enables you to do is to rapidly configure train and deploy your AI Solutions and really this puts it in the hands of a business analyst type of persona as opposed to strictly an admin developer type of persona it's a very guided methodology for setting up the different models that we offer on the platform and for a lot of organizations this is going to help them reduce that deployment anxiety associated with rolling out an AI solution you will also have the ability to decide when you want to move from let's say recommended field values to autofill field values so you can do the crawl walk run type of approach and lastly you're going to be able to directly measure the hard dollar business impact of deploying these AI Solutions so let's say that you roll out something like a case categorization you may be you will be able to to say hey this resulted in you know 720 minutes of saving per month as an example and have a real hard dollar value calculation that you could take back to your business stakeholders to prove the value of these Solutions so with that said I will hand it over to Alexandra thank you Andre so let's first understand it as intelligence admination for this exercise the goal is to drive agent productivity by helping them to first save time recommending the category field and predicting priority for all cases using the case's short description and description will guide the agent in the right direction to prioritize predicting the cases sentiments will help the agent to act ingly and three improve routing predicting the language skill for the case and integrating this prediction with advanced work assignment will route the case to the right agent so we have a lot to cover in this exercise and we will focus in three steps first the Asian experience with task intelligence then we will set up past intelligence and finally we will get a glance of gas intelligence Analytics so let's begin by seeing task intelligence in action in the Asian experience so what we have here is we have in our left side we have our consumer he is ready to send a case um in the right side we have our agents in the top part we have John who speak English and in the bottom we have net uh the speak Spanish now if I send this case now from my uh consumer what does intelligence is doing right now is predicting the category field the priority field the language doing the routing based on the language and uh even when we are not going to be touching much on it today we have document intelligence set up to extracting the vendor from the attachment of this case so let's see how the Asian experience uh this ifia said this case what we are going to see is first we can see how the task skill was added to the case so the English skill was required for this case and that is the reason for why he was assigned to it if we look into the details you can see how the priority has been predicted and the category has been recommended which is a different behavior when we predict we save the value in the case when we recommend we will just display it to the agent for them to take actions we can see in the left side as well we have the language because we decided to put it on the case form two and in the right side you see how the current sentiment is negative for this case so let's say that this agent want to reply to this phase and we are going to make a comment saying please disconnect the router for three seconds and try again we are going to pause this comment so they are what consumers can see that if you notice sentiment is not being recalculated that is because we don't predict sentiment in agents uh comments we will just learn in the customer or consumer so let's check how that looks for him he already our consumer has gotten the uh comment from our agent after he tries he realized that I was working and he is going to be making a comment saying thank you finally works he will send that to our agent and our agent now can see how the current sentiment is positive and he can see as well the sentiment over time is improving so the last piece of these is if you notice in the top bar there is a review window can tell so that means the award um case has an attachment and we are able to work on it so if we look at it right now he's saying that in this moment is still striking the values so it will be available to the agent later on but if you want if you cannot wait he can um work on it right now so let's say no I can't wait so this is in general how the agent is seeing that interaction now this is about how the agent work but what about the admin So Listen by the Admin to our screen and the first thing I want you to notice is when they when the admin is working in these cases they can as well use something that is called the assign word assignment workbench so let's think in a case that was written in Spanish if they want to use the assignment or bench then they just agents the match the skill will be presenting to them that is because we are saying that the Spanish skill is mandatory so with that we kind of conclude the agent's side of it and how does intelligence interact with it so let's think how our admin is going to be selling all this up we will navigate to Thousand candidates and we will go to the setup portion of that let's maximize this so it's a little bit more clear for us so this is our main console this is where the admin will be interacting with how to set up the models in the top part you always will have a performance live so it will be showing you the predictions that has been done in the last seven days you can speed it by each of the models so you can see that uh performance being live now on the right side on the task intelligence admin console you have help and this will be consistent in all the exercise you will see how that is really useful in the middle session we have any model that you have deployed in this case we have the language the sentiment and the categorization as we mentioned and in the bottom side you will have the cards that you can use to be able to set up each of this model so for categorization or for sentiment or for language in this instance we have everything set up but I really want us having the flavor of how all this is being set up in task intelligence admin console so what I'm going to do is I'm going to be deleting uh the solution and we can we do it together if you notice when a model has been deployed you can edit the model you can export the model which it is a important function when you have a model that is performing really well in an instance and you want to move it to a next one you always can explore and import the model in the new instance for you to to work on and finally we can delete a model so let's go ahead and delete the language now how we set it up so we will go to the language card we will say setup model and here we are in the left side you will have the steps that you need to take for um setting up a model in the right side you have the help so the first one is testing your model the reason why we are testing a language detection and this is the same for sentiment is because we have pre-trained models we don't need your data to be trained to be able to predict languages and as you can see we are predicting 20 languages uh good question for you on on that so the 20 languages that are supported in sentiment and Analysis are also supported in case categorization or what's what's the story there yes that is uh that is a really good point Andre thank you the first thing is these 20 languages are for language detection for sentiment we support just English and for categorization