Enterprise Document Intelligence Automating Insights and Detecting Anomalies

Table of Contents   

1. Welcome and Session Overview

2. Oracle AI Stack and Enablement
3. Types of AI in Focus
4. AI Implementation Approaches in JD Edwards
5. Leveraging OCI AI Services and Orchestrations
6. Demo Walkthrough: Document Understanding in JD Edwards
7. Capabilities Highlighted and Future Features
8. Other Use Cases for Document Understanding
9. Q&A and Closing Discussion

Transcript

Welcome and Session Overview

Hello everyone and welcome to enterprise document intelligence uh automating insights and detecting anomalies uh our session here this afternoon at ERP suites AI week. Happy to have you join. I'm here with a colleague of mine Drew Rob who's our AI advisor for ERP Suites. My name is Manuel Nara. I'm an AI product uh vice president uh and AI at ERP Suites. Uh we're happy here to talk to you about um a particular solution that has gotten traction in the marketplace. Uh it's an interesting area uh certainly one of the areas that I I touched upon yesterday in the keynote uh regarding you know some of the some of the functional areas where AI adoption has some traction across industries uh and that is specifically sales. So we'll jump into that. We'll provide some context today and we'll also have a demo today which will be which will be nice. You can see the solution in action and we'll also give you some forward looking type of information and where we're looking at taking this particular solution. So um with that let's go ahead and uh move to the next slide please Drew.

So the agenda is we we'll talk a little bit about you know understanding AI within the context of JD Edwards uh give you a little bit of context uh as well as uh the the technologies that are involved in a uh sales solution for with AI that we'll be discussing. Uh we'll talk a little bit about embedding AI document understanding uh and the demo. So you'll see a live version of that and then we'll also cover other use cases and we'll have some time for questions at the end.

Next slide please.

Oracle AI Stack and Enablement

Okay, just kind of tee this up and uh we we saw this yesterday in the in the keynote and then earlier today where the JD Edwards team presented information uh about JD Edwards and about their AI enablement strategy. Right, this is the Oracle OCI AI stack as you can see right quite a few uh services as part of their portfolio. So pretty expansive rich capabilities and of course as pashant mentioned earlier uh infrastructure right the AI infrastructure is tantamount right so that not only uh can you get results but you can get performant uh you know answers and insights so uh this is this is kind of the breath and we've been spending a good chunk of our time working with Oracle AI uh not exclusively but definitely a good chunk of our time going here uh for the context that uh for the solution that we'll share, we will narrow it down a little bit more to give focus and and you see what technologies we're using for that solution.

Next slide.

Types of AI in Focus

So, types of AI and there's there's quite a few types of AI, but we've we've summarized four of the key ones. Vision uh so if you if uh document understanding is is what we have the session uh title or at least a portion of the title today. document understanding falls under the category of vision uh AI capabilities, right? So, being able to read read documents, read images and be able to harvest that information that is within uh those types of uh files.

uh two rag uh which is uh a capability that allows additional knowledge bases to be uh incorporated as part of your data as a customer and may be able to use. So this is something that we are looking at going forward uh in in the possibility to enrich the functionality of the sales order process. So more to come on that.

Uh next uh machine learning, right? Being able to learn from patterns, being able to uh both both positive and maybe not so positive patterns. Uh you can learn from both as as we all know, right? As humans, we we learn from those two type of experiences. So does AI.

Um and anomaly detection. This is the other component that we have embedded in the solution. looking for anomalous type of scenarios uh as well as conformity type of scenarios that will help enrich the solution. So uh when we talk about the solution in increasing detail and see the demo keep in mind that these are uh at least three of these four technologies are things that we have incorporated one that is on the road map.

Next slide.

AI Implementation Approaches in JD Edwards

So just a little bit before we jump into the specifics of the solution AI implementations if we want to simplify outside of it outside of the the the technology and the cloud and some of the other components that we've discussed are the actual implementations. How do you implement AI? And as we look at JD Edwards there's two major options if you will. One is a digital assistant that you may have sat on other sessions today and even yesterday where you talk about AI that comes by way of an intern right that Mo Shu Jat at ERP Suites likes to refer uh the agent as sometimes I'll call it a companion that is sitting next to a JDR's application and you can interact right with the data in JD Edwards and have the base app next to you.

