my name is trisha cornish and i am a solutions consultant with the now intelligence group here at servicenow in today's live session our webinar is jumpstart and deploy your predictive intelligence for it service management um i don't think this is going to come up today it might come up in questions but i always like to put the safe harbor statement uh up in front of all of my customers because sometimes i get crazy and i talk about things in the future if i happen to make those statements those can't be held against us or used for buying decisions that's all that we're saying here but i think we're pretty safe today because we're talking about uh current capabilities um welcome to live on service now at today's webinar is just about predictive intelligence and one example of that but we have webinars and meetups across a host of topics whether it's upgrades itsm specific topics i did one a couple weeks ago for performance analytics hrsd there's so many more we're going to post a link in the chat so that you can learn more about these interactive events they're really helpful and um you know whether they're webinars or meetups hopefully we you find information that helps you with your deployment and adoption okay some housekeeping tips and i think this is key this is a virtual webinar different from a meetup we want to have a great conversation with you but i have a big presentation ready for you so um of course feel free to ask questions in the q a and chat and please stay on mute um just so we can avoid noises but if you would like to jump in and ask a question unmute yourself um and introduce yourself we have a global audience here and it's great to see you all so uh thanks for throwing out your location and um we look forward to an interactive conversation for those that are concerned the session will be recorded and posted to the community later this week so now is your chance to opt out if you don't want to be recorded okay a little bit about me my name is trisha cornish i'm a solutions consultant here at servicenow what that means is i do demonstrations i present how the software works uh the art of the possible uh i am like a vanna white if you will if you're familiar with wheel of fortune i've been at servicenow for a little over four years i've spent 20 plus years in software specifically erp service and support delivery not all of that time has been doing pre-sales presentations i have done implementations and my focus is really helping customers envision what's possible so being able to facilitate product adoption as a passion of mine i really want to help you realize the value of the servicenow platform that you're all using and really fostering that community that we have okay let's start with a poll question how familiar are you today no judgment and there are no names how familiar are you um with predictive intelligence in itsm and it's okay if this is your first introduction to it maybe you've heard about it but you're like oh i don't know protective intelligence sounds heavy maybe you've tried it out you've poked around but you're like trish i've got it i've got a model but i'm not sure how what to do with my model and then maybe some of you are actually using it which makes my heart happy how do we get better well hopefully i'll have a little bit of something for all of you looks like we've got oh a couple people using it fantastic a few people familiar um but this is going to be new so hopefully this is a good introduction to you know just dipping your toes into predictive intelligence um all right let's go ahead and end that poll got a second one and this one's kind of fun um what is your role at your company uh are you a data scientist are you a business analyst a process owner or a developer administrator consultant and here's why i'm asking this when people hear machine learning and predictive intelligence of course they think of the data scientists and we at servicenow employ those data scientists we have those big brains i am also not a data scientist um i am you know as i introduce myself i'm a solutions consultant then the equivalent of a functional analyst and a process owner i know how the processes should work and i can identify where i want to see automation and improvement now we are going to use machine learning algorithms and machine learning models to make that happen but i'm not the data scientist that built that model we have those big brains here at servicenow that provide that and then provide a tool so that i as a process owner or developer can implement that and so really that's where i want to focus for for that data scientist on the call i am envious of your knowledge and expertise um and uh there are plenty of um posts on the community about specific algorithms and vectorizations and so hopefully that will meet your data scientist um questions but we're going to focus on the process today so thanks for answering those questions and hopefully um will help you see the transparency of a purpose built machine learning solution within the servicenow platform alrighty so here's our agenda i'm going to go through it pretty quickly just a brief overview of the frameworks and and how we see machine learning working within a process i'm going to use predictive intelligence the classification solution as an example of predictive intelligence for itsm it's usually how customers are getting started and so uh i'll just talk about some of the prerequisites for getting those models going now i'm actually going to jump ahead to the end where we see machine learning as part of the task process because i believe the model definition and activations there's a lot of information in labs and exercises on now learning and i want to cover where we see this played out in the process before we go into definition i hope that's okay with you and then of course at the end resources where you can find those posts on the community those now learning training labs that are really going to help you see the whole picture alrighty so let's get started ai powered workflows boost business operating curves so so