AI Functional Use Cases Procurement

Table of Contents   

1. Opening and Introduction

2. AI as a Toolbox
   

3. Integration with JDE and OCI

4. Types of AI for Procurement

5. Quick Win: Document Vision for Procurement

6. Anomaly Detection and Quality Assurance

7. Exception Handling and Human Feedback Loops
   

8. Agent-Based Solutions 
   

9. Predictive Analytics and Risk Management
   

10. Conclusion and Closing Discussion

Transcript

Opening and Introduction

We are going to go ahead and jump into this. So, um I will start by sharing my screen. And one thing I'm going to say for everybody is uh please keep this uh feel free to keep this interactive. Um you know, I I I have a couple ideas and stuff that we're going to go through in this session, but I by no means do I want this to be a um you know, just a me talking to you you all and you all kind of listening kind of session. So, please uh type things in the chat um you know and and let me know if you have any questions or anything like that. Is everybody able to see the screen? All right, let's go. Great. So, thanks for joining you all. um you know if if any of you heard the keynote earlier right I I am going to be spending most of the week talking about AI use cases and that's kind of what um I'll be focused on mostly with um the participants in um in uh AI week right and and the reason being is because I think where my passions with AI really lie is in how it's actually used right how it really impacts people's lives right I I know there's everyone kind of has their own little thing with it Some people love the the technology behind it. You know, for me, it's more about let's put this thing to work and let's see how it can make things better. Right? So, that being said, you know, I'm going to start with a couple of um you know, quick agenda. So, we're just going to go through uh an overview of AI. Uh nothing nothing uh nothing too crazy, uh nothing too detailed. We're going to talk about AI for procurement and then we're going to go through some fairly uh detailed AI use cases. Now, uh, some of these use cases are ones we've already developed and some are more kind of theoretical, right? So, I'll kind of I'll let you know that this is something that's already been made uh or or being made or this is something that's kind of been ideated with a customer but is in the process of being is is in the process of being developed, right?

AI as a Toolbox

So uh one of the things that I always want to talk about is that you know a lot of times people focus on AI being a singular technology right that it's just this you know we're going to use AI right and and honestly even I'm guilty of it because the I did it earlier to a lot of you in the keynote saying hey how often do you use AI right now if we were being sticklers right there's many different kinds of AI and many different kinds of technologies behind it and it's really about how you learn to use them together. That's that's where the power comes in, right? Um if any of you heard Manuel talk earlier about, you know, kind of how AI and orchestrator get together, uh those, you know, that that that coming together of those two those two tools is really where the power lies, right?

So oftentimes, you know, people have this tendency to think that like, oh, you know, oh, AI is a great solution to every problem. You know, AI is you if you go in YouTube now, AI is being used to solve every problem, right? You you you get AI enabled this and AI enabled that. And and one of the points that I really always like to make is that AI is really not the solution to every problem, but it is a part of the solution to many problems, right? So like like earlier today when we were talking about this with with um with Manuel, AI has been around for all for a long time. If any of you have ever shopped for airline tickets, you know, even as early as the 80s or the 90s, you've been using machine learning algorithms, uh, you know, even that long ago, right? So, you know, AI is just another tool inside of the solutioners toolbox. And that's you all, you all are solutioners. That's why you're looking and you're listening to these webinars and these these these uh conferences, right? You're you're people who naturally come up with solutions. So, you know, AI is just one of the many solutions in your toolbox, right? And inside of that AI umbrella, there's so many different there's so many different tools that you need to learn how to put together to be able to create those solutions. So, so we're going to talk about that today in the context of procurement, right?

