good morning good afternoon and good evening everybody Welcome to servicenow AI Academy my name is loik and I'm an outbound product manager at servicenow on the platform AI team and as always I'll be your host for today I want to wish you a happy International Women's Day to all the incredible women out there and today of any day is a great day to celebrate the women in our lives in today's session we'll continue learning about Ai and Hyper automation this is the second and last part of our Series where we explore the role of AI to cut through the noise of data and fight automation opportunities two weeks ago we introduced that concept and we covered automation discovery and again today Yost and Timo are back with us to continue and dive deeper into the techniques to to do that before we get started remember that as always in addition to the session today we have additional resources for you we write articles and so your questions on the AI and intelligence Community forum so I highly recommend you check that out to access that one simple link ascent.works slash AI and if you're watching that on YouTube you can subscribe to our Channel also I want to let you know that we are not the only group doing academies and if you're interested in a virtual agent and nlu or process optimizations they also add their own academies so if that's something you're interested in check out the YouTube channel or on our community to find the link to register just before we get started please take a quick minute to review our Safe Harbor notice and remember that sometimes we may make a forward-looking statement if you wanna when we want to provide more context but we should always ask with your accounting before making a purchasing decision and if you are new today a quick review of what AI Academy is we bring you fresh ideas so that we provide you a better product understanding and practical guidance because of the field of AI is so wide and new we also sometimes cover more of the research or the foundation of AI so we are balancing between Hands-On sessions and general understanding of AI as we mentioned the sessions are recorded and then posted on YouTube so if you are watching that on YouTube or if you want to review the content later you can go to YouTube and if you want to send the session to a co-worker that's a great way to do it but if you are live with us today is great opportunity to ask questions so that we can answer them and for that you just have to use the Q a panel on the zoom interface and we'll answer all your questions today at the end of this the the event you'll get a survey let us know how today went what you liked what you would like to see more of and if you have a specific topic you want us to cover feel free to let us know as well uh just to let you know this year we already covered a AI research some topic around AI search and then later this year we'll talk about document intelligence we'll cover everything that's new in Utah for AI search uh and we are going to introduce task intelligence later as well if you're looking for more of the get started type of content uh feel free to look at the content from last year so last year we covered how to get started with AI search as well as how to get started with document intelligence so I highly recommend you check that out if uh if that's something you are interested and with that I'm gonna hand that up I'm gonna hand it over to our experts to get us started with today's session so I welcome everybody uh this is as low week mentioned part two of our hyper automation Journey or how to use AI to Kickstart your hyper automation Journey my uh my name is Jose from popo I'm a solution consultant for platform intelligence I'm based in the Netherlands and uh joined today like last time with my colleague so welcome everybody thanks so as uh I don't know we've mentioned we have we have a packed agenda um we uh we want to do more of an exercise today compared to last time so we we quickly do a recap of what we discussed last time sort of a framing and then we want to talk about uh Pi mpo so we we are full of acronyms within servicenow uh one of them is predictive intelligence number one process optimization um the exercise will be on clustering and process optimization but first a quick um framing of why are we using those techniques and we we are using those techniques in the context of of hyper automation so what we were looking at last time was this uh Wheel Of Fortune or Circle of Life of hyper Automation and um what we see here is that hyper automation we see that as a continuous cycle but it consists of four steps um discover build and digitize automates and apply Ai and optimize as sort of the the um it's not the exact consecutive Circle you can you can do multiple things at the same time but it's what we want to establish here is that in order to build and digitize or to automate and apply AI with all the capabilities on our platform it's sometimes is difficult to know where to get started and that's where we use the Discover phase and in the Discover phase we have a lot of techniques like automation Discovery we discussed that last time clustering and also process mining or process optimization within servicenow um and then you find for example bottlenecks a bottleneck is an opportunity to automate or to um improve processes by removing approvals those kind of things you do that and you build and digitize phase and eventually you uh you want to see if you if that actually made sense so you you automated something and then you want to see in the optimized phase did I actually did a good job in those things and then you can maybe go again and discover phases see if there are new opportunities uh Based on data generated when you implement processes within service now but as mentioned we are focusing on the Discover phase here and we are focusing on those three capabilities last time we did automation Discovery today is uh clustering or clustering analytics or something sometimes also something you you hear as a term but clustering and Pros optimization and really the the target of those three techniques is find those automation opportunities as we want to find opportunities that we can automate or that we can oh this