here's where we see the value of artificial intelligence ai and these ai powered workflows really do enhance the user experience because they're constantly learning so for those of us on the call who are familiar with servicenow say assignment rules it's very rigid we do provide there's a lot of value in and i'm using picking on assignment rules is one example there's a lot of value in those pre-set configurations but it kind of plateaus there's no ai and so when the process changes when you have new categorizations or new assignment groups or even process changes someone has to go in and fix that and so yes you are receiving value from it but there's a point where it just kind of levels off and and there's no continual improvement now the beauty of ai is with ai enabled we're constantly using the data to inform and improve on its own it's continuously learning and so the workflows move the curve up because what we're reducing is the effort but we're increasing the value one of our customers university of maryland really transformed their service delivery with now intelligence because what they did was they used the models instead of the rigid rules to correctly route incidents so just that if you think about just triaging incidents the time it takes well while our team members might be correct their time is valuable and we want to make sure that we're taking advantage of that um but with that automated assignment we were able to improve resolution time because we're no longer first triaging then assigning we're just jumping to an assignment you can see how we could reduce our resolution time and of course there are other elements within predictive intelligence like similarity for surfacing knowledge to agents that help the university of maryland reduce their incident volume improve the resolution time by pushing solutions to to the users and so when we're talking about ai powered workflows the example i'm using today is predictive intelligence with classification but as you're going to see we have multiple frameworks and multiple scenarios where machine learning within the platform really changes your digital transformation experience okay so broadly predictive intelligence encompasses these technologies and these artificial intelligence frameworks and this hour that we have together is not enough to cover all of them but uh if you're searching on the community for specific information about some of these ai capabilities i'm going to point you in the right direction so the first one let's go back sorry i didn't mean to hey i searched in natural language processing this is where we can understand um the human statement we allow humans to interact with me with the machine using natural language perhaps everyone here on the call understands what we're talking about when we talk about the classification and categorization of incidents but you know what business users don't and they might not know that their request is actually an incident or vice versa they can just make us an utterance in our virtual agent and the virtual agent can use our machine learning solution to understand what the user is really trying to get to and present a solution for them same with ai search within our portal and our virtual agent we provide inform access to information with the minimum number of clicks it really takes advantage of our natural language processing in the platform now these ones that are remaining these um similarity and clustering are unsupervised frameworks and and we develop solutions using these capabilities like automation discovery or knowledge demand insights and what's great about the unsupervised learning capabilities within the platform is and i'm going to pick on clustering first clustering is just grouping grouping data into clusters based on their um and i don't like using the word similarity um but you know based on a common theme uh for lack of a better statement so as an example we can cluster together incidents based on short description to surface what common themes are perhaps identify um in our automation discovery what virtual agent topics might be most useful to address the most commonly used most commonly created incidents we might use clustering to look at opportunities within our customer uh survey comments as an example clustering is grouping unstructured text the string fields um like short description description resolution notes now we don't cluster journal notes or work notes but if you think about the string the unstructured text in columns there's a lot of information there and clustering can help group that together same with similarity similarity we're going to identify data that's similar to a common target now clustering and similarity together are part of our knowledge demand insights solution and so together we can group uh incidents together and compare them to outstanding knowledge articles to see where our gaps are and so that's an example of where we're using these technologies together to provide a solution for clustering we use automation discovery in our virtual agent to look at our incidents by our short description to see what topics might be most beneficial so really you can start seeing where we're adapting these frameworks into delivered solutions so that you can get intelligence from your data in the platform now where we're going to focus is actually our supervised learning uh examples today classification regression are supervised learning models and with supervised learning models we have a defined output and we're going to take specific inputs and uh the we're going to create a model in which we can predict which class or category the target data belongs to now with regression it's very similar the difference is it's numeric value so today's example that we're going to walk through in this getting started is around the classification the difference between classification and regression regression into this layman non-data scientist is for one we're predicting a class or category like a choice or reference field with regression it's a numeric value okay some examples and i was