Um, so that's a digital assistant. An embedded AI is a a an implementation where there is a premium on the AI functionality being embedded within the JD Edwards application itself. So the user the let's say sales rep or customer service rep that is working in sales order entry is interacting with the application and we per and and and the decision is made at company to enhance those capabilities through AI. Well the the the ideal situation would be to have that AI already part of the sales order entry application for example. So that's embedding.

Uh again I'll repeat uh the embedding of AI is not a customization of the base JDR's application. Uh it is extension of capabilities through other methods so that you don't have you as a customer don't have to take a step back and being able to take patches or enhancements for that application from JDM.

Next up

Leveraging OCI AI Services and Orchestrations

Absolutely I'll take it over here. Um yeah absolutely thank you Manuel and and just giving some background there. Um just wanted to give some more information before I jump into the demo. Uh mentioned this in a few other uh presentations I I had at AI week. Uh but we really try to authenticate with these OCI AI services because it is very seamless. Um especially mixing together with orchestrator, right? because the power behind these is definitely orchestrations um and building out orchestrations to get data in and out of of JDO of the JDO system is is very important and very viable um when building out our solutions and I just also wanted to list as well kind of the AI services that we've looked into and manuel harped on them uh genai or rag right being able to read financial documents or large amounts of documents training materials are something we speak about with customers do customer onboarding Um, we actually have a great session tomorrow on code assist to help developers uh develop and test their code. And that's got to be big in the way of the JD Edwards world world as we start to think about, you know, code assist helping us build out orchestrations so we can bulk mass produce orchestrations and start doing that kind of work. Still in the future, but something we're very thinking of very heavily here with using uh orchestrations, especially with our AI services.

Manuel touched on the digital assistance. We have a manufacturing and a financial digital assistant. into the financial loans earlier today that Manuel did. Um and then also we have speech and language um which we haven't harped on too much yet. Haven't really found the place yet in the enterprise. Um but the one we'll be focused on mostly today is is vision really document understanding and being able to read uh PDFs, TIFFs, JPEGs. We're going to focus on PDFs today on reading those um into um JD Edwards.

Um vision just real quick going back to that one is being able to read images as data. Um one example we like to use there is is for anomaly detection. So um for example sending in a part of a pulley um and detecting and testing and detecting whether it's it's defective or non-deective it's complete um is something that we use for in the vision um space. So again document understanding is what we're going to be focused on today in the demo.

Um so with that um just real quick uh we will be doing a sales order entry demo. Um this demo will be inside of JD Edwards as Nwell said. Um we do not want to do any custom customizations. We want to use standard JD Edwards and enhance it right with buttons and make it very seamless very integrated. Um and again just connecting orchestrations um with the OCI service we'll use today is document understanding. So, with that, I'm gonna go ahead and pull that demo up.

Demo Walkthrough: Document Understanding in JD Edwards

Give me one second. Share my screen.

So, with that, I will say that I had to zoom in quite a bit because our faces would not go away, and I know you guys like seeing our faces so much, but I wanted you guys to also see the screen. So, I might be zooming in and out here a few times. Um, but I wanted you guys to see all the features and capabilities here.

Um, so with that, um, as you can see, I'm inside of JD Edwards here, just in the customer service inquiry, the P4210. Um, you can see that we just have two buttons here. So, we have order from image single and order from image multi. This is where we'll be doing the upload of the sales order documents um, inside of JD Edwards.

Um so with that I I just want to show a few of the documents we'll be uploading here. Uh probably just do a few. I have four total but just wanted to walk through them. Um so as I said before um these documents can be you know PDF, TIFFs, JPEGs. Oracle is working to do better at um word documents, Excel documents. Uh new capabilities to come. Uh but as of right now uh you can train those various ones.

Let me quickly just stop sharing my screen. Oh not that one. Sorry, my apologies. Back. There we go. All right.

Awesome. So, yeah. So, as I was saying, so this is just a standard U sales order here. Um, you see at the top here, we just have some information, some more tables. Right now, it's just, you can just see it's just touring bikes. So, you have a red touring bike and a used sport bike. Um, you got the ship two sold two up here.

Um, and then what I really wanted to point out here, which is a cool capability we've been working on, is you'll see there's this is our item number and this is the customer item number. And a cool capability we we brought into a document understanding is the ability to cross reference um when you have these when you don't have an item number in there.