is something we we can apply one of our AI capabilities to predict uh for example an assignment group those kind of things so um that's what we're aiming for last time we also quickly mentioned uh or mentioned that if you start with discover it's uh you always first need to understand how your how you are using uh service now or how your processes are being implemented so you want to answer well a lot of questions we have some examples here and you want to do that with the following methods doesn't have to be AI you can simply run a report to get some kind of insights uh if we typically talking about clustering there are several algorithm algorithms being used within clustering and some algorithms make more sense when you look at machine generated data so incidents for example that are being generated via alerts um so if you set up a clustering solution you might want to filter based on what you understand in this in this space so what we said there is you use understanding as an input for those discover techniques on the other hand we also mentioned last time that sometimes those those things that you get out of understanding by looking at uh performance analytics dashboard or reports sometimes it's enough to have a direct call to action to actually start improving or to build and digitize Supply AI um also a real small recap what we did last time um so why we use machine learning and that's basically because sometimes data is too big to analyze manually by hand if it's too big you cannot make sense of it so recognizing Trends in in six records is very easy because I can easily see okay out of those six records almost all of them have to do something with connection or VPN and only one has to do something with password or resets account unlocks so quick Insight is there okay 83 VPN 70 password issues however as we said last time that's never the case six records you probably have more in your in your instance so if we're looking at uh 96 000 or 100 000 or 1 million records you cannot do that anymore you you cannot see Trends you can you can try to but you probably will miss something and that's where we want to use machine learning so we want to find Trends in thousands of cases we want to match those Trends to automation capabilities or we want to look at processes process executions if you have a lot of process executions in your systems so running workflows you want to see where those processes are stuck so that's what we're going to look at today we're going to look at um how we can discover more with predictive intelligence and process optimization so we're going to talk about clustering and uh produce optimization those two will be briefly explained before we dive into the instance so we'll I will quickly hand over to Timo to talk you through clustering do you want me to share Timo or yes please okay so clustering first of all clustering is part of predictive intelligence and what clustering is doing and for that I um we use a very very simple example so if you're just clicking one time you see that we want to use and identify patterns with unsupervised machine learning so as you was explaining we have a lot of data in this example we have a lot of fruits just to mention that the raw data the unstructured text is now displayed as fruits um apples and different colors bananas cherries and so on and we have algorithms and what the algorithms are at the undoing that try to structure the text that try to identify a pattern to have some of such raw data have at that point three buckets three outputs one with the cherries one with the bananas and one with apples this don't doesn't need to be all the time the same that's the reason why we have this not very green apple there's blue Apple in there so what we do is really with clustering we're using a large amount of data of unstructured text and try to find same things into that if we go into the next page we're displaying that a little bit different basically this is what it doesn't look like at the end so we can use that to identify which words which fruits are belonged together in such amount of text to uncover very very fast such alternational opportunities visualize those clusters also with Group by functionality and understand the the customers or the employee requests what are the people asking for what is what is incoming in a way of our unstructured text but in a way of what is happening currently with catalog items what is coming in with free text the last one year the last six months or in that channel one the last 15 minutes so very very easy way of having a good machine learning algorithm find out what is the data about where are the topics the concepts about the data back to you used okay thanks Timo um yeah and then the next thing that we're going to discuss today is uh process optimization so also very brief context there uh what you need to think about is when you have every process anywhere within organizations it has been designed so uh there is some kind of flow being designed in mind on how this process would actually run best and then reality happens so then people like are starting to use in your process and they are following all kinds of different routes that not necessarily has to follow the design path um and that's if you if you see reality you also see if the reality makes sense is it more optimal can we do things better uh do we need to maybe redesign our process or do we need to steer reality towards the design process so how do we get from the design parts to a visual representation of reality while process optimization it uses the uh event blocks so any record within oh sorry any records within uh service now like task-based records have a certain life cycle and that has been captured in the event block and that's actually where process optimization gets is um insights from so we we use the event pipelines for all of those records and present them in a discovered process map so we use the event block to eventually visualize uh reality into a so-called process map and that's what we're also going to show you today so you can you can see how your process has actually run um like for example the last quarter or we can see Trends in data for the last quarter with uh with clustering and when