talking through examples as we were going through the frameworks but here are some other examples across the different workflows and you can see where we start adopting these frameworks into delivered solution um if you're using the agent workspace you might see similarity on our agent assist i'm going to show you examples of classification where we're looking at incident categorization and incident assignment but our change intelligence also takes advantage of our machine learning solution and with clustering like we were talking about you can we are using it for our knowledge gap analysis but we can also take advantage to identify instant clusters um within the day so we can surface common themes of incoming incidents sooner than a report might be able to generate and identify that so some interesting examples for itsm let's so we're going to talk about clustering today okay so with or sorry we're gonna talk about classification today i should take a breath um and before we get started how are we doing on questions and comments any questions so far uh do we have pi algorithms to find out the correlation and tell us uh the customer that these alerts might reserve okay um akansha thank you for that question so actually um a conscious question i'm just going to read it aloud do we have pi algorithms to find out the correlation and tell the customer that these alerts might result in a p1 proactive intelligence so today we're talking about predictive intelligence for what i call human assisted tasks and transactions but we do have ai ops and machine learning for proactive notification on alerts and events so unfortunately i'm not going to be able to speak to that today a cotton cha but we do have a solution uh within the platform that can help you identify that and i hope that answered your question do we have any other questions um did i miss any questions above alrighty well let's get started so classification classification is taking inputs and identifying a defined output and the outcome of this is you're going to have faster resolution times for just an improved experience not just for the agent but also just to reduce errors um when we are automatically categorizing and prioritizing that helps us facilitate better routing and so as an example we're going to take our short description and we're going to correctly identify or correctly predict correctly predict the category and the subcategory perhaps we might even predict an urgency or an impact or a priority column the output is really up to you and defined in your model we're going to take the inputs which is our short description and identify what those outputs are and based on whether we're defining a model for our assignment group or our categorization that output field is going to be automatically populated what we're doing is we're reducing clicks and i'm going to get to this in a moment our goal is to reduce clicks for our agent and to improve the results that we see on the screen um when we categorize work we're we're taking advantage of a historical data and this is going to become key when we get into the process analysis when we're looking about at our historic data we really want to think about like good data in produces a good model so this is actually one of the things that i focus on when i talk to customers like how's your data going into the model we're going to take this historic data and i'm going to cover your data requirements in a moment that becomes the basis for your model and your and from that model we create an algorithm and then as new data is entered then that's how we get our predictions so if historically um email incidents were automatically going to the software category in the service desk assignment group then that is what the machine learning algorithm is going to predict for a new incident coming in with that same short description the difference is the time saved so we're going to identify the data and we're going to make sure that the machine learning algorithm can learn from it okay so that's just a brief overview of the example i'm going to use today in classification let's talk about some prerequisites so um you do want to um at the end of the webinar i do have some resources for you for the community posts that are helpful as well as now learning um to get started on defining your models and so we really encourage you to go through the training and um look at the product skills there's also information on now create for getting into those adoption and that's where i grab these from so you'll want to make sure that you check the box on training you've checked the box on the technical requirements for instance um if you're an on-premise customer we do have on-premise requirements for enabling machine learning um if you are a government uh customer we have uh separate requirements for enabling machine learning within that controlled environment and so really just want to make sure that you're familiar with the machine learning architecture and the technical requirements before getting started this last one is the reason why i put this on this slide we want to set expectations so our goal with predictive intelligence is to match human assignments so predictive intelligence will not be able to predict 100 of the time it's not going to get it right that is your fixed and rigid assignment rules so let's just set expectations that we're not going to be able to predict 100 of the time with 100 accuracy what we're going for is time savings as long as the solution you build can match your human accuracy that's a significant value because what we're saving is time time and effort so the example i like to use when i coach customers on starting predictive intelligence imagine you're bringing a new employee in on the first day as an agent and you're gonna sit them down and you're gonna say okay you're gonna start taking um incoming incidents and when someone describes the incident you're going to put you know their issue in the short description a little more in detail in the description and then you're going to identify the right category and