out of the box we have four languages there is French Spanish uh um German and English if there are other um languages that we can support but for those you will need to contact a service now thank you for the question so now anytime that you have a question like for example the one day Andre has you always can click and explain this explain this will bring you to the right side it will bring you a more detailed help of the step that you are taking so for the modern name for language it's already set up for you you don't need to change it because like we say is pre-trained now in this session you we will be choosing up to 100 cases to be able to do a testing so give you a flavor of um how the predictions work so for example I'm going to say I want to choose a hundred cases for these they have been created um after you know last month so I'm going to launch that and like I said In This Moment is doing uh testing is not doing uh training the whole point of this exercise is making you feel comfortable with the predictions so in the top part is telling you how many predictions were done for cases in that period of time in the bottom side what you have is that sample of cases that you can um that we have predicted so if you want to have a little bit more uh an idea of what what that prediction went so you can click on the case you then will be able to see the short description and the description they generate that uh prediction and the whole point is feel comfortable you know make sure that those predictions are matching what do you expect for this model once you feel comfortable you will go back you will save and continue now it's time to set your preference this is a really important step of the process you will know if there are two main uh points the first one you can use predictions in new cases which means this can affect uh your case or case form and in the bottom one you have monitor predictions in the background only so you always can just test this model once you set it up as background only what we will do is we will save the predictions for you to do analytics on it and feel more comfortable with it before you really deploy into production now in this exercise we are uh adding the language uh to the case form and we added the language as a skill to influence the case routing using advanced work assignment or assignment workbench as we saw once this has been set up you will say save and continue and the last step is deploying the model before you do you make sure that everything what you have done is being captured in this summary and what you have done that you will deploy and that's it from now unknown every time that you create a case they share description and the description will be taken in consideration to predict your uh language field I'm going to stop here for any questions about language before I move to the next next session Alexander no questions on language um but we did have a question regarding um what we're referring to is CS and um in this case CS is customer service management so keep an eye out um our first release of this is customer service management we do under Safe Harbor expect additional areas to be supported with task intelligence um and then Alexandra as we're on the model page this may be a good opportunity to maybe talk a little bit about what's unique to task intelligence because the question came in um you know what are some areas that are new in in relation to predictive intelligence which is the solution that came out um several years ago with classification um so the greatest thing about the test intelligence and the console is they now there is no doubly Set uh um definitions right now what we have is the opportunity to create on the Fly whatever model we need for whatever fields we need and we have a more interactive way to work on it right before we have the uh solution definitions then we will be giving you out of the box to predict assignment group or category or priority now you have the freedom of uh selecting whatever you want and uh deploy it dynamically I don't know panu or Andrea do you want anything that you want to act yeah if I were to summarize it quickly I would say for me the admin console kind of democratizes um AI in in the sense of now it's in the hands of a business analyst type of persona they can Light It Up have it monitoring in the background and as it's becoming more effective and their their confidence in in it grows then they can flip it to autofill so to me it kind of moves it from the the domain of technical people that understand the language of AI Etc and puts it in the hands of business stakeholders that's the way I like to think about it thank you so uh if any any other question Marcelo or should we move on to categorization sir let's move on okay so for categorization what we are going to do is let's edit the model so that we can skip the training part uh I'm going to edit this model you notice that now instead of four uh steps we have five the first one is the find your purpose and this is a really important step here what we are doing is deciding which data we are going to be using to train our models and you have two options here the top part is saying do you want to use emails as a source of training your data this option is really good for those customers they do any kind of customization when they are creating cases from email and they are doing any kind of modification in the short description of the description so they may prefer to go directly to the source the shift email table or the email table to be able to get the predictions in the second bar in the bottom part you have the opportunity to do training in cases or case extensions and um you notice that for both of them there is now a checkbox and that check box is include attachments you can use attachments now as an input for your predictions again if you have any questions remember you can pick a screen here and you are going to have a complete help only now in our use case we say that we are using your short description and description so no need to change anything here then you go to train your model you the process step is you will put a name in your model once that name has been set you cannot change it that is the reason for why I cannot change it here because we are editing this model now the second step is uh if you want to train your data you can set up any conditions for example you can say I want to train my data in the last year of data that I have in my case uh table then you will select what are the output fields that you want to predict we say that we wanted to predict priority and recommend category so that is why we have these two Fields here um now in this session you can select the short description and description they are going to be the inputs for the prediction and finally in the bottom you have the number of Records they are going to be used for that training they match that condition in my case for example because I didn't put any condition is bringing the uh the total amount of cases I have in this instance so Alexander is there is there um a minimum number of records that customers will need in their instance to train these models yes I'm great we out of the box we have 10 000 records uh as a as a default but as you can see here I have less than that right I have a thousand and that is because we do have um a property there is called mean training limit that we can set up if we want to lower that threshold but you know we encourage you to have at least 10 000 records because we are the model is being trained in your data so it's