hello and welcome everybody to this first edition of ai academy we have a full agenda today but before we get started if you have any questions during the session feel free to ask them in the q a panel and we'll be here to answer that for you this session is recorded and will be shared later on our youtube channel so share that with your co-workers and colleagues if you think they might be interested uh some additional resources as well uh be aware that we have an ai and intelligence community where we share content from experts and people on the field as well as a the servicenow now community youtube channel so if you watching the recording you might be on youtube now uh if you like what you see just press the like button here and also subscribe to the channel so you get the latest updates today's session we will be covering how to get started with document intelligence my name is loic sanchez and i am joined by jean-michel bodwen today for this session a quick agenda for today we'll go through an overview of document intelligence and then we'll have a live building exercise and we'll take questions that we can answer along the way so with that being said i will give the floor to jean michelle thank you lorik and thank you all for investing time with us this morning to learn more about this exciting new servicenow ai capability document intelligence is an ai capability built for organizations who in the course of their daily operations must process large volumes of digital documents be it pdfs jpegs or pngs while some industry verticals rely more heavily than others on data extraction from digital forms i'm thinking about insurance banking health care almost every organization's accounting legal and compliance departments struggle with document processing methods that are resource intensive rigid and very much error prone the challenge of automating such processes is two folds first document formats and structures are too variable and second they change too often over time for template uh for your typical template-based approaches to perform reliably at scale the solution uh that that was released in the servicenow store with san diego leverages predictive intelligence and requires either a financial services operations pro license or an automation engine pro or enterprise license and we'd like to cover before going into the product in and of itself we'd like to cover a few common use cases that were targeted when building this uh this ai capability so the first one is your client onboarding know your customer kyc for short whenever we we within the process we involve uh verification of client ideas being passports or drivers license this is one typical case where document intelligence is very very useful purchase order validation for faster revenue recognition invoice verification for easier expense reconciliation mortgage application processing insurance submission processing and medical bill claims processing as well but instead of taking a typical rpa document extraction approach to these use cases duck intel takes a phased approach that minimizes upstream configuration work and minimize business disruption so instead of having specialists pre-process configure and maintain fixed document definitions that too often break as document templates evolve duck intel removes that step altogether with its proprietary few shot learning algorithm once deployed duck intel lets data clerks handle digital documents as usual within a more convenient user interface than your typical back office system of records and it's one that is purpose built for rapid data extraction and validation as you will see once we get into the doc intel exercise with every different document format that gets processed duck intel learns in the background and the more data clerks use the tool the more the ai is able to assist them with accurate and timely recommendations gradually taking over the load for for the data clerks the more your organization repeats this teach review and monitor cycle the faster it can process documents and focus constrained resources on value-added tasks and this in turn translates into direct positive impacts on the bottom line now without stealing lloyd's thunder here's a a very high level overview of duck intel flow so it first starts with upload uploading digital documents either manually in batch or through a complete integration with your system of engagement be it email crm erp duck intel then runs optical character recognition and its propertary ai models on each page of each attached document this step then returns scored recommendations for every entity that your data clerks are looking to extract and can provide them with the recommendations they need in time and with every human validation of these recommendations and and fields the ai model gets updated so the next documents being processed can benefit from each data clerk's inputs future learning allows organizations to start with absolutely no data so it provides value from day one once you've processed five to six documents of the same equal template duck intel will be confident enough to pre-fill information for data clerks so that they can review the information rather than do the data entry at that point and it does all this while keeping the human in the loop so organizations can minimize their data entry errors and gracefully adapt to change over time without having to go back to the rpa configuration that were set prior so again the more cycles you complete the less effort is required to handle large volumes so you can take on more volume if you if you so wish or you can focus your resources on value-added tasks uh with both approaches directly contributing again to your bottom line so without any further ado and before moving on to the exercise i think like you had a poll we'd like participants to take part in yes i'll start the poll i will launch the pawn now there we go and i will leave it there for a few minutes in the meantime we had a question in the q a it's more related to our forum so i'll take that question pierre is asking us which forum to use because there is an analytics intelligence and reporting forum that we had in the past and now we have a new ai and intelligence forum we understand that our new content for document intelligence and ai search and