hopefully the assignment group uh will come up so a couple things you know you'll give them the rules for which they're going to find that information in effect that's exactly what we're doing here but the best thing about this is if you have the machine doing it we're reducing clicks for the agent and we're approving accuracy and speeding up the intake process and so that's the value of it okay another prerequisite are the roles there are roles specific to machine learning both for creating the models we have a machine an ml admin and an ml report user so that you can look at the results of your admin and user and then we if you're using the predictive intelligence workbench there are some additional roles for using the workbench so we've got our roles and then our data requirements for classification specifically we need some data and so we recommend you know 30 000 qualified records now here's what i mean by qualified records you want to make you want to establish your data health so we want to make sure we're looking at closed instance you know they're through the process they are they represent um the assignment group or the categorization or whatever output field you're using actually is correct and so there is an exercising just cleaning up your data help and that would happen with any machine learning solution the difference is we don't have to reformat that data we don't need to reorganize that data we've already got the data in the servicenow instance in the format it needs to be so now it's really just data health do the do the descriptions and the categories match to what your expectations are so then that becomes the basis for your good model so the example i give is if you want to automate incident assignment based on caller location as an example you need resolved incident data and with accurate groups based on the color location um and an identified source for identifying that location like like the dot walked color location field so you want to start thinking about the process and i like to visualize employing a new agent how would i train somebody on this that's the perspective i take when i train my new model okay so let's stop for a moment i'm gonna take a breath alexander asks what is the best way to identify progress of pi out of the box dashboards are pretty hard to understand and use alex i've got the answer for you right now okay so let's talk about process integration first because i think the teams are pretty solid like if you have started looking at predictive intelligence you will cover a model definition in a moment but i think understanding it in the process is key to the overall picture so alexander let me know if i answer your question in this next section all right so let's talk about the data elements together so i'm going to use our incident process as an example what we have are we have solution elements we have our data and i like to think about the data from like a rolling forward now in my demo instance i have to use a fixed set of data because i'm not entering incidents every day and i'm certainly not entering incidents on the volume scale and diversity that i need for a good model so my data is fixed but for customers i encourage them to think about recent and relevant and diverse and of course as you are thinking about rolling this into production it's going to be rolling forward so you're not going to set your data on by fixed date you're going to think about it from a rolling and continuous improvement perspective and then we have our model definition now there are two places where you can define your model definition for classification and that's the predictive intelligence workbench which is our preferred solution for classification and our predictive intelligence models i call this a legacy model definition it's the home page and it's where you can identify models for similarity clustering and regression and so just understanding that there are two places where you could identify a model is going to help perhaps identify downstream which reports to use and how how and if you want to migrate over to the workbench and then activating it when we turn it on i can create models all day long until it's activated it's not useful and so the great thing about this is i can create models and nothing breaks um when i activate predictive intelligence there are a couple different ways i can turn it on i can activate it by way of activating it through business rules activating and the predictive intelligence workbench or and there is a plug-in for this i can use predictive intelligence in flow designer and the reason why that is helpful is then i can use predictive intelligence for task records that are created through say the virtual agent or perhaps i want to calculate regression and surface those results to the user things like that the and this is no small thing the task itself and by when i say task any record i'm going to use incident record as my example but from a classification perspective we want to classify and then machine learning is out but you have a task life cycle to think about where are we turning this on for our users are we going to re-predict at specific points within the process when we how and when we populate the output field with our prediction is going to be part of this activate and life cycle process do we like the engagement with the user do we like the notification that i'm going to show you in a moment um or are we just going to have it in two separate columns how are we going to surface these results to the agent or are we and then of course results we're going to look at the results how accurate is this model um when we look at the results we actually have another business rule in place to make sure we're updating the final output the you know the final uh field in a closed incident back to see if the prediction was right or not and then we have pa dashboards so the reason why i want to point out these five elements is because this whole thing goes into the process of predictive intelligence and then that can help you identify your retraining frequency