you know that you can start applying automation you can sort of tackle bottlenecks coming into your organization to be doing a better job than you did before um and that's where we're going to run you through in the in the exercise so we uh we want to dive into the instance just some some small prerequisites and this can also be be found in the docs but a few things that we want to mention here are the prerequisites for clustering so it's depending on predictive intelligence there are a few plugins that need to be installed in your instance by default the maximum amount of data has 300 records the minimum is two records however that doesn't make sense so like Timo just explained um you want to do this for a lot of Records two records you probably can find a Trends very easily license wise it is part of a pro license so predictive intelligence is part for example of ipsm professional or CSM professional um and what we also we have here are the languages last time when we discussed automation discovery we uh we said those languages are being handled by automation Discovery clustering has a few more so that could also be a choice of using clustering instead of automation Discovery you will see more but you can also use more languages with clustering for process optimization sort of the same thing we have a plugins to be installed that's actually something that um well it's very cool process optimization process mining on the on the on the now platform is a plugin so we don't need to forklift data we can install a plugin we can install a Content pack and then we can actually start process my do process mining on platform data uh license wise this is available with the Enterprise license so itm Enterprise but this can also be purchased as a standalone SKU so you can for example buy process optimization for HR or brothers optimization for itsm for app engine for CSM so there are standalone skus now um the next thing is we want to go into the demo so first thing we will do is that Timo will run you through how to set up and work with clustering so over to your Gmail okay thank you can you see my screen and is it big enough yeah perfect you're good okay first thing that we are doing um I will show you the plugins because plugins leads are limited to loads um I already prepared them here and these are the plugins that we mentioned in the slide deck so it's predictive intelligence um in this project within reports just be aware if you're using that for example for itsm you can also install the predictive intelligence for itsm or for Incident Management and then it comes with a predefined clustering solution for Incident Management for example so there are a couple of examples available that we can also add to our slide deck later if you can find that and download that it will be in there then it's very easy to find just type in clustering from predictive attentions in the menu so we're going to solution definitions and if you can see here in my instance I have a lot of them already configured some of them are for change requests um some of them are for search events so you can use that also for the search results in your instance what are people searching for um you can use that for catalog items for cases for problems for all kinds of tables you can use clustering what I will show you today is something I already prepared as the AI Academy incident all and it's very easy to configure because we describe what to do so the first one that we need to do is we need a name a label that is my AI Academy incidents all then we have something which is different to our automation Discovery the main difference with clustering we have a word Corpus the word Corpus is somehow the vocabulary that we are using to find which fruit types we have on the instance so how many do we have bananas and apples or do we have pies and Elena so so what do we really have on the instance this is the word Corpus if we're going to open for example that one you can see it it's also have a name and we have data into that so in that word Corpus which is um yeah per default used for this incident cluster article and incidents we're using incidents that have been created the last 12 months and we're using knowledge base articles that have been published and for those we use the fields for short description description close notes and for the Articles we use short subscription text and description we have also different types of yeah of the algorithm that we can use some of them are more meaningful on machine generated data some of them are better for free text like the paragraph factor is very good working on this free text approach that I'm doing here with all incidents um it could make sense that we are modifying those word corpuses um in regards to if you have really different departments working with the same process with the same incident process for example the results May differ but if you want to start you can just really start working with article and incidents which is very very good starting point if you're going back to that step number two is select the input data for the clustering um it is in our example incident but we can also choose any other table which is here available so we can use that on foreign so all tables here are available in our example we're using incident and the next thing is we identified the fields that we want to use so what are the texts that we normally use for clustering which is here short description which is good starting point sometimes short description and description could really make sense that we are using those two together sometimes also location makes sense to add location um so you can play around with that for the first time I'm always using just short description to see where we are going and then it depends on how the agents or how your end users are using the short description this is very template based for the sort of scripts is always the same then you may want to go to the field where the free text is so for example description um or if you think that in the resolution notes that is the point where the really the the resolution the really the problem of the users of system that you may want to use the resolution notes field instead of the short description