good morning good afternoon good evening everybody Welcome to AI Academy please remember that we have additional resources available for you the AI and intelligence Community Forum on the now Community is here with a ton of content from experts on the fields to answer your questions so please remember to go there ask your questions and find the information you need if you're watching this AI Academy on YouTube remember to subscribe to the servicenow now Community Channel so you can get all the latest information and the letters Academy sessions just a quick statement before we get started into academies we tend to only cover what's existing and available in the product today in case we do mention a feature that's coming in the future roadmap please don't take that as a purchase decision and today Welcome to our session on what's new in document intelligence my name is loik Sanchez I'm an outbound product manager for AI products and with me today is Joe hey how's it going my name is Joe Wilmeth I'm a senior principal outbound product manager with automation engine so our goal for today is to give you a brief overview that's only going to take a few minutes and then loik's going to dive into some of the details on the new features that we have available in 2.0 and 2.1 for document intelligence document intelligence is a new application on the now platform enabling organizations to automate and accelerate the process of extracting data from documents these could be both structured and semi-structured documents and integrate that data into larger automation workflows extracting information from documents as you all know is typically a manual time consuming process and it often leads to a lot of errors and rework having to go in and change values or make sure that got the correct field so to address this business challenge customers need a way to automate this process to drive speed and efficiency and to free to employees such as customer service agents for higher value work they need to quickly extract data held in multiple types of documents such as PDFs scans paper documents driver's license IDs so they can act on it and use that data in their digital workflows so with document intelligence we have ai first design that reduces processing time minimizes data entry and is resilient to how that document changes over time and through continual learning fueled by user feedback the AI models learn in real time and this is significantly increasing the accuracy of what is extracted from those documents a document intelligence requires automation engine professional or Enterprise and is also available in financial services operations for Banking and insurance in the latest version of document intelligence we're now on 2.1 it's easier to get set up and get going it enables the path to True automation with little human interaction it offers more features enabling different document and data types such as table extraction and we're looking at better optical character recognition and artificial intelligence producing better results thanks to the new flow designer templates for document intelligence it is now even easier to automate a process in the end when embedding document intelligence into your workflows it reduces configuration time through a low code setup and configuration pre-populated flows are created from templates with the populated data that goes into them based on the task definition this requires you to Simply click and activate the flow to make use of it straight through processing is a new feature in document intelligence that takes document automation to the next level values can be extracted with less involvement from the human agents if a certain confidence threshold is met the values are automatically extracted without validation this allows the business to stay in control by defining that confidence threshold and what that is the confidence score is calculated automatically in the background but when the warning threshold is set manually by the Admin user is seen here on the screen the warning threshold is then compared against that confidence score behind the scenes if the admin user puts 0.7 that means you will see a warning every time the artificial intelligence is less than 70 confident that it got the correct answer and now in this version it is possible to extract tables from documents this is a new data type in document intelligence and agents can extract items from tables even if the number of rows is not predefined and now we have time for our first poll we want to get a little bit of feedback from you all on what's going on out there with document intelligence and how you are looking to use document intelligence whether that be template based forms contracts other use cases HR documents invoices purchase orders a quick question for you Joe uh in specifically in the context of automation engine what are the other products of automation engine that we can integrate docking Terror with yes so automation engine is currently composed of integration Hub RPA Hub and document intelligence and automation Center So within those workflows that you would design with your integration Hub Integrations or spokes within the flows that would go into how you interact with your RPA Bots you have the opportunity to utilize document intelligence for any document extracting that you would need to do or data extraction okay great thank you we have a question in the Q a what are the use cases of talking document intelligence in the HR space so as of today there is no out of the box true integration between our HR products and document intelligence but as Joe just mentioned it's possible to create your own workflows using the automation engine Suite of tools correct all right I think we are ready to close the poll we've got many HR documents and invoices as the main answer let's get moving to the next section cover some of the new features of version 2.1 so document intelligence version 2.0 was released last August and actually today as part of our Talk Show release we're also releasing documented divisions P 2.1 so what's new in v 2.1 there is an enhancements around the simple tables extraction so it is now possible to autofill table extraction under certain circumstances so it means when we add an explicit grid and the table is on a single page and we already processed 10 documents then table extraction is actually automated now and that's coming with v 2.1 when we do not automatically extract the tables the data validation process is a lot easier we have real-time recommendation after the first row and I'll show that in the exercise later on also it's possible to have a reference field between the two tables that we use to extract our documents so we'll