these incidents that i'm creating today become the basis for my model that i retrain say next month or that i update a little bit more frequently so you want to think about how often the model takes in new data to create a new model okay uh i'm going to take a moment and just look at the questions that came in um so i kind of unfortunately don't have any of specifics for predictive intelligence for alerts but hopefully we can talk about that a little bit more so i'm sorry i can't answer your question about um ti and driven to specific events i think we really do focus that in hla um so we alexander i think we're getting through your question so victor is this something that we have to go through the development process of dev test and prod um so what we do when when you're defining and refining your model um we and actually it is dependent on the model for instance nlu um which is tied to our virtual agent that might be a dev test pro prod migration versus your machine learning model you can bring in a machine learning model but then you might update and train in just production anyway i think if you change the model definition that might be an opportunity for that migration process we might have to victor we might have to follow up on you on specific community posts around that process ramesh asks why limit 300 000 as data i don't uh relevant it's a relevant data set and i believe they're um this goes back to the safe harbor i believe they are changing those volumes as we expand that um but i i uh i actually don't know why i don't know why we do that i just know we do it so i'm sorry ramesh i can't get i can't give you a good answer and ramesh got disconnected when i answered his question ramesh i have no idea sorry i don't know why we are limiting um ramesh had asked why limit the 300k as data um and so that is part of our our training and scheduling grab from our data set as we build the model we are improving safe harbor statement we are improving the volume um the volume limits but we do have volume limits that you do need to be aware of so i don't know why we have volume limits i don't know the true reason why but we i know we do have volume limits okay let's talk about the process um and this is just an example process for classification by contact type and um i like to separate event-based incidents i'm going to push those off to the side versus human assisted incidents that's right that's where i live human assistant instance how do we save our agents time how do we improve call resolution rates but if we think about the way we bring in incidents it's depend it might depend in my scenario it was dependent on the contact type so if we think about it when a user calls in and speaks with an agent the agent is one that's actually populating the information in the short description and the description field they are they have been trained in capturing relevant details right and so that information might be a little bit more accurate than say a user who sends in an email with help i need help with my email but they don't actually say what help they need you know you think about i like to think about the process as part of my machine learning planning because it because based on where our incidents are coming in from and who's responsible for it might change the models we create or the data we use as part of our data set um we for our chat and our portal created incidents those might be templatized and so that might change um our data set in our model definition so when i think about the process in total i also want to think about our sources of information coming in and what i want to use for those sources and that's going to make a difference when we set our business rules or when we activate our predictive intelligence model through flow designer now of course there are going to be situations where we are not using machine learning at all perhaps we have incidents coming in from a legacy system that has transformation rules so there's no machine learning so i think it's a good idea that when we're thinking about machine learning it's not a one and done and this is gonna drive your data in your data set and then of course what happens downstream because there might be opportunities where we have a reproduction going on uh things like that so as an example in my phone-based incidents the agent is adding specific context we might and i call it fire machine learning we're gonna fire it up fire machine learning where we classify the product from the color and text and we might classify our assignment group from our predicted color our predicted product in our text and so we're going to take those inputs and then boom we've got our right assignment group the right assignment group resolves the incident and then they close the incident and then this becomes the basis for information in a table that we call the predictor results this is where we're capturing how accurate our model was now whether the um whether it went to the right assignment group or not whomever gets it might correct the information and then it goes on to the right assignment group and they resolve the incident and close the incident we're going to see that change in our predictor results table but i think that's key to understand that we have a table where we're capturing our original prediction and then our final output value and those two together really help us report and identify what's going on in process now as an example by comparison for email we might choose to fire machine learning with the best first guest and then when the incident is updated we might choose to re-trigger on update because predictive intelligence can be based on business rule activation and so i like to make sure that everyone's aware of the process overall and how our business rules from our solution elements kick into our process okay so step one we do want to understand our data health and i think it's helpful to just create some operational reports on data distribution this is and there is no judgment i don't like to overthink this i want to make sure that i have an understanding