newer ai capabilities moving forward will be hosted in that new ai and intelligence forum and eventually we all transition everything that was related to intelligence into that new one so yeah uh we starting to get some answers from the poll this is great all right keep it coming we have a uh give it another 30 seconds and in the meantime uh we'll we'll be good i'm gonna close the poll now and we're gonna jump on the exercise thank you very much michelle alright let's jump into our exercise for today so in this life building exercise we'll see how we can digitally transform a legacy process that is based on pdf to process invoices as was said in the overview it's a process that tends to be very tedious and that creates a lot of errors so we want to leverage document intelligence for this use case so before even starting on that i gathered a few sample invoices from the existing process and i want to integrate them on the now platform through document intelligence on those invoices i identify the fields or the values that i want to get from those invoices in that case we'll start with the name of the company and the total amount on that invoice in this first part of the exercise we are going to configure document intelligence what that means is that we are going to create a definition and define the keys which are the values that we want to extract from the documents and then we'll look at how to process our first document and go through the ui for ai assisted data validation we also locate where to find those extracted values and then in the second part of the exercise to take a step further we will leverage the fact that document intelligence is a solution that's fully integrated on the now platform so it can essentially trigger enterprise workflows so to to do that we'll create a reference between a document and a task record on the now platform and then pull the extracted values into that record so that and then we can also update fields such as assigning that task to a specific agent and managing the states so let me go to my instance now so before starting i'm just making sure that i have my plugin installed so i navigated to my plugin page look for document intelligence that plug-in here there's a dependency on predictive intelligence and that macro component for which is the ui that was installed as part of that plugin once that's done i can start with configuration of document intelligence so i'll go to my menu here and look for document intelligence and i'll bookmark and mark it as a favorite so it's easier to access my menu as well and then i'll go to task definitions and i will create my task definition so my task definition is essentially my use case here my use case is invoice so i'll just give it the simple name i'm going to save that and then i'm going to create my keys so again the keys are the values or the entities that i want to extract from my documents so we said there are two of them the first one being the company so i give it a display name and then a more technical name i can submit that and create my second key which is going to be the tool and i can submit that so this is that's it for the configuration of my task definition i can start processing documents now so i will create a first document task here and give it a name so i will process my first invoice invoice number one and i will attach this file to this document task attaching the document invoice one it's a pdf and i can click on process task now i back to my task definition on my document task list here i see that i have my task that was created and i am waiting for the task to be processed by ai by looking at this flag here is processed uh we had an article that we wrote on the community to explain a little bit further all the steps that are happening when this is being processed so as we already explained we upload a document and then we run a set of models optical character recognition and ai to extract the values based on the keys and that take that step takes a few minutes because the task is being processed on the prediction server so that's with we are waiting now but once this is ready i'll see the flag is processed to be turned to true and i'll be able to do my validation as the agent or order the data clerk as we said in the overview after the validation is done the task is submitted and the status of that document task changes to done and we'll use we will remember that and we will use that when we build our flow later so i will go back to my task definition just to check if my task is ready and it was processed it was not processed yet so i'll wait another minute so in the meantime i won't get started on that flow so i will open flow designer and i will create that new flow for now i will just give it a simple name and submit it all right let's go back to the task to see if it's ready so now we see that is processed just changed to true which means that the task is ready or the document is ready to be reviewed and validated for extraction so i can open that document task again and open it in doc intel through that button here which opens the document intelligence ui it's an ui where the the agent can do the data extraction with assistance of ai and since it's our first document we won't see any predictions yet so we have to teach a little bit ai what we want so for the name of the company i start typing the name of the company that i'm looking for and i see already a few suggestions but the predictions are all at zero percent since it's my first document but i know which field i need on that form here so i'll click and select it and then for the tour i will do something similar by starting to type the amount and then selecting uh the area on the form for which i needed that value then i'm submitting that document and by going back to my interface i see that the status was changed to done so what happened to those values that i extracted i'll go back to my document intelligence menu in my navigation here and go to extracted values and this is where i see my extracted values so we extracted two value from that one document that was tied to a document task and we stored the actual value of the the piece of data in the data column so we had the company name and the the tool and these are referenced with a link to the key that was tied