So, for example, if I move over to 2-1, it's the same exact uh order, but you'll see that the um item number is missing. The customer item number is missing there. Um so, with that, um we also have another one here. This is just a different company. So, Cloud9, uh shipped to, sold to a little bit different. Same same uh items um in here though.

Um and then lastly, we have a customer PO sample bad. And by bad, what I'm what I'm showing you is we're missing the ship tool right here. And what we're actually um starting to work on um with this solution is the ability to handle errors uh that might arrive and give you actually um reasons why a sales order might not run through because as you know from testing models, they're not always perfect. People miss things, people miss data, but be able to catch those mistakes and better train on them is is definitely something we're looking to do here.

Um, so I'm going to jump back into JD Edwards again. Um, and as you can see, I'm just going to do a quick order from image single right here. And again, this is working with OCI services and so seamless that I'm just going to click this drag or drop. It's going to pop up the same orders up here. I'm actually going to insert this P2-1 right here.

And then I'm just going to click okay. And as you can see in the top right here, it's with the power of orchestration mixed with OCI document understanding. It's actually reading that document, pulling out all the data, and actually going to input it into the P4210 that you see here. So, just give it a few moments for that to come through. Um, and when it comes through, it'll show a nice submitted uh popup here.

So, as you can see, it's been submitted. The order number is 2009 right here. You can see a sales order. Um, if I do a quick find right here, um, what you're going to be able to see is that the input has been uploaded completely right here. All the information is in here. Get the correct prices as well from JD Edwards. Also uploaded the extended amount. One thing I showed you before was uh, the second item number.

Uh, so as you can see, we were missing the customer item number right here. And as you can see, we were able to cross reference and actually bring that into JD Edwards correctly. Um, another capability I wanted to show just real quick is I'm going to rightclick here and I'm going to go to order attachments.

And what I'm going to show you is that we actually were able to upload the order into JD Edwards. Um, so as you can see, we can do data validation. The order has been uploaded here. Um, and as far as that, there's no work done actually at OCI even though we're using OCI document understanding. Again, this is embedded inside of your JD Edwards environment. Again, it's standard just with a button. No customizations needed.

Um, one other key feature I wanted to show you, and again, we're working on the multi. I can show that today. It just might be a little uh janky showing up here just because you'll see multiple in the same order, but I wanted to show you kind of what we're what we see with error handling. So, again, I'm going to go back to this one. You're going to see we're missing the ship to address or the ship to place right here. I'm going to jump back over to customer service inquiry. I'm going to do another order from image single. I'm going to pull that one in.

So again, just a simple drag and drop as I showed before um right here. And you're going to see the customer PO sample. And then I'm going to hit okay.

And then again it's going to try to grab that data using the power of you know OCI document understanding again uh pull that data and try to import it into JDP4210.

And as you can see at the top here it's it's pretty cool. So you'll see the error executing this specific orchestration that we use for this document understanding for sales order. If I hit the dropown number or the dropown button, it actually determines could not determine ship to name from customer file and it tells you exactly what the issue is. So you can go through and have the human in the loop that could actually change this and better enhance your model um in the future for sales order process.

So with that I'm going to exit out and again really quick seamless demo. I'm going to jump back over to the presentation. So give me one moment here.

Capabilities Highlighted and Future Features

Oh, I saw a question pop up. Do you offer orchestration training that would cover AI components? Manuel, you want to take that one real quick while I open up my presentation?

I will take care of that question. Thank you.

Can everybody see my screen? Okay, good to go. Awesome.

So just jumping forward uh just want to highlight a few capabilities that I went over um during that presentation and what we have future looking what we're looking into doing more with the document understanding piece. Um one thing that we showed you was automated data entry using images. Um the thing is we've been doing this with with other images as well, expense reports, um sales orders, quotes, you can do it with pretty much anything. So think are the possible there with how we can start to automate these images um into JD Edwards and automate kind of these processes.

Um the second one is embedded in E1 as you saw quick button no customizations pretty standard there. Um as far as that capability um single easy to use um attach order source doc as media image. you saw that as well. Um all the work was done inside JD Edwards and you were able to see that attachment link in the top right um as part of the as part of the demo.