Integration with JDE and OCI

One of the other things that I I I really want to touch upon for this is that AI uh and orchestrator and and JDE are are very much natively integrated with each other, right? So, so now with orchestrations and if any of you caught Frank Jordan's session earlier today, if you haven't, please check out the recording. um you are are going to be you you are able to authenticate natively with OCI services using orchestration. So orchestrations can directly authenticate with with OCI, right? Um and and what that means is I because the AI technology we're using and we're working with with Oracle, it lives on OCI. It doesn't mean you have to be hosted on OCI. Doesn't mean you have to be an existing OCI customer. which means that the AI kind of lives on Oracle's cloud uh technology stack, right? Because once orchestrations can authenticate with them, orchestrations can also consume any OCI services that are are used to enable AI. So you know for example if you are using object storage uh for data if for files or data if you're using uh autonomous data warehouse if you're using any of these other OCI capabilities orchestrations can consume those uh natively right and the nice thing is that OCI's AI technology can then extract data can format it and then send requests in a way that's very much readable and consumable by an orchestration.

Types of AI for Procurement

So here in this example kind of the flow as I've laid out is you know orchestrations can natively authenticate with OCI. Somebody can take a picture of a receipt right, that can be sent to an OCI service right which is the document understanding service. Document understanding can read through that service and return back to the orchestration saying hey these are the valid these are the the data points that I need you to be aware of that need to go in and then orchestration can then enter that into JD Edwards as an expense report. So you can kind of see how orchestrator JD Edwards and OCI really are coupled together as part of this right they're all multiple pieces of the puzzle. They're not different pieces, you know, they all kind of fit together.

So, one of the things we'll talk about is all of the different types of AI, right? And again, at any point, if you all have any questions, please type them in the chat. Once we get to the use cases section, I would love to hear people's thoughts around this or if you have any other use cases that you've come up with or you've seen or you've thought about that you'd like to talk through here as a group. The reason I want to talk about the types of AI is because it's really important because not every type of AI really fits with every type of functional area for solutioning. So, if you think about something like augmented and virtual reality, this fits really well into other areas like sales orders and inventory and picking and manufacturing, but doesn't really fit as nicely into procurement, right? Same thing with robotics, you know. So, you know, there are many different types of AI. We're going to be focusing on ones in this session that fit for procurement. So for that that's going to be mostly these guys right here which is around machine learning, predictive analytics, digital assistance and computer vision.

You'll hear me call these different names throughout. So sometimes you'll hear me call computer vision document vision or document understanding. Those are essentially similar things. They're essentially the same thing. You'll also hear me sometimes say machine learning and data science predictive analytics at the same time, right? So those two kind of end up being a little bit more interchangeable when it comes to functional applications. And then digital assistance and natural language processing. So anytime you hear me say the word agent or you hear me use the the new buzzword agentic, right? Gotta love the buzzwords, right? Assume I'm talking about digital assistance and natural language processing. And by no means are the use cases that I'm about to walk us through the an exhaustive list, right? These are use cases that we've come across, we've talked about, we've thought about, or ones that we thought would be important to discuss here as a group because, you know, it kind of gets the ideas flowing, right?

Quick Win: Document Vision for Procurement

So before we go forward, any questions, comments or anything before we start getting into what are some use cases we've seen or have talked through with customers for procurement and AI? Okay. And please, like I said, feel free to chat, type in the chat. Would love to hear comments or questions. So this is probably… Okay. Sure. Sure. Sure. And as they come, as questions come in I will go ahead and answer them as they're coming in too. So not a problem. Yeah I don't see any yet but there is sometimes a delay so… Oh okay cool. Thanks Scott.

So one of the use cases, and this is actually my favorite one because it's really straightforward and very um very it's actually quite simple to do. It's one of the ones where we talk about with quick wins because once we have the base foundation of this made, it's very easy to expand this out to other areas, right? So this is one of the use cases where you can actually take purchase order documents or you can take invoice documents or you can really take contract documents, upload them into an analysis tool, get them over to an OCI service right like we have computer visions and so essentially what happens is the computer vision agent runs through it, extracts information from that contract or that purchase order document or that order form document that you may need to create a PO for. And what it'll do is it can actually enter in service billing or purchase order information right into JD Edwards for you. It sends that to an orchestration, orchestration receives that data and enters that right into E1.