or description field it really depends on how you are using those fields where the data is stored then we are defining the amount of data the amount of data we're doing that with filter and our demo instance is over 80 000 incidents activists folds and short description is not empty but for sure we can also use as we discussed last time for example Channel um is email then we would use doing just a clustering for all email incoming emails or we could use um created last six months or last three months when I would recommend based on my demo data I can show you with a click on that update count it will tell me in last six months generated on this instance one incident this is not enough the minimum is two as we learned um so I will remove that do it again and see that we're at 80 000. then we're using a cluster inside table um this is yeah creating some kind of reporting for the solution that we can have a look later we gotta use Purity Fields um and we're using automated priority Fields what that means is that for each cluster we're calculating the percentage of things like how many assignment groups or which assignment was percentage with category so it gives us really more insights into the data into the cluster that we can find in our example we use group I and we use Group by Channel um we can Group by any other field I will show you results later then we have what is the update frequency in the training frequency but that we can a little bit play around do we want to see the trends of the day for example um this use case is coming normally from CSM um with what is trending issue what is a trending topic but for ensure can also be done on Incident Management level um stopwatch is what words are not getting counted at the end very easily explained um you can add your custom ones company specific ones that you can see like you're using in templates for example and then you have the processing language which you can use if you have mixed languages English is a very good start point for for itsm because a lot of those words that are being using in Incident Management and itsm processes have a lot to do with English right teams Outlook VPN issue password all the things that we pretty much the same um I choose the minimum number of Records per cluster if I have 80 000 yeah I don't want to see classes with two incidents in so I don't want I just want to see the bigger clusters um and this is what we can do here therefore a lot of advanced solution settings so for example the percentage of data that we're processing normally you're trying to do that for 50 of the data for the first run to be very fast and you can change a lot of settings here um and this is how it looks like so you're just gonna tip on update and retrain or just on train and this is what you can then see I have three versions here um so I played a little bit around for you today I go to my active it's true so my version number three I have my solution statistics here so how many records are clustered how many classes I found and here for example is our class of visualization um I can see it now here for self-service and in self-service I can see okay in self-service my biggest one is account access issue user unable with a very high quality that you can see here oh I have a lot of um with email issue Outlook access server urgent working block fix please or is your report Mac so you can also use for example the email one and then we can see again email support login Su portal lock now ee learning so if you can really see it's a clusters it has a word concept and all word Concepts we can see here with the biggest volumes so for example we see um lesson number 66 we can never look at we see our cluster name we see the group by value um we can see all the incidents that have been put to that we can also open that easily and just check what are the incidents about double check if that makes sense what the machine learning algorithm did so what is this about what what the thought description again this is our demo data um we have that one here where the coupon values um if I'm now pressing on generate cluster sample it will also fill me the cluster field number one so I can see the short description immediately here in that list for each time and I can give the thing a name with a name I would say I can go back to show you that in a more so so for example that one is account issues and if you're going in there and start analyzing the data you will find for example if you're going in um for accounts that you can find 10 accounts related things um you can now check them and see oh okay how many are there is the quality good enough so normally I would say um 70 is a good starting point of using and everything less is very but if I was sorted but everything will over 70 has very good quality of saying okay this is really email servers now support welcome learning portal external account developer this is something to do which we can give name for example it's service now support and if you're going through those ones you can combine them you have really a comprehensive look on this amount of data have so many issues related to account issues or related to again this is access issues or also account issues you can then Group by and just show okay these are the ones and you can see that you have almost 2 000 issues in that amount of data and that amount of time have something to do with account issues and then you can start creating your continuous Improvement record try to find out how to deflect those things with that just to give you the half of the time that we are studying through each other today um I will pause it here if there are any questions happy to answer them in the Q a section so I will give again back the sharing to you used I thanks Timo thanks for this overview on clustering um I will I I hope this this all made sense if you have questions we'll answer them in the meantime we'll now switch over to um let me see where is my rights one of my thousand opened there it is so um we're not we're now going to switch to process optimization some things out of the way here so when you are starting with process optimization we uh we the first thing that you want to do um is go to process configurations