so good morning uh good afternoon from uh the uk uh my name is del cheeseman i said earlier uh welcome to this event how to get started using predictive intelligence to radically improve your i.t service experience and what we're going to do before we get started we're going to go through how our housekeeping so again we always put this up this is our safe harbor notice that we may or may not discuss forward-looking statements or directions of the platform and the products please make sure that this is our legal statement to say please don't plan your deployments features ahead because again things could move so without that next slide uh this for me personally is my la my last live on service now for this section of the year uh but we will be doing future webinars and meetups on the virtual agent predictive intelligence and performance analytics and more and the more that i'm going to call out is i've actually put a request into the team that one of the next webinars that i will do is all around the natural language understanding the nlu that's a big portion of predictive intelligence and i think it has enough uh for it to be on its own so that's what's coming up so please keep a check on upcoming uh webinars and meetups from a housekeeping perspective i could have please ask that you stay on mute there will be q a after the presentation please use the q a feature to ask any questions i'll be monitoring that through the session uh when you if you do come off mute to ask a question please feel free to introduce yourself uh what company you're from uh you know where you are in your deployments and the session will be recorded and shared on the servicenow community after the session and after the session ends you will be prompted to fill out a short survey we really do appreciate your feedback so as i said my name is dale cheeseman and i'm going to depend on my camera very quickly so you can all see my ugly mug here i am good afternoon and i work in the now intelligence team and as a senior advisory solution consultant so our remain responsibilities is really in the hyper automation view and for people who don't know about hyper automation it's our new strategy which encompasses the core technologies that we look after which is performance analytics virtual agent predictive intelligence ai search process optimization so you know we do take a look at all the cool products so with that i'm going to move through to my first slide and the moment people think about ai they either think of the film that has bruce willis in and that child who's a robot or they think about this film because they think it's a lot of code that goes everywhere and it all just magically comes together so i always like this slide just to sort of ease us in but we're going to go through ai we're going to go through the core competencies but this session is more in the back end of how easy this thing is to actually deploy at speed and get the benefits you know rapid so a couple of things i always sort of see a lot of is when we first talking with customers you know challenges faced in these organizations is you know we start off with their mean time to resolve is poor and we start looking into that and one of the things that predictive intelligence can definitely come along and help is you know improving those tickets that get stuck in the in the world of incident or ticket management and when a ticket gets incorrectly assigned it can sit there for a while until somebody picks up realizes the wrong queue and moves to the new queue and two hours could have gone by so that's where a lot of customers you know one of their challenges is that they're getting calls incorrectly assigned which is slowing their mean time to resolve and ultimately it's also you know not helping on their css goals technical assistance to sport staff you know again where can you've got this this wealth of data in your platform but again providing that relevant information to the agent at that point in time helps expedite the agent so as opposed to the agent looking for it you know the predictive intelligence presents it to them so again this is helpful in major incident detection in you know presenting relevant records providing similar changes or recommended changes and so on so it really does help and for me personally you know this is a real big win and is a real way of speeding up your delivery performance straight away because similarity framework actually does no changes to the system it's just presenting information and then the other one that we see a lot of is unable to see larger reoccurring issues because if you're in an organization that's generating you know 30 40 50 000 things a month or you're generating over a million a year you know you're seeing the big major ones but you know throughout the year you could have smaller you know what we say not important issues but accumulatively over the year they are a problem so helping proactive problem management be able to find those cluster of records and as well as that being able to use one of our models or engines to be able to say you know what we want to see where we can shift left where are our virtualized our virtualization records where are our virtual agent potentials so these are a lot of the challenges that some of our customers face on a day-to-day basis so the significance of ai and machine learning is you know we all talk about the human mind is great in solving difficult issues and problems um but when you're faced with mountains of data patterns are hard to find you know humans don't find that very easy and so what we want to say is is that imagine a world where you know you can instantly recall solutions to known issues and problems quickly find those needles in the haystacks of relevant information automatically make accurate assessments such as you know where's the best place to put a task or where's the best category so this is where we start looking at your employees your employees will work faster and smarter make better decisions fewer mistakes and giving them that time to focus on more meaningful work for your business that translates to higher cut higher productivity lower costs increase csat