Um and then future fe features I mentioned a few before. So quotes um sales orders expense reports when you actually upload a different document um document understanding has the ability to understand various different documents decipher which one it is. Um, and with an enhanced error handling, we can tell what documents what, if it should go to the right place, if it's not going to the right place, how it's read, kind of all those things. Um, it can really understand different document types. So, we can build a lot of different processes, um, using this specific AI technology.

Um, and I showed you the cross reference. We didn't, you know, I said future right here, but the cross reference, being able to cross reference a customer item number with your item number in JD Edwards is also very important.

Um this third point leverages uh supplemental databases. Now with this one um what we really like to indicate here is that it actually is has the ability to understand historical data. So as you train the model and certain customers send in certain things maybe you know we're really training on handwritten um you know files as well and I mentioned Excel and Word are coming up with with Oracle as well. they're still working on getting those getting those read through document understanding and better understood and and more complete.

Um, but with leveraging supplemental databases and and master data, it's really understanding historical customer inputs. Let's say someone spells something wrong or an item number is just misplaced, but they continuously do it. Um, we can use a thing called vector databases to really keep track of all the historical inputs, help train the model. Um, so it provides better better um better features, better capabilities, and provides better um results in the end.

Um, so that's another key feature we're definitely working on. Um, you saw the multi-upload button up in the corner there up on the screen that I showed you before. Um, so being able to batch load multiple sales orders, batch load multiple expense reports, really anything. Um, we actually have that tested and working just fine as well. um just tweaking the error handling on that.

But really um that's all around leveraging. So you saw how I submitted that that sales order and it popped up and you got a little notification there. Really you can leverage E1 notifications as well especially when you upload multiple of them. Um it will show you multiple of them bas basically in the JDwords page. Um you can also set the E1 notifications to send email messages to notify other people when certain sales orders have been uploaded as well. So that's another key feature we're working on um in the future and then we actually can leverage it right now.

Um, so a few of these we got done before AI week, which we're very excited about, but still looking forward, still seeing what we can do as as Manuel mentioned before, you know, potentially connecting this document understanding piece with, you know, once we get the data into JD Edwards, use a digital assistant in the end, you know, that a territory sales manager can use or something along those lines, right? the possibilities are kind of endless and and really growing AI from square one of you know crawl walk run automation prediction autonomous ERP is where we want to strive to be.

So yeah, so with that um going to move on.

Other Use Cases for Document Understanding

I want to talk about a few other use cases and I may have mentioned them already um talking with you guys, but really quality management inspection and and I mentioned it before. It's the vision tool. Um being able to read images as data um send them into JD Edwards for quality inspection, uploading that data into JD Edwards as well. Um is it definitely another piece we're looking into.

Um we actually have a session tomorrow on anomaly detection that we will we will start to look at some of that stuff. Um not so much using vision but more anomaly detection stuff but definitely another piece another cool technology.

Um and then we have supple uh supplier and customer automate compliance as well is another one. Um recording operational transactions. So expense reports um financial documents uh mentioned a few of these before. quotes. Um really anything you can think as operational transactions is where we like to start um with this document understanding piece. I mentioned invoices, receipts, PO labels, um asset inventory recognition and again it's it's all enhancing um how you currently do it. Right? So right no more manual inventory tracking, no more manual uploads, data entry really help with um you know it's more than third party OCI that you can uh train for one thing. it can read multiple documents, cut down on manual efforts, more automation um and it can lead to bigger and brighter things um in the AI world as we start to seamlessly move these tools together.

So yeah, that's uh anything you want to add there Manwell and then just on the right at real quick is just the AI considerations for this tool. So we mentioned vision document understanding kind of go hand in hand there obviously enterprise one orchestration is the backbone for getting the data in the JD works and then you know we we actually move the files using into OCI buckets and once a file is read it's actually moved to another file so that prevents that prevents duplication um with the files there and then lastly we eventually want to hook it up to Oracle analytics again that's that's more the possible I said digital assistant but also Oracle analytics too do some more dashboarding with the data and and and analytics and insights and and all that.

So that anything you want to add there manwell that I may have missed as far as use cases, images, data or the possible anything there?