of what the spectrum of the data that is going to be the basis for my model is going to be it's neither right nor wrong i just want to make sure that perhaps if 80 of my incidents are assigned to the service desk my model is going to have pretty much one prediction group and then everything else is kind of an outlier so i think a that's where i was talking about a diverse data set for your model you want to have an understanding of the distribution of your data secondly and i do think this is key you want to identify the problem that you're trying to solve and this is where performance analytics can help you when you're identifying the problem to be solved what what are we looking to improve are we hoping to lower reassignment rates are we hoping to um to reduce our aging time our resolution time that um can be found with our performance analytics and we can identify our goal we can set our target and then we can just keep reporting on our incident life cycle to see if we're seeing that improvement based on activating our predictive intelligence now the performance analytics for incident are different from the performance analytics for machine learning our machine learning performance analytics is truly looking at the predictor results table how accurate that model was where it's being applied how it's being used here i'm talking about the broader life cycle how what is our incident um what are our incident metrics what are our goals for our task life cycle and where are we hoping to see improvement and are we going to meet our targets and so this i'm the only one here that's doing this for fun everyone else is trying to improve a process and performance analytics really helps you identify the problem to be solved okay so we are going to touch on quickly creating your model or creating your use case now within predictive intelligence if you go to the menu um and you have your machine learning rule uh you will see a predictive intelligence menu with a home page which i'll go to in a moment and then you'll see your predictive intelligence workbench and so between these two i just want to call out that for classification you can use the home page and define your classification models here but we make it a little bit easier in the predictive intelligence workbench where we tie it to a use case and a business outcome and then in that predictive intelligence workbench dashboard we show you how we're doing against that business outcome now for me because i'm also going through similarity and clustering and regression i tend to stay in the home page to be quite honest but the workbench is very very helpful to compare models to help refine and it's and it's more user-friendly it has a guided setup so i'm going to show you that okay turning it on and i think this is key when you turn it on we have business rules the business rules are really what drive how we apply our machine learning models and if everyone's familiar with business rules then you know you can set your conditions and that's why i covered the process map first because if there are incidents where machine learning does not apply okay then we make sure that the conditions where machine learning does apply are set and so that's what i wanted to bring up to you when we're turning on our business rules first when we're activating but then there's also a business rule for update prediction results and the update prediction results business rule allows us that if a prediction occurred to report on that prediction so two business rules at least in your business roles that will help you identify what's useful okay so we've defined our data we've assessed our goals we have set up our model and we've activated now what now we go through our process we just started entering incidents we there is batch testing on the model definition which i'll show you but in effect when you're using it now our goal is just to get agents started going through their process perhaps we're removing a couple clicks for them or presenting suggestions on where this should go that they're validating but in effect they're going to go through their process until the service is delivered and then we're going to report on the accuracy okay so this table is key the ml predictor results table and alexander i hope this is answering your questions on how you can monitor um your machine learning in progress this is where you can see where prediction occurred initially and then if it was successful on the clothes and so in doing so you can see the multiple options that were available the confidence level the threshold level things like that i'm going to stop for a moment i know i just covered a lot um oh and then we have our dashboards real quick on the dashboards so if you're using the home page for your predictive intelligence we have a prediction results dashboard and this uses the predictive intelligence definitions that were designed or defined from the home page which i'll show you in a moment but we also have a dashboard called predictive intelligence for incidents which i tend to rename so i know the difference between the two the reason why this is helpful is because this helps us monitor our models against our business cases that we define in that in the use case which i'll show you in a moment but knowing where you defined your model is going to drive perhaps which dashboard you use out of the box of course you're always welcome to create your own dashboard okay so how are we doing on questions everybody feeling good i should have had a poll right here um let's go to okay so let's start start with the end in mind so let's create i'm on my service operations workspace and i'm going to just show you quickly um the end result and so i'm going to grab a definition here and in the service operations workspace which is new as of san diego we move this short description up which is very handy um so that we save a couple clicks of getting down to the short description we want our agents to be able to enter the content first perhaps the caller's already identified from an inbound email and populate the