And so if you think about this, like normally rather than having to have a human being look at the document or look at a contract or read through a contract or read through an order form, you would just be able to take an order form, you drop it somewhere, an agent reads through it, it's able to pick up your vendor number, it knows that that vendor number is this address number because it knows to go look that up. It knows that this item is this item number. It knows to go look that up, right? And the way that these things know is that it's not that it just knows these natively, right? We've taught it these things. We've taught the AI agent these things. We've taught it that when it sees this contextual information to go look at this other information. And the nice thing is that as we teach it, it learns these things, right? So we use these reusable essentially services or or bits of skills that we can kind of reuse at multiple places and it just allows us to expand that onto, you know, kind of expand that. So that's what I mean when I say you know once you create a base foundation for it it's really easy to go expand that to somewhere else.

Anomaly Detection and Quality Assurance

So like for example with this we first created a base foundation for it to understand sales order documents or form type documents and then once it kind of learns to extract things like address book information or anything else you know you can reuse those skills in other places. So this is actually one of our more common use cases when we come across this. It's part of what we like to do when we stage these and we talk to customers. This is a really good quick win kind of example because it's easy to roll out. Another use case that I think is really really cool, we did this a while back for a customer. And it's unfortunately it's a bit proprietary, so we can't share too much of it, but I'd love to be, you know, have to do it, you know, get a chance to do it again, which is capturing images of product that you receive on a receiving dock and then using anomaly detection and computer vision to determine if there's any quality criteria or quality issues with that product.

So computer vision has gotten very good to the point where it can detect anomalies in physical goods at a fairly detailed level. Now, I'm not going to say that it's going to be able to distinguish between, you know, three millimeters and 3.5 millimeters, right? So, if you need calipers, you probably are still going to need some calipers. But there are, you know, even with very minute changes in color or size, anomaly detection can really detect a lot of that and the reason it can is because you originally have to feed it like hundreds of thousands of good examples and thousands and thousands of bad examples and then you teach it how to identify the bad examples then it can continue to do that right. So then if you have any of that quality data or quality issues you've captured you can then enter that into JD Edwards, send that to an orchestration and then that orchestration updates quality tests if you're using the quality module or supplier conformance data if you're using the supplier performance submodule inside of procurement.

Exception Handling and Human Feedback Loops

And one of the things you'll notice as we're going through these is you know within any sort of AI solution there's always this possibility of drift, hallucination or non-recognition. Just like a human, it's capable of making mistakes or sometimes just not identifying something. Not to put it colloquially, but sometimes AI can have a brain fart, right? So, you know, the more data and the better data you feed it, the less likely that is to happen, but it still happens because it can come across what we would be like a very anomaly type situation. And for any of those, what you'll notice is we always build in sort of an exception review or a correction process or correction queue because that's very important. And it's important for two reasons. One, you want to make sure that for where the AI has drift and has nonrecognition that a human being is involved. Number two is we want to create pathways for the human to provide feedback back to the AI algorithms so that they can improve. Those are very two very important things. You'll notice that that pattern will be in almost every use case we talk about. We always want to have that pattern for exception handling and correction.

Agent-Based Solutions

Procurement Agent

Now we're going to go through some of the more agentic, agent-based solutions. Right? So one of these, these are again ones this one we have not yet fully created but it's one where because we've created our sales agent, we've created our finance agent, you know the procurement agent unfortunately gets pushed to the last one. So this is going to be one of our last ones we create. But again, the point being is that once you create some of these agents and you teach them skills, it's not like the skills go away. You get to take those same skills and you get to teach those same skills to those agents and reuse lots of chunks of it.

So for example, one of the use cases that I really, you know, we're really working with for our procurement agent is to be able to place orders. So you've talked about in how here in the last couple of examples how the sales agent, using the document understanding, can put in sales orders, how the manufacturing agent can update routings. So it's only natural for the procurement agent to be able to create requisitions. This is kind of the vision of where the use case is that a user can actually engage with an agent to place an item. And again, you can build controls around a lot of these. So if you just have certain catalog items you want users to place items with, the agent can look through those catalogs, find those items. It can confirm that, hey, does the repetitioner actually have the authority to buy this thing? Do they actually have the spend authority? And if they don't, tell them, hey, this is going to go for approval. And then if the agent is able to do it, it confirms everything. It's very simple to get the agent to take that information from the conversation of the user, extract the context of item number, vendor, things like that and enter the requisition into JD Edwards.