Well, there there's a lot of possibilities there um in in in this space. One of them being right other types of files, images, being able to harvest information from there. Um, as well as incorporating some of the other components we mentioned early on in the presentation, right? Which would have been, you know, you know, kind of more machine learning. um as well as uh suggesting to the CSR that maybe you know importing these orders or quotes could be as well uh and being able to get some recommendations by the AI agent in terms of upsell cross-ell type of opportunities in those kind of scenarios and and when I'm talking about that capability I'm not specifically talking about spec you know the functionality that JD Edwards has it would be enhancing those capabilities with further kind of suggestions uh that maybe would be beneficial definitely maybe for a newer CSR but also valuable to more experienced one. So certainly there's there's some potential there.

Think we're all good there.

Q&A and Closing Discussion

Yeah, absolutely. So uh yeah, thank you for, you know, hearing about our our new solution. You know, happy to build it. You know, we we love building it out. We love seeing all of the possible and implementing these tools using Oracle services and and you know the are the possible just keep thinking that for sure as you as you go through the rest of AI week and and all the other sessions you go through.

Um with that I mean any questions in the chat we didn't see or we need to answer. I'm sure there's a few Scott.

[Music]

um

yeah. Did you get to the um AI document is a Oracle service, right? I mean that's the one we were using, right? But yeah, that's correct. Yeah. Yeah. And yeah, it's it's called document understanding is the Oracle service that's used.

That's Yep, that's correct.

And can this also work with configurator or configured sales orders? JDE configurator.

Yes. Yes, it can. um it can work it you know the the particular example that Drew demonstrated was not for preconfigured sales orders but uh it's it's the same type of implementation to be some just some tweaks to enable it for those set of use cases that functionality in JD Edwards.

Another comment said sounds like this can replace thirdparty software companies that offer OCR.

Yes that is certainly you know in in another session I had with one of my colleagues mojat right that that kind of question came up as Well um you know legacy right is OCR technology now we're looking at AI technology uh a big differentiator in that kind of you know kind of looking at those two solutions not just modern to say modern right but the AI solution would have uh the capability to learn so it'll be learning and be able to incorporate those right so uh and and and one example of that is also handwriting right so right now I know Oracle is working on that we've tested tested some of those things out and Oracle is improving that. That does require deeper learning uh you know algorithms that they're working on so that they can you know certainly with my horrible handwriting and being left-handed too it can read but it's it's it's pretty impressive how close they're getting to that. So you know those kind of capabilities uh are are kind of you know kind of triggered or kind of pushed for with the AI technology. OCR may would have an issue with the kind of handwritten type of notes.

So that's all I see. I was just going to jump back up to the other question about orchestrator training. Um, and Manuel, you answered that. So I think people can all see that reply, but certainly yeah, that's something whether it's regular orchestrator training orchestrator training with AI. And obviously this is AI week, anything with AI. So head over to our website um fill that out and and set up a quick conversation where we can find out what it is specifically with your organization that you're looking for and and we can find the right people. You know, you you've seen many people today speaking so we'll we'll reach out to that right person, bring them on the call and and we can find out how we can help.

Um can it scrape data from the email body? I don't Okay, that's great question. Do you want me to tackle that one, Drew?

You got it. Yeah.

Yes. So, good question. Um, it can do it for certain types. This goes back to Drew's comment about uh because we're using Oracle OCI AI um functionality. The the additional document types, you know, Microsoft kind of suite of of uh of products, spreadsheets, and whatnot in the queue. They also have a queue for different types of formats of of email. Uh but text files, it can certainly read text files. Uh so if you have that in that format, it might even do HTML. Uh but other types of formats for emails, it it will if it doesn't do it right now, it will soon do it. It's in their road map.

Yeah, I think I mentioned like Excel was one of them as well. It's it's in the road map. Like we we work pretty closely with Oracle. partners for you know we as we test these tools we we try to get as many things that we hear from our customers what how their processes work and that's the biggest thing you know it's like making sure it can read all these things but as man well said it involves deeper learning so but in the road map and hopefully soon.

So so another question could this arguably replace AP automation uh the answer is yes um yeah it can be a replacement.

We'd have to discuss what you know what kind of capabilities you're using and whatnot and we can get into the details right of how we would implement I I imagine folks uh with some of these questions you have insights well how are you doing this how are you doing that right uh you know obviously with orchestrator in the mix that gives us a lot of flexibility and what JD has done to kind of enable the authentic uh the native authentication of OCI services uh but AI has a lot a lot to offer as well right from from an automation perspective. Um so yeah, possibilities are are endless.