and more motivated employees so it's a less business risk so this is where we start looking at what the what the ai solution is and as you can see it actually spans quite a wide range of our platform technologies what we're going to focus on today is the machine learning frameworks which are the four here and we're going to show you how to get started with these very very quickly one of our other workshops that we're on our other live webinars we're going to do is going to focus on that middle section that natural language understanding in there you've got you've got the workbench you've got the natural language processing you've got querying you've got understanding so that's a big part of the virtual in the virtual agent and then over on the right hand side you've got the search and automation where you're looking at the ai search enhancing your search capabilities bringing in index material showing those genius results and then the other one is the automation discovery so for people who've been on the virtual agent live webinar we talked about the automation discovery being able to look at your data and present information of where you know you could look at shifting left on some of this so this is the core purpose built ai which is directly embedded on our platform so hopefully this is all making sense to you but what can a lot of customers say to me you know where do we start how do you drive the champions transformation and we put it down into five steps establish evaluate identify automate and optimize now again the first one is pretty much where the mistakes or the success is gonna come because you have to establish the solid data foundation you have to understand your data now what i mean by that is you know if you're looking at a modeling you know you understand the data what you're looking at it's that old adage of junk in equals junk out so being able to sort of look at your data and go okay i want to filter out this chunk of information because that could skew my results i've had scenarios where you know we've looked at two years worth of data but halfway through the year they re they recategorized so suddenly we're trying to do a category model based off two technically different sets of data skews the results so understanding your data foundation is is the key to the success then you look at evaluate and what we talk about about evaluate is look at your service performance baselines you know where are you performing right now before you start bringing in this ai data this can be achieved with pa performance analytics but identifying you know what is your current mean times result what are your current open without updates what are your reassignment counts and so on then you start getting into your planning and your modeling so identifying goals and pinpoint areas for improvement so identify from the previous step you're now starting to look at okay i'm seeing a reassignment issue i'm seeing that my mean times results are taking a long time why are they taking a long time and so on so when you start identifying goals you want to pinpoint so i want to reduce mean times result how can i do that get the tickets there faster i can present records to my agents because that will help them resolve the calls faster so this is where you're starting to pinpoint these so with the automation piece obviously utilizing something like clustering to be able to understand your automation candidates and actually applying them into the virtual agent world then you're looking to optimize because once you've done it you know you and you've been running with this looking at your evaluate phase now looking at your optimized phase are we seeing a shift are we seeing that this thing is moving the dial so these outcomes at the bottom is 42 percent of all requests in 2020 fell within 10 categories this is an outcome 75 percent of use cases were identical to the other servicenow customers 21 million dollars saved by automating the top 10 manual processes over five years the virtualization the sixty thousand hours saved by automating the top ten manual processes each year so as you can see you know to achieve these successes predictive intelligence does play a call in some of these so what we're going to focus on again today is we're going to look at the core of these frameworks which is classification similarity regression and clustering now for some people who've been on some of our pi courses already webinars you may have seen this a bit more but the session that we're going to do now is a bit more under the hood behind the scenes the modeling the training the training service and things like that just to show you don't be scared by it it's very very simple so again classification being able to get the ticket classified categorized assigned by basically looking at information coming in from a customer to be then showing you you know that what priority it should be what category should be what assignment group it should be then we're looking at the similarity now the similarity framework for me personally coming from an operational background of over 25 years the framework of this capability is you know what challenges and i've already touched on it but agents spend too much time manually searching for incidents you know or related so to help them to resolve an issue for a specific customer so the solution is to be able to automatically recommend relevant incidents represent them with similar changes or recommended changes detect major incidents i remember many many years ago when i was operational is there was nothing worse than your customers calling you and telling you there's a problem and you didn't know about it so having the model in where it can see a pattern and then alert you to that to say in the last couple in the last 10 minutes i've noticed three of the high packed impact incidents coming in you may want to you may have proposed a major incident so having that ability so similarity for me is one of the quickest wins the clustering framework now the clustering framework uncovers opportunities for improvement with what's called unsupervised learning so solutions that identify patterns and continuously segment them