to the task definition so we'll use that also later in our flow let me go back to my task definition and i'm going to create a second task so now that we have configured doc intel and we are able to extract values from document we want to start integrating that into a bigger workflow and one of the way we can do that here is when we create that document task and we're going to fill it a second document here i'm going to save instead of processing it and i'm going to create a reference between that document task and a record on the now platform and essentially that other record will be orchestrating my workflow so for this i created a task table that i named invoice task and create a new record on that table so we see my record is all the default values it's not assigned to anybody or any of that so i'll just submit that and i created that reference here between the document and the task and i can stop processing my task so again that step is to take the step that is going to take a few minutes so in the meantime let's build that flow so we want to automatically grab all the values that we extract and store them somewhere on that record and trigger a workflow so to do that we are waiting for the document task which is a di underscore task in the system to be triggered when the status is changed to done and once we do that we want to grab the reference to that invoice task that we assigned there so to do that we create a flow variable and we'll name it our target record id and on my first step here i want to set that flow variable to have for value the reference that was stored on the document task so it's the source record field on my document task i want to do a quick check that if my flow variable is empty it means there is no relationship here so i don't need to continue my flow but if not i want to grab all the values that were extracted so to do that i will do a lookup of all the records that i can find on the extracted value table that are tied to my document task so it's the document the task that triggers my workflow here and once this is done i can pull the values and store them so for now i will start with a flow variable and i'm going to put those values into a flow variable so once i located all the records of my extracted values i can iterate on them so i'll do a for each loop for each of the lookup records and for each of them i want to pull the value and store it in my flow variable so i will for now concatenate them into my flow variable and i want to show um the display value of my key because that's the the nice string that describes my my value so it's the display value of the key that's on that extracted value table and then i want to put the actual piece of data that was extracted and if we remember that one it was stored in the data column there we go we will also add our flow variable so that we get a concatenation of all those values once we are done concatenating the value let's add some text as well so that this text makes bit more sense and it's going to be value extracted and then all my values there and once this is done we have all the values that were extracted from the document and i want to pull them and move them to a to my record that was linked to my document task here and we stored the cis id in my flow variable so now i can do a lookup and i'm looking for my invoice task and i have the cis id and i store it in the flow variable all right and once i have my record then i can trigger my workflow based on task and to do that i'll do an update on that record and i want to pull the values that were extracted from a document and store them into the description field of my task extracted values there we go and i also want to assign this task to apple since i know is the finance manager and it needs to know what values we what what is the the amount and the company we need to pay for this invoice and finally i'm also going to set the state to be pending so that he can see his task on his list of open tasks all right and that's done with this flow so i'm going to activate it so to summarize uh in the first step we set up document intelligence to extract the values and in that second step we just did here we grab those values and attach them to a task record on the now platform so we can orchestrate bigger workflows for our company i'll go back to my task definition and you remember i created my invoice number two a little bit ago so i should refresh the list and it should be available to be processed now just making sure my flow was activated so i can open my invoice number two i will open my invoice task record in a new tab and i see it's still blank so i will process that document so invoice number two was a different invoice than number one but because i processed invoice number one previously i see the predictions here are starting to make a little bit more sense and they actually it's way faster now to grab the first value because the prediction is uh starting to get more and more accurate and then i also want to grab the tool here and i can submit this task and once this is done the stylus of my document task changes to done and it should have triggered my flow that i just built and removed the values that were previously stored in that pdf document i really couldn't do anything with now they are moved into a task assigned to somebody and with a specific state so that way we can orchestrate a bigger workflow and that's it for the exercise today if you have any questions feel free to add them in the q a and if not i will close this session today i'll give you another two minutes um again this session is recorded it's gonna be shared on youtube make sure you register for the next ai academy session and subscribe to our youtube channel so you can get all the latest sessions and in the future any questions so somebody is asking for the youtube channel link uh let's uh follow up on that it should be on youtube it's the now community youtube channel and then we probably will create a playlist for ai that is not yet there all right uh i see there is there we go thank you victor victor is sharing the link of the youtube channel and we'll se we'll share that also on the community but that's it thank you then that's it for today thank you everybody for joining today and we hope to see you in future sessions