if you go to process configurations you will actually need to set up a configuration per table that you eventually want to mine so you apply process mining on tables on the event blocks for example on the event logs of the incident table or the problem table or the change request table um I mentioned you can install process optimization via plugin and then there are several plugins for Content packs for process optimization so if you install the plugin process optimization or the content pack for process optimization for itsm then you will get those configurations for incidents out of the box and then you can modify them what we see there is several things I want to take you through you will get settings for automated root cause analysis you will get settings for automation Discovery so automation discovery that we discussed last time in our session two weeks ago is also integrated into process optimization and cluster analysis so what teammate was just talking about clustering is also integrated in process optimization what you do there is you select fields that you want to analyze from a root cause perspective you can simply select any field from the incident table in this case and you can select if you want to auto run this with the model generation So eventually when you start mining it will apply this root cause by default when you mine the model you can set up shorter automation Discovery for pros optimization when you start Mining and again it's sort of something the same that we explained earlier you can say Okay I want to do automation Discovery I want to run this on this short description field so unstructured text Fields also again auto run and the taxonomy so remember from last time we have several taxonomies within automation Discovery you can choose them here and also uh you can use clustering in process optimization and here you can Define which solution definition so when Timo was setting up a solution definition you can select any clustering solution definition that you want that makes sense for in this case the instant table you also get one out of the box if you implement the itsm content pack for process optimization or the series and content pack then we have several uh findings that we will see eventually in the outputs so we will have several findings out of the box again through these content packs but you can also set up your own content packs of your own use cases and basically what you need to think about is for example I will open up one is that you set up certain conditions that are being checked when you start mining so for example solution rejection issue this is something when an issue has been resolved and then the original caller reopens it and so it goes from results to uh work in progress those conditions can be set up and then all those use cases will be checked when the process is mined foreign that's the first thing that we want to do so that's the first thing that we need to do one time when we want to mine the incident table this so this is a one-time setup that you do for approaches optimization then the next thing you want to do is you go to um projects or all projects or create new projects uh here if opens all my projects for process optimization and what you can do is you you can click on new and then you can create a new mining project I will open up one that I have already created and what I see here is um well of course I can give it a name show description I can also give it some kind of categorization which goal I want to um attach to this mining and the next thing I want to do is I want to say on which tables I want to apply this mining so in this case I want I want to mine on the incident table but I can mine on multiple tables so I could for example also mine on incidents and change anti-propub table and then I need to connect those tables via for example what is caused by change so then I can link the incident and change table and then you will see both tables in the output for the Simplicity of only have one table here and actually this is also quite simple so you give it a name in this case the instant table you select the table best practice is always to include approvals um and you will you will state which records you want to mine so think about lost water uh last year as you can apply any conditions on the instant table that you want you select preview here you will see how many records you will mine so approximately mine 31 000 records if I have multiple tables I can apply those rated list conditions um and then if I go to uh what I'm what I want to mine is how I can add uh pills here so in this case I added the state and the assignment group and actually the only only the state we will eventually see in the output but I can select any kind of transition within from a life cycle perspective in this configuration so in this case I say I want to see how State went through the life cycle of these incidents of these 31 000 incidents I will see how State changed and I can also apply breakdowns so I can sort of preset breakdowns that I eventually also will see in my outputs so any fields of the instant table I can select here and then I can say Okay I want to see contact type assignment group or priority in the eventual outputs you'll see that when when I go to the Mind model and then when you have have set up this table configuration you can simply press uh generate model in this case sample or generate model fool if you do sample it will take a small sample of those 31 000 records to see if it's working if it's working you can press full model and then it does a full mining on those 31 000 so this sample is only to sort of test if your configuration is working when done you will get an output and you can jump to the analyst workbench so you can open up the analyst workbench you can also open it up from your menu if you open up the analyst workbench you will go to your process optimization workspace where all your mind's projects are if I open the uh project that I was looking at then what I'm going to see is first of all a dashboard it's loads there it is you sort of Let me refresh this so we'll see a dashboard which is game available via the content pack I have some indicators which I can as of Utah I can very