and group them with similar ideas so in essence the clustering framework is is doing clustering analysis and shows you that clustering visualization so it's learning from your records and it's look creating what's called clustering concept and grouping records together to sort of let you see the big picture so when i talked about the million records a year and you've got these small amounts of records throughout the year the clustering will pick up on that and group them together so it actually helps proactive problem management it can help understand where your virtualization is and a lot of my customers in the in the uk who have deployed this they actually use clustering you know every quarter to look at okay we virtualize these conversations let's run clustering again and omit out anything that's virtual agent and see where our next roadmap of conversations for virtualization can come or you can also sort of say you know what noisy records have we got how many item records have we seen in our incident space can we get rid of them you know and put that down the item or item room so this is what the clustering framework is about and then the final one that we'll look at is regression so regression is the common supervised machine learning technique that can be used to predict continual continuous values based on a set of given criterias so what's the challenge here so time to resolution is pretty much a critical metric to ensure that you're efficient and your service management teams and to deliver the best experience for your customer a customer who asks a question and gets a good answer back within a few hours will almost always be more satisfied with the interaction so the need here is organizations can improve their experiences for customers and efficiency across service management by correctly setting expectations monitoring example time to resolve helps set customer expectation identify bottlenecks and areas where more attention is needed so identify more complex incidents that require additional resources and then flag the internal process that are associated with longer resolution times so in essence what i'm getting at here what some customers are doing with this the modeling is learning that numerical calculation of resolve time and some of my customers are presenting that to their customers when a record has been raised for example in the virtual agent the virtual agent can call the regression model so imagine the virtual agent you know trying to resolve an issue then it gets to the stage where it has to log a call because it can't fix it the ticket system is then triggered it logs a call but then it can then at that point call the classification framework get it to the right assignment group it also can call the regression model and say based off your information we're predicting that this could be fixed in 1.2 days or 1.5 hours because it's looked historically at your fixed times and it's making a calculation prediction based off that so hopefully that all makes sense so what we're going to do is before we get into this the demonstration as such or the back end viewpoint just think about these discussion points you know if you're on your journey already some of these may or may not be relevant but getting started where do you where do we start to see benefits with pi and starting place for deployment well i've give you a clue data but getting started for me the quick win again i'll say it again similarity it doesn't change anything in the system it's just providing information and just getting that information to your agents faster can actually help speed them up speak your course speed your agent up resolves calls faster improves me time to resolve improve sees that so then the tuning can we tune these models have we already done this so that you know building a model can i tune the model and then the one that i always finish on is continual cycle do we just set this and forget it no it's a continual improvement process every time the model's retraining is the model better or worse and this is where we talk about tuning and continual cycle they are sort of closely linked so what i'm going to do is i'm going to just stop sharing for a second while i switch screens while i just do this do we have any questions okay so i'm going to stop sharing and i'm going to reshare on my other machine okay so hopefully you can see my instance now again visually just to for transparency this is in san diego and this is the new ui uh visuals so don't worry if yours doesn't look like mine mine is just the new visual uh that we've got here in the format so the first thing i'm going to talk about is back end you know the back end of the solution and i'm also going to touch on well where's it seen in the system so i'm going to talk about these so classification similarity clustering regression this is how you get started now out of the box we give you a workbench in this workbench it has all your use cases so when you look at use cases you can see here create new template so we actually give you templates so if you wanted to get started quick what i always suggest to people is remember you can train a model and it not actually impact your system so training the model just to see the results so here as you can see i've got a few i've got the full system so here i can basically see i can predict category for instance i can predict assignment group i can predict hr these are all guided cell built ready to go okay so i've got this modeling ability this workbench just to say if i wanted to set up quick i could just say i want to do this one i click start and it starts me off on the process but what i want to do is i want to take you through the the long-winded process so you can understand how this comes together so in the classification you've got what's called solution definitions and solutions your solution deaths are your designs your solutions are the results now what i'm going to do here as you can see i did one earlier for european webinar so i'm going to open this one up because this is probably either trained or it's in the middle of training now but what i'm doing is i'm actually basically telling this i want