see that clearly in the exercise as well next slide we have a question do we have apis for automating the extraction so we'll see today in the exercise how you create your document intelligent use case how you train your model with a few documents and then how you integrate that with flow designer and that should cover your end-to-end workflow and with straight through processing we have automation of the complete extraction process so that and Pro designer should cover your use cases going back to your enhancements so some other enhancements that we have in 2.1 is optical character recognition and AI improvements so as of v 2.1 we can now support document Rotation by 90 degrees increment as well as adding up to 50 attributes now for task definition as opposed to three in the past and also it is now possible to predict whether a field is present or not what does that mean is that for example when you are extracting addresses you know that you have zip codes or state code that are different from a country to another country so the AI would be automatically able to pick up that for specific address type the field is not needed so it won't block the automation there the the OCR engine itself is also better with some improvement extracting text from from document as well as an improved PDF passing engine so that the experience and the results are better all right and with that now we can move to the exercise all right for our exercise today so we're going to be using Dr v 2.1 if you're not sure which version you're using just go to your plugins in your instance and check what version you have installed I see here it's 2.1.0 and that's my version 2.1 we will extract values from invoices today and I have a few use cases and this is the type of document I will be using it's an invoice I have a date and invoice number the company that sent the invoice and to who is what it was sent as well as a table that's what we call a table in the context of document intelligence uh it's uh it's a list of items that clearly could have one or or many items and I don't necessarily know how many items are going to be on my table and that's where I would create a specific configuration for my table extraction here and then obviously my toe the total amount of my invoice is going to be something I want to be looking at so before I start I gathered a few of those invoices so I can put them in my in my dock Intel all right so let's get started with that and I'm going to navigate to Doc Intel task definition and I'm going to create a task definition if you join our previous Academy on dock Intel you probably remember that I'm going to show you some of the new features of virtue specifically but the seller process starts the same way I'm creating a task definition and I'm giving it a name for today invoice and then I'll see my task definition so first thing I want to set up on my task definition here is to Define my target table and that's the table in servicenow that I'm going to be using to both trigger document extraction tasks and extract the value and store them someone somewhere once the extraction is done so in my use case here I'm actually using a table that I called invoice invoice task but depending on your use case just use the table that works best for you so here invoice task and then I'm going to create my keys these are values from the document that are not part of the table so you remember if we go back to our invoice document every invoice is going to have an invoice number every invoice is going to have the name of the company that sends the invoice and every invoice is going to have the tool they might be laid out differently on the page they might be uh yeah located differently in a different format but for all my documents I know I need those values and that's how I set up these as keys so just save my my task definition and then I'm going to create one key for each of those values I need to extract so the first key is the invoice number and what's new here is the Target Field field I can see here that relates to the Target table that I just set up on my task definition and what I'm saying here is one once I extract the value from the document then store it on the field of my table and that way the automation or the workflow works properly so for my invoice number I'm gonna set I'm going to Target the invoice number field I'm going to create two more keys one for the invoice the company that sends the invoice and I'm going to use the invoice company field and finally I'm going to create a a key for the invoice total amount and I'm store that in the invoice amount field all right I am done with my keys let's move to configuring extraction of the table the table extraction is configured via the key groups related list so I'm going to create one key group per table I need to extract let's go back to the document once again in that specific use case with that document I only have one table so I'm going to create one key group my key group would be my my list of items items basically my list of line items here and I'm going to specify a Target table and what we see here is that that table can be different from the table that I set up in my task definition and that's because I might want to store those values in a different tables and that's the case here I actually have a table to store my line item I call that table invoice line item and I'm going to use that to store my table values I'm also going to use a parent mapping to field so I have a reference field on my table that link back to my invoice task that I can use so that I don't so that my records are linked to uh the name table the other main record so I submit that so I create one key group and then I need to open the key group again and I need to create keys and I'll create one key for each value of the elements of the table in that case I am going to create so let's go back to the document again I want to extract the the name of the item and the line total okay that's what I need here so I will create a item name field I'm going to attach that to my task definition I'm going to make sure it's in the right key group and then I can select the Target Field and I see here the the table that is targeted is the table that was linked to my key group so I can use an item description field so there's one key and then I'm going to create a second key again attached to my key group and that would be the line poor value and we'll set that up with the right key group and I'm gonna link that to the line item 12 film all right I can submit that once I'm done doing that I am ready to test this out with a first task so I will create a task a doc Intel task here so that we let give it time to process and then we'll