information here when we exit out of our required fields i do get a in the service operations workspace i get a recommendation with our classification our assignment group is automatically identified to the software group now this is still editable if the agent does not agree that this is the right assignment group i change it here but i'm just going to save this in the service operations workspace by the way a little bit of introduction to the rest of predictive intelligence within itsm we've seen our classification on our assignment group but here over in recommendations and agent assist we are going to get our similarity and over here in agent assist we can look at similar knowledge articles and um information like this so when we're looking at the broader solutions for predictive intelligence for the agent it's not just about classification we also have a lot of capabilities over here in the workspace and of course we have a ui action that identifies the notification that we've predicted this that's really all we're going for and it should be a non-event right the agent's internet we've got a correct prediction they're going to go ahead and go through troubleshooting it and creating it and when i go to my predictor results i'm going to see this going through the process and i can see in my predictor results perhaps what similar incidents were applied i can see this i'm going to just limit it to a single incident so we can see all the work that was being done for agent assist versus our recommendations for classification now i'm going to clear out my source id and i'm kind of rushing through this because i have a little short on time as we go through the process in this ml predictor results table we are going to identify what was predicted correctly versus what the final output value is so here if a user changed the fi the predicted value versus the final output value we're going to be able to see that and that's how we're going to calculate accuracy but this table is really where all the all the magic around was our prediction correct do we need to refine our model do we need to improve our model what was the final outcome okay stop sharing uh oh wait no i need to continue to share meant to uh go back to my powerpoint so we only have a few moments left i just want to go through model definition and activation so the first thing is of course to activate the plugins so when we activate the plugins just a couple i wanted to bring up to you this is not an exhaustive list it's pretty much the basics we have the core plugin and then we have the enhanced ui and that was the original home page um predictive intelligence enhanced ui for the for the home page listing we have a plug-in if you want to use predictive intelligence for flow designer and there are some great labs for incorporating predictive intelligence with a flow so we have predictive intelligence for flow designer if you want to use predictive intelligence there and the predictive intelligence reports for our dashboarding and then of course if you are interested in the predictive intelligence workbench then that plug-in is available as well so let's look at this really quick so if i go to my predictive intelligence menu so here is an example so if i go into my home page this is where i can see my structured learnings i can refresh my solutions we're providing a lot of information before we even dig in to the solutions but you can see here on what this i probably shouldn't call it legacy but this home page solution is really showing me all the solutions i have available here comparatively the predictive intelligence workbench down here allows us perhaps a more user-friendly approach to classification and here i can create a new model from a template and so what's great about this is we provide guided setup help so if you're new to predictive intelligence and perhaps new to the surfs now platform our guided setup tours really can help walk you through um getting started for um setting up an assignment group you're going to give your use case a name and it's going to be something that you want to accomplish you're going to give your model a name and then you're going to identify your input fields if you go to advanced setup this is where you can identify your additional input fields the example i was giving earlier let's say you want to identify assignment group the output by a color location you're going to want to identify your input conditions so that you can find that color location how cool is this that you can dot walk to the location it's pretty amazing so i think it's helpful to know what's available from the workbench and what's available from the home page and those are two solutions depending on the predictive intelligence solution that you're starting so we've talked about our home page okay really quickly the configuration is mostly the same our goal is the output field so that's our goal and then how do we get that output field from our data what are the inputs that we're going to be looking for right here i'm just using short description but i could be using color location perhaps it's ci perhaps it's something else based on your business process remember this is process built so you want to focus on what is the process we're trying to improve and then of course you have your filter conditions and then of course if you're working in a multi-language environment you'll want to identify your language and your stop words and your training frequency the same occurs on the predictive intelligence workbench where we have the same input fields our output field but we also have a guided setup so the guided setup is very helpful so here's our output for our predicted fields look for your filters for so from all your incidents what is the data we're going to use to build the model and then of course we have our input fields now one of the reasons we perhaps don't want to use all of our incident data is because we might have millions of records it might be a couple years old pre-pandemic which is different from pandemic and post-pandemic so based on your process