this to train based off my data my category structure so what this is do what it's going to do it's going to basically accept now what i want to talk about here is understanding that you you as the customer are the data handlers we are just the data processors because when i say this statement when you send a solution for training it goes to our training server now the training server resides in the same dc as your instance so we don't load balance we don't send it off to a different dc it stays within yours i don't know why my internet's a bit slow i apologize so what on what's going to happen here is that you're going to see that you're going to specify what information you wish to send to the training server so for example if you're building a model that's going to learn the category okay based off short description and description for one year's worth of tickets and that one year equals a million or let's say another million let's say 250 000 then it's going to send 250 000 tickets but it's going to send the description the short description and the category it's not sending the entire actual solution itself i'm sorry about it i don't know why i'm modeling slow just let me just open up another instance just in case i'm just waiting hello apologize about this so i'm just gonna find one that i need i want so i'm gonna just sort by the incident table so i've got incident and they go so categorization so hopefully this one will be a bit quicker okay so what you're looking at here is as i was saying before is that you're looking at the category that's what your output field is you're learning from the short description you're sending it's a small amount of records here and this is creating the last two years so what you're doing is let me just edit this record so you can see it properly so what this is doing is that's going to be sent to the training system so that's the amount of records that are going to be sent so as you can see it's very very simple and as you can see short description category incident table okay now what this is saying here created on the last two years active is false and state is one of resolved and closed so it's only going to learn of those records okay now again you've got my processing languages what i want to process the machine model in and then i've got this training frequency now this is my demonstration instance so i'm only going to run it once but depending on the volume of data that you're learning from you may want to say if i'm right if i'm creating 60 000 records a week you may want to run a model every seven days because the more data it learns the more intelligent it gets okay so once it's got this you can start to train the model but then when your results come back you end up with something that looks like this so this is my category result okay so once it's gone to the training server and the training is complete it comes back now what you're then presented with is you can test the solution so you could say mail not responding i can do something like that give me the top result so i can click run test it's basically telling me based off my modeling software with a 96 confidence and the threshold was in 65 so again this is what it's telling me but you see down here my category structures are here and it's showing me my models so based off this is where you start to about tuning your model so as you can see this distribution is telling me the volume of records that it had against that particular category so what it's telling you is the precision is that's how precise i'm going to be my estimated precision and my estimated coverage going forward so with a 93 precision i'm going to be able to get an 82 coverage estimated going forward now if i look at this one that was software software here is a 96 precision with a 96 or 97 percent nearly coverage but if i go inside of the software you'll see it's giving me other options so at the moment i'm at 96.44 which is here and it's giving me a coverage of 96 but what i'm trying to do is if i push it up if you look the percentage of push up if i push it up so i'm at 0.44 if i go to 0.6 my coverage is going to reduce so what your tuning is here is you're saying for this category you can be less precise or more precise so what this does is by changing that model and you can just change it by checking the box and a click apply the value once you apply the value gives you a bit of a heads up to say you're changing it you click ok once you've done that you'll then see that you're now at 96.6 with a coverage what's that done to your model well at the high level what it's actually done is it's pushed your precision up or you cover it down because it's like the yin and yang effect so this is all about the volume of data it learns from against a certain category and it gives you this precision and coverage but then gives you the ability to be able to look at you know do i push the precision higher or do i am i happy with the coverage and so on so this is what this does so very very quickly now from the use case perspective you can just go in and say i want to create new from template so calling the templates that you've that you have you can just say i'll just use this way so doing it this way if i said as a category so if i said predict category so same thing i click start give it a name give it a model name advanced setup and i'm just going to basically say here my advanced setup is i want you to say that's two years i think it was in here because i've not got a lot of data in here last two years activities falls now you'll see in a second when i step off this will actually jump in a second and change is one of resolved and closed and then if i go to here and then say look at short description so what this is going to say is and if i want to do a different language so this is what this is doing is false resolved and closed so you see here i can click on this and this will give me a list of the records so you can see here it's 6 000 at the moment so if i so i need to play with this a little bit so if i said remove that and step away it will change so it will actually sort of say to me in a second that's actually a bit a lot more records depend