move on to integrating that inflow inflow designer so I'm going to give it a name it's task number one I'm going to attach my invoice to it and click on process task and we're gonna give it a few minutes to process in the meantime let's integrate that into flow designer and so the way this is done is via the integration setups related this here and I'm going to click on new and I'm going to see the creation form and I see here I have two available types I can create a flow integration to process my tasks as well as a flow integration to extract the values once I'm done processing my tasks and we'll start with the process task so I give it a name process invoice process task and every time my task is created as active I wanna a new Doc until task to be created as well make sure the create flow checkbox is checked here and then I can click on submit and that will bring me back to my task definition there we go and I created an integration setup to process my invoice with the flow I'm going to open that flow here because even though it was created as we saw I just need to activate it just checking that everything is created as I expected so we basically have a template now that created my flow based on my inputs from my task diffusion so when my invoice task is created then create a doc Intel task and process it that looks good to me so I can activate that all right I activated my flow designer task here and then I'm going to create the second piece of my integration and that's after I am done extracting the values I want them to be stored on the record that triggered the flow and that's going to be the extraction process and I'm going to give you the type of extract values there when I specify extract values I see I don't have the condition Builder because it basically triggers every time my documental task is processed make sure the create flow checkbox is enabled and then I can click on submit is this available in Rome now dock Intel is a product that is available starting in San Diego all right I will open my flow anything here I created a flow for the extraction process task embedded where studies change to done and then so that's my template that was pre-populated based on my documentary again integration setup what it does is that once I'm done processing my task in the the documentary UI and extracting my values then I retrieved I retrieved with exact values and then I'm going to set them in the record and the values are populated so that's good I can activate that okay that's done all right let's see if the the flow works properly and to do that so I will go to my invoice task table and this one that's where I create a new record so what I'm doing here is I'm using a an invoice task as a way to trigger my doc Intel extraction and then I created Fields here on that records so that I can store the values that I extracted from the document so it's a way for me to integrate the document intelligence extraction Engine with end-to-end workflows on the platform and so with the way I do that now is that I create a invoice task a my invoice to it and then I can submit it and so I created a new task now I'm going to go back to my task definition and look at the related list of tasks and make sure that here indeed I create the the flow that I set up in the integration setup here indeed trigger the creation of a doc Intel invoice task so I can open it I see I have a relationship here between the doc Intel task and the invoice task that triggered that and I can show that in dock Intel do my extraction so I get the invoice number there get the invoice amounts that's the number all the way at the bottom and then I can extract my table values so consumable get 35 460 and then the second line is my Hardware and I have a 22 000 452 or I can submit that It's Gonna Save it and then I can close that all right now if I look at the task the invoice title I created I see already some values so let's take a closer look at that and that's what the second part of my flow designer triggered right after I am done extracting the values they are extracted retrieved and then set on my task so my invoice task here as the the value of the invoice the company the invoice number as well as two line items here that referred to the consumable and Hardware items from the table and the invoice all right when we also want to show is straight through processing so the straight through processing setting can be accessed via the task definition here right here the checkbox to enable I'm going to call it STP and then I can set a threshold here so let me show you on a other instance on tasks that actually train on more than one or two invoices so here I enable STP and I set up my threshold my threshold is the value I decide is my confidence level so in that case 0.7 that's a 70 confidence level in the values that are predicted by AI if that confidence level is achieved then the values on the document are extracted automatically without having to do the manual validation so how do how does that look like if I look at all the list of document intelligence tasks that I have here and I look at the is STP processed filmed here I see that for some of those tasks even after 21 invoices some of these tasks actually were automatically processed it means that I submitted my PDF as an invoice here and the confidence level was achieved 70 here and these values the invoice number the inbox company the amount and the line items were extracted automatically without having to do any manual data validation we can end the session now and remember this is recorded this is posted on our YouTube so if you think that was helpful uh and you want to look at it later go to our Ai and intelligence Community forum and feel free to share that with your colleagues who are also interested that's it for today thank you for joining everybody and we'll see you next time for next session of AI Academy thank you thank you Joe foreign so I will show you how that looks on a in a document that was previously already extracted because uh I don't think we are see any progress being done here but just to highlight what we built today uh we still can extract the company name and the toll that is on the invoice and now we can extract values from from the tables and the way you would see them down on you that's being processed by UI processing properly all right let's try all right yes I apologize for that I don't think we're gonna see the tasks being processed but at least we saw how to set that up so if you install doc Intel on your instance today and upgrade to B 2.1 I'm sure we're going to be able to repeat those steps and make it work for your specific use cases all right if there are any questions please post them on the Q a