important that we have enough data to produce good predictions now this step number three is assess your model in the top part you have the number of predictions that we did they could be done if you have deployed this model during this period of time and this is just to give you uh an idea of how many predictions so that you can do your own calculations for example if you say every prediction is saving me 10 seconds if you multiply those 10 seconds for the number of predictions that we are presenting you you may have a rough idea of how much time this model could have saved if we have deployed to that period of time in the bottom part we do have a kind of like an overview of how each of the fields um has performed during that predictions so you'll see here we are saying that the same as the agent is 100 and this is just because we have pre-publicated data with a thousand cases if this won't be the case in your real life but you will have uh what is the percentage of predictions that were the same as an agent what were different and which predictions will skip it for any questions we have the help so it will explain you exactly why but remember with categorization we are training program to data so we do have a way to compare our our predictions uh execute versus what the reality on your data is so you always can go to the view samples resource and you will be able to see that so if you are curious about one of the cases you can click you can see the description as your description just like we did in language you can see what was the option um that the agent choose what was what we predicted and we will let you know the comparison if it was the same or it was different again all this is for you to feel comfortable with those predictions once you have done that you can go now to set your preference just like in language we can run this in background and nothing will happen but if you want influence if you wanna save your prediction now you have some options remember that we say we wanna recommend category so here is where you can say I wanna recommend I don't want that being saved directly to my case form or my case we just want to recommend it to the agent for them to take the decision if we want to fulfill then we can just say preview if in any moment of those predictions you are not feeling comfortable with one of them you always can uh disable that field and we won't do the predictions over that field so Alexandra let's say that you're in addition to the category and priority let's say that you're predicting a field like assignment group as an example would there be a handshake between um the prediction and AWA routing like can can you maybe explain how that may or may not work uh so Advanced workers we are not integrated out of the box with advanced work assignment you will have to add that integration but yes we can use a category uh the fields that we are predicting with categorization we can use it as routing and the way that you will do that is in our case we have a field that is called the prediction status that prediction status is telling you when the predictions has been completed so by adding that to your queue uh in your Advanced word assignment you will be able to hold the case until the predictions are done thank you for the question Andre okay uh now once you have set up your preferences you can deploy your mother again just like in language you have a summary of everything and you will be able just to pick deploy and the predictions will be ready for you to do uh before I just give you a glance of sentiment do you have any questions in categorization okay two questions yeah we do have a question on a little bit on the model training and running um so the question is do the models sit within servicenow or the external disservice now I'm happy to take that one or Andre Banu or Alexandra yeah go ahead um uh Marcel but as far as I know the training happens in um the servicenow data center wherever whichever data center you're located in and I believe once the model is trained then uh your specific uh the customer specific model is then pushed back to the instance but I could be wrong that that's the extent to which I know yeah correct it's all within the servicenow infrastructure um so it goes from your instance to the prediction server so no risk of leaving servicenow um related infrastructure there thank you okay perfect so now for sentiment you will notice is uh pretty much the same the language the only thing is here you can say do I want to predict sentiment for all cases or do I want to predict cases sentiment for just case types so in this one you have the option to say I just want public sentiments for order case if that is the scenario that you want to cover then it becomes pretty similar to language we will test your model you can have up to 100 cases they will be showing you what sentiments are um is being predicted then we will assess the model so now you can see here for example why this case was negative or neutral just like in language once you feel comfortable with it you can come to set your preference and in this case it's a little bit more simple is do you want your sentiment be saved to the case do you want to run it in background and finally we will deploy the model so that is about Eid about the setting up of these uh models in task intelligence I Wanna Give just a brief glance of the analytics because it's important right we have these models already deploy they are already doing predictions but how I feel about those predictions it is uh you can come here and click view all metrics so when you go to be all metrics what is going to be doing is it's going to be presenting you with um this charge where you can see what are the predictions that has been done for each of the models you can see the mptr and more importantly in the bottom you can see how many predictions the agent has accepted and how many predictions the agent has replaced that is because every time that the agent close the case we capture that um that change if they were not comfortable with the prediction we captured that as a feedback to be able to display it to you and with that I conclude the presentation I'm not sure if there are any other questions yeah Alexandra at this time um the the only question is under Safe Harbor is Task intelligence uh the future predictive intelligence for itsm hrsd um the answer is yes we are expecting additional capabilities to be uh shipped in subsequent releases but the first version of that is the task intelligence you see today so we have that available for customer service management you can use that with document intelligence which we've had previously recorded webinars on that so feel free to check those out and at this time we'll take any additional questions and conclude the presentation for today so if anyone has anything else please type it in the chat and while we're waiting for questions in the chat I'll just uh call out a call to action here which is to get you guys to go and download task intelligence from the servicenow store and light up whatever feature whether you're talking about sentiment analysis categorization document intelligence Etc and give it a go because as you saw in the demonstration you don't have to deploy it immediately into production you can just have it run in the background see what potential impact it could have to your organization and incrementally light it up as required so that's our call to action for you folks great thanks everyone for attending thanks for your questions your participation and we'll see you in two weeks