on the volume of data i've actually got in my system so i've only got that minute so so what you're doing now is you just basically say train the model and that trains once the models have trained in this instance if i have any here you can see predict category for incidents so now what you've got here is start the model test the model like we did with the tuning and then integrate the model so it's just an easy way of doing it now what i always say to people is where does this surface in the system it surfaces in the system like this i'm sorry minimize the wrong tab so when you look at the back end and you see the modeling that it's trained where do i see it well from a classification perspective if i go into a new incident here and i basically say i'm going to create a new record now again i'm going to say it's for alvin and i'm going to leave that let me set up to well i'm gonna set up to say hardware i'm gonna send it to the i'm gonna send it to acme refrigerator support i'm gonna say something like um outlook not responding okay that's all i'm gonna do so at this point i'm just gonna step off now you'll see something behave here something up here but at the moment alls i care about is i've set hardware admin refrigerator support and outlet not responding and i'm just going to click save when i click save you'll actually you do actually get prompted with the blue box that tells you it's made a prediction but as you can see here software and i.t client systems engineering because it's learned from your historical models now this is also viewed in here so here's what i prepared earlier so even in the virtual agent when i'm logging the calls this was me logging the call a while ago when it goes through the process in the va when he gets to this point you'll see here it's created the incident but again i wouldn't show this to the customer it's showing my modeling this has gone to the software support group been categorized as software and it's been prioritized as five for planning so again learning from my model so it's relevant here some customers also say to me that's all well and good but i don't use the agent workspace yet well worry not because it's in there as well it's actually in the existing workspace in the existing ui so if your agents are creating records in here and they create a wreck a call here and they say exactly the same say outlook not responding and i'll leave it blank here this time and i'll i'll actually set that to cloud management once i click save you'll see immediately this will also kick in here so you don't have to configure it differently it's working where it should be okay so we've got a question is predictive intelligence workbench covered under isis and pro licensing yes it is is it not showing up in that row right now if it's not showing up in rome it is there so somebody's asked the question predictive intelligence workbench is it available yes it is but they're not seeing it well this is it gets overlooked so often if you go into the plugins and i'll just open that in a new tab what you've got to remember is predictive intelligence isn't just one plugin depending on what you own there's a lot of plugins now i can tell you now the predictive intelligence workbench is but you can turn on the workbench and still not see the templates because there's another plug-in you turn on to give you your itsm conversation your itsm stuff uh another question while i'm just waiting for my page to load i'm sorry if i missed it but but what my coverage do you mean the percentage of records that i've predicted is it the measurement is it the measured in numbers or length of period it's measured in numbers so what this is saying is the coverage is going forward i'm gonna get so if you've got a thousand it's looking at the volume of records that it's learned from i'm going to go forward i i predict 97 of these records i'm going to get right and remember if it doesn't pass the confidence threshold it won't set the record it won't set the field so just going back to this one on the plugins it's a dead it's dead simple type in predictive and spell it correctly um you'll see i get 36 okay now when you scroll through these you see all these different ones for predictive but as you get down at the bottom you've got the predictive intelligence engine then you've got all this stuff it's it's really down at the bottom there's predictive intelligence there's pretty much incident management you'll see there's two in there if you keep coming down you'll see predictive intelligence major incidents predictive universal requests then you get to the workbench now under the workbench you see if you click on this there's the workbench and there is itsm content so if you enable this one it should give you both that one and that one if you have hr as well you can go down here and install the hr one but if you've already installed the workbench but you're not seeing the templates if you see here it sits on it can sit on its own so you can install that so that will give you that entire workbench setup so that will give you this uh that will give you this workbench and it will give you all of this as well so there there's this is here once you've enabled the itsm one you'll see all the created use cases okay now just moving on the similarity framework similarity again i'm a massive advocate i've worked on a service desk a long long long time ago and this for me would have helped me massively so what i'm showing you here is how do you build it where's it shown how you build it is really simple because what you're actually doing here is if you look at similar open incidents what you're actually doing is is basically saying show me similar records in the same table so this is me building my model in the incident table look at your description and compare it to incident your description and i only want you to look at resolved and closed uh so what this does is it's got these records now if you wanted to do similar changes table would be incident test table would be change so this is going to say when i'm creating a new incident look at the change table and do me a comparison is there anything similar so this way you could