easily add my own indicators for the instant process but this is just reporting what I'm interested in are those in the first of all those Improvement opportunities so remember those findings those findings we were looking for example at this solution rejection issue this is being presented as an improved Improvement opportunity in the outputs and you will see that out of those 31 000 records I have 675 that were reopened or actually rejected by the customer and so I will I can apply any use case and it will be presented here if applicable the next thing I can do is I can for for all of those findings you can open up the analyst workbench by simply clicking the cards I can also open up the analyst workbench in total and start analyzing my process and what I see here is the visual representation of my process and I will see all the uh how the incidents went through their life cycle so I see those 31 000 records I will see how they went from created till process ends what I see here is not the full model so I will I will only see in this case I will see 20 of the connections and I can zoom in or out in this process map so I can zoom in to 40 and I will see more notes um if I fully zoom in for those 31 000 records I will see 1700 routes will become completely non-readable so you want to have a high overview of the biggest routes what an ability you can see here is something like a waiting color info so sort of state that I don't want to see over represented in my model and I actually see here that um almost 10 000 records of those 31 000 went into a whaling color info so went back to the original caller I don't want that so I can start up investigating what went on here and remember we can we we set those root cause analysis um Fields so I can I can see okay what's we're sort of leading influencers to go into this bottleneck and I see for example that the channel portal is steering this so almost all records that had initially the channel portal eventually went into this inefficiency I see here that I have a lot of Bo demo user but any kind of leading influences that pop out will be uh presented here so you see from an original point of the records when it was created uh are there any things that we could see okay all of those went into this negative bottleneck what I can also do is clustering so we can do clustering on a full set of incidents but now I'm I'm only clustering on those 9.6 records that went into the waiting color info so I not only know that things went into a negative State I also know what kind of Records went in there so we can open up this cluster analysis and I can see okay there are a lot of things related to uh email issues um I see here again email issues so I guess some kind of insights okay things that go back to the caller have something to do all the time with email a lot of time with something VPN uh in apparently in India I have a lot of Mac report issues so those insights give me an extra information in my call to action to start solving this particular bottleneck another thing that I want to show you is that we can also um let's see just looking at the timing one last thing I want to show you here is I can also start comparing things so there's a compare feature in process optimization as well so you can for example say I want to apply one of my breakdowns and I want to see email now I see all the records that came in via email and I want to compare this with a process of everything that came in by for example phone so now I can compare two filters with each other and I can see okay how do how does email compare to uh phone for example and I see that's in this case phone is three days faster uh but probably it's also a more expensive Channel and you could do this for any kind of filter not only on filters within your model but also uh filters over time so think about this hyper automation loop again when we mine a model in January we can compare this mines model in January also with a mind model uh in June so you can compare over time if your process did improve over time so see if the automation that you applied in the meantime approaches improvements actually made sense so that's a that's a quick run through on how to set up process optimization how to run your first model and how to sort of navigate into uh this analyst workbench lasting to sort of close back to the original session or the first session is we also have one tab here automation opportunities so if you start mining you can always choose also to run automation Discovery together with process optimization so you will see okay those 31 000 records I can also see what kind of automation opportunities uh are available within those 31 000 records so this ties back to a session of two weeks ago so everything is on one platform everything is studied together and everything will give you insights uh in how to automates improve your processes with the capabilities on our platform that's it for now for for the demo I'm not sure if there are any questions left Timo so some things popping up nope everything answered okay then I quickly go back to my um my PowerPoints um so I I hope you uh enjoyed this quick run through of those uh of those two techniques um and get some insights on how you can use this within your own organization um process optimization and clustering is a big topic uh especially Pros optimization there's a lot more to talk about no week mentioned it in the beginning there are more academies uh one of them I want to point out is the process optimization Academy this is available although this will be a monthly um session on topics within process optimization um so you can you can check those out as well those are also recorded can be played back on YouTube so same concept as the AI Academy and with that's I would like to say thank you to everybody see you know questions left steamo if there are any we can still answer them oh good okay thank you yours thank you Timo that was a great session uh thank you everybody for joining and if you have additional questions feel free to join our community Forum to ask the question and we all make sure to answer those and with that being said I hope we see all of you in two weeks and hope you have a good day bye thanks everybody