that's where i say your data wants you want it to be recent and diverse and applicable it's human versus event driven as an example and then from there because you want like for like you want your inputs and your data to have data in those input fields to drive your output fields okay so we activate the model through the business rules which we talked about in the process and that is pretty much it and i know we've got about 10 minutes but here in our creative use cases i'm just going to go back to our created use cases and i'm going to open a predictive intelligence model that i have right here so i've already trained a model i've made my definition and i have the option to create another and i've got a model right here and i and i have the option with all of my models to tune values and here's why this is meaningful this again goes back to what's going on at your company in your process i had a customer that was saying okay most of this works but my network team is three guys and they are overwhelmed already and the last thing they need is an incorrectly predicted and assigned incident so i need to improve my precision and reduce my coverage so if we get it to the network if we get a ticket to the network team remember prediction is never 100 but if a ticket goes to the network team i want it to be accurate versus coverage which is we're going to assign it and if it's incorrect we're going to leave it to the humans to reassign it and so that's really when we're tuning our models we're looking at our precision versus coverage to say i'm going to change this now for each class here i'm predicting my category i only have five category columns but my model is making my best guess that doesn't mean i can't change my prediction so that i can identify perhaps a slightly lower coverage and a slightly higher precision rate or vice versa and here in the solution on the home page i can do that as well so if i go to my category and i open my incident here i can see my overall solution precision recalling coverage and down here i can see the same classes but i can of course change that information where i'm drilling in and perhaps seeing what the other options are available to me and then if i want to tune and refine i can choose apply values and then if i do this at the class level it's going to recalculate at the model level so it's helpful i think to think about when you're tuning your model to think first on perhaps not the overall um value and my demo data is mostly accurate but to look at the classes that are getting assigned and think about the situations in your environment the network team i think provided a great example for me to be like oh okay it's got to be more precise if these are really technical people that are already overwhelmed so i'm gonna make sure that if it gets to them it's spot on okay so that is our recap we covered a lot um so data health that's key identify the problem to be solved create a definition test it um and then just go through your incident process review your results and then look at your data and then update and retrain your model that's i mean once we take out the the i think the heaviness of machine learning i was like oh i got this okay so where do we learn more about this and i think the now learning has some great classes just on the fundamentals of predictive intelligence i learned a lot from just understanding machine learning because i'm jealous of the data scientist on the call i don't have that knowledge but i know how to apply it in this environment and these now learning labs can also give you a safe space to try model definition and model execution i think it's really helpful and then of course we have the community for we have a subforum for predictive intelligence um it's within the itsm pro forum and then of course we have topics from predictive intelligence and i believe we're posting those in the chat those links thank you very much okay how okay so kataya asked how productive intelligence is different from assignment rules thank you so much for asking that question so with predictive intelligence we're predicting an output based on historical data assignment rules are rigid and fixed and so whatever i assign even if it's incorrect it's going to go to that assignment rule so you are going to want to think about where we're applying predictive intelligence because it's prediction that's basic activated by business rules and where we're going to apply our assignment rules um alexander did i answer your question about the pro monitoring the progress of pi i just want to make sure because i saw that in two places okay and santosh asked if someone is using advanced work assignment for auto assigning the incident what benefits will be by utilizing predictive intelligence okay great question so advanced work assignments are running to the right queue based on skills context uh things like that and so what we recommend and again that goes to your process if you're using advanced work assignments because you have additional inputs based on the context of the inbound incident perhaps and i'm going to use virtual agent as an example um you're not going to want to use predictive intelligence for assignment for auto assignment group you might use predictive intelligence for categorization you might use uh the predictive intelligence for similarity for the agent but for classification for the assignment group field you're going to decide either advanced work assignments or predictive intelligence they don't work together i hope that answered your question other questions how are we how are we doing how are we feeling thoughts statements questions who's excited to try this today make make wednesday a predictive intelligence day okay well uh i think that's about it tune in for more live on servicenow webinars and meetups these are great interactive sessions i really do appreciate the um the questions the engagement i hope you found a lot of value uh within the talk that we had today and i am going to wrap it up and give you four minutes of your day back does that sound great thanks everybody