get similar changes so again the modeling is basically built here of doing comparisons training frequency i've got to run every seven days update frequency every 15 minutes but the only thing that's different from classification at these two fields and this corpus this corpus is mandatory so the corpus is i always say imagine the pattern of wordings in your organization so you're putting a set of wording structures together so when it learns your similarity framework it's looking at how you're positioning word structures together in your organization so when it learns from this when it comes back it gives you a solution now very very simple here there's no real tuning of this the only thing you can do is set your similarity threshold so what you're basically saying here is the degree of similarity between two records records with the similarity score higher than the threshold below will be returned by the solution so what this means is if i test the solution if i say outlook is not responding give me the top five results run test this is basically showing me there's my threshold these are way past the threshold so if if your agent was logging a call or searching stuff i'm going to give him these ones because they're passing the threshold if i then say something like wi-fi it's down give me the top ten run test so you see here unable to predict say top five run test so it's not giving me anything well maybe my thresholds if i push my threshold down to say sixty percent like so test solution wi-fi is down is it gonna give me anything still don't give me anything but if i go back to outlook is down it's not responding and i say given the top 10 you know i'm seeing all other bits of stuff now that might not be relevant so i've got 99 it's still really high so the only thing you can really do here is change this threshold okay now what you can do in here you can look at your examples this will just open up your pages of showing you the similarities of records in your cases so a question from jess should you use the same word corpus for all solutions or is there a reason to have a new covers for a similar um no you can if you've got if you if you create a word corpus for clustering to say look at all incidents in the last six months that are only closed you can use that classification you could use that for similarity but just remember it's it's pulling its wording structures from just those six months of close records think of it another way um if you think about um if i'm building a model to let's say look at classification and in there let's say i'll give you this is a silly example but let's say um you're a gardening company now again you've got a technical department and you've got a gardening part of this now if you basically say to the uh corpus just learn from the incident management's piece so in its eyes a router is a technical piece of rooting software information so in your network but if you've got a customer side of it it could actually be an engine part in a lawnmower a router could be the same thing so if you're building a corpus and say i want you to learn from the incident table and i also want you to learn from the knowledge base so your corpus can look at multiple sources because it's the more wording patterns it gets the better the modeling is going to be because it's understanding in your structure but in essence jess you can use the pattern the close to corpus sorry the word corpus across the the same solutions and i do that very often um yeah so what you'll do is just if you're if you're building so if you're building a brand new model you were basically under the word corpus you just click it and select it and it will be there because you actually build word corpuses separately so they're actually down here so they're actually in predictive times word corpus so your word corpse is residing as you can see i've done this quite a while quite a few times so these were your word purposes resign so all incidents i've got articles and content tasks articles and work order tasks so i've got this looking at multiple things i've got incident and knowledge base last 12 months so you can i can reuse that one multiple times just have to add it in when i'm building a new solution so similarity where does it present itself well you've actually seen it because if you're in the agent workspace okay over here if you go down here and select similar result incidents it presents itself here so this to me is where an agent is logging the call so let's say yes i'm not going to come off mutant role play but me and jess jesse is the customer i need i'm the agent i'm new to this organization she's telling me what's going on now normally i would say log the call and i'll call you back and i'll go and research it because i'm new to the organization so i might not know some of your interfaces now being presented i'm being present with two in pieces of information the only piece of information i have to ignore is that the hours so let's say that says one hour now i'm being present with two sets of information so if i log in i've got i've got jess on the call on the phone and she's talking me through it and i'm being presented with information okay let me just take a look at this one is this similar and i can go into there and copy the incident resolution i can look at the incident itself but i'm also being told imagine this we noticed too high similar impact incidents have been created in the past one hour hold on a minute is something starting to happen i may need to propose an mi so ah thanks jess so so what we're seeing here is the agent can be expedited the agent can be told there's a major incident going on so similarity framework is presented here but i could also say similar knowledge articles is there any similar changes is there any open major incidents so again it's giving me all this information now some people then say to me plus uh i don't have the agent workspace yet we've not deployed it well again you've already seen it because in the incidence table it exists so in incidents and spell incidents correctly as well if i look at creating a new incident so if i want to look at new incidents oh that's not the one that's considered here so if i look in into any it doesn't really matter if i'm in this one the similarity framework presents itself here so if this is outlet not responding and i go to similar open incidents it presents itself here as well so again i can see that as well okay and as i will stress similarity is the quickest win you will get okay so uh what we're going to do now is i'm just going to do clustering because clustering is also a very powerful one so under the clustering section here exactly the same as building um classification what you're doing is you're building the model to be able to um group your records the results that come back look like this so here you can see i've run this a few times uh so if i go into this one this is where i've got to incident all incidents that i've got in my system and i give my cluster visualization and it's grouping the records into clusters i can basically and filter on these so then basically say show me clusters with 400 or more and i can then say show me the quality of ones the quality is the relationship of the record inside the cluster i rarely work here i sort of work down here so what i do with this is the ability to be able to see where my clustering records are put together if i go into one of these it basically shows me what they are so and i've just put this clustering model is just learning off just your description but i could get this to learn off your description description and category so the more data fields it's looking at and it's clustering together this is one cluster with 198 if i built it to cluster on one field or short description description category i could end up with two clusters and these split between two different clusters because the more information information i give it the better the clusters could be so this is what helps me understand from a prior to problem management where i've got reoccurring issues but i actually use this and there is a addition another uh there's a an update set you can apply that gives you what's called clustering analytics now i love this because what this enables me to do is use our predictive intelligence engine to give me this this set of records and imagine doing this for yours and let's say for example let's use our imagination that this one here is password visa i can press this create recommendation button and it puts into the my recommendation space i can then make some calculations and then i can go to my my bosses upstairs and say hey you know what we need to do we need to put in these virtual agent conversations based on our data and in making some recommendations we can potentially see the savings over a three year period if we put virtual agent in so clustering is helping us improve our service expediting our service bringing it together clustering's very powerful and i'm a big advocate for that and then the final one before i just pause for questions and hopefully you've all enjoyed this solution of regression regression is exactly the same as classification but it's numeric so the actual design of it is exactly the same as classification you're basically saying what fields you want to learn from what fields do you want me to set so as you can see it's exactly the same the thing of it is here is predict resolution i've gone for so my output field is resolved time short description the amount of records i want to learn from my modeling once it's trained comes back and here i get a model now again there's no real fine tuning here it's just your confidence level so again if i use that outlook is not responding set 80 and run test it's giving me the worst case scenario in seconds the best case scenario in seconds and then it's making a point estimate so i always do this for customers now again this is demonstration data so ages is a big factor for us but this is looking at this and what that then gives us if i press this into here there it's actually telling me that it would be 56 days which would probably be pretty disastrous if it was a live environment but this is what it presents now we have customers well i have customers that are utilizing this that when a customer logs are calling the virtual agent the virtual agent can call this regression model and they say to the customer we're going to log a call we're going to send it to the hardware team and just so you know this is going to take 1.5 days or 1.8 hours to resolve our expected resolution time so this is what regression is it's more of a numeric but it's not just resolve time as you can see here it actually can do temperature stock prices and so on and so forth so that's what this is used for so i've overrun a little bit here so i'm just going to stop there and i'm going to come back and i'm going to turn my camera on just to scare you all so that's the core competencies now remember we're doing more live webinars and we're doing more and i would ask is all you know join the nlu one if you're looking at virtualization because there's a lot to benefit from that one but i'm going to stop talking and just see if we've got any questions i know we've had questions through and jess thank you for the beard comment um so uh thank you all right so i hope you've all enjoyed it i hope you've got something out of it um if there's no questions uh just as i said tears i'm just gonna show my screen one more time and this is what we always asked we're asked to finish on is please be mindful and again i'm just going to go back to the slide i need please keep an eye out for the future webinars we have got more coming uh and again we really do appreciate you taking the time have we got a question so if there's no more questions i thank you all for joining please be uh take the um feedback at the end we do appreciate that um hope to see you all at knowledge there's a knowledge event in vegas there's a knowledge event in the hague in amsterdam and there's also one in new york i will be in the uk at one of our other events but please you know if you're getting along to the events drop by say hello there will be members of the now intelligence team in all of those uh knowledge events so with that i bid you farewell and have a good day take care everyone