Buyer’s Agent

One of the other agents that I think really has a lot of tremendous ability to add value is a buyer's agent. And the reason I think a buyer's agent is important is because as buyers, you know, and I've worked with many a buyer in my clients and during my time as a distribution consultant, buyers are always trying to stay on top of their orders. They're always trying to figure out what's coming in now, what's going to come in next week, what's coming in sooner, what can I expedite, what do I have to push out. They're always trying to stay on top of their orders. So, the idea behind this agent is that it can relieve a lot of the manual work that comes with that updating and upkeep of information in the orders.

So in this example, if a buyer engages with an agent to ask questions about a delivery date, the agent can be taught to go to a carrier's delivery tracking API, get an update, and update that delivery date for the buyer. We can build in questions that say hey do you really want to update this or do you not. In this world of APIs it's very straightforward to do this. Similar to consuming an OCI service, OCI or orchestration can consume a third party API as well.

Supplier Agent

Another use case agent that I think is really important is kind of on that last side of the procurement cycle, which is the supplier agent. This is an externally facing agent which can interact with suppliers to provide payment information, to provide catalog and pricing information or updates. So let's say you maintain a catalog with Hilti or you maybe you maintain a catalog with Staples or someone else, or maybe you maintain a catalog with a smaller local carrier or local vendor for cleaning supplies, for anything else. If your suppliers need to update those catalogs, update pricing, add items to those catalogs, we have agents that can interact with a supplier to do that and then update any of that information.

Now, the reason I like this is because — and you're wondering, you know, why do I want to make it easier for my interaction with suppliers — there are two reasons. One, you free up your buyers’ time. That's the main important thing, freeing up their time. Suppliers have a natural relationship with buyers in AP. When they have questions, they're calling those people and they're eating up their time. Now, that doesn't mean we don't want them to talk to each other. We just want to see if we can drive those groups to have more value-add conversations rather than conversations on, “Hey your department hasn't released payment on this yet, can you push on your end?” or something along the lines of, “Hey I've got a catalog that updates, can I send an Excel file to your IT team?” Those things we can move off of the human being's plate and move onto the robot's plate or the intern's plate. And the other reason I really like this use case is because it gives an opportunity for us to be proactive about information from the supplier. For example, the agent can ask the supplier for updates on delayed orders and then update the delivery date on those orders.

Predictive Analytics and Risk Management

The last couple of use cases I'm going to talk around are all around predictive understanding. And these are really cool and some of my favorite ones and I think the ones that apply the most to the procurement area because it's a lot about understanding and analyzing procurement data. So, you know, we talked a lot about master data and I think this one is a really important one where you can utilize AI tools to identify master data updates. We can create an anomaly detection and pattern recognition agent which can analyze delivery data. It can then identify any outdated or incorrect reorder points or lead times. It can provide those as recommendations to the buyer and then the buyer can accept those and then orchestration will update those for your users, essentially allowing you to create algorithmically probabilistic driven lead times and reorder points.

You know I've met many a customer who always told me, “Oh we haven't updated our reorder times for a long time,” or “We haven't updated our lead times in two or three years.” So having this kind of information staying up to date — and actually the bigger thing is having it be based in reality, not based on what someone is telling you — is very valuable. Similarly, you can also use this to identify orders that are at risk of delay. So you can use similar technology to identify past delivery data, identify suppliers or items or periods or times of events. You can also pair this with geopolitical events or things that are going on in the world, sentiment, news. For example if a shipping lane is blocked because of terrorist attacks or something like that, again you can take all of those things into consideration. These AI agents are very smart. You can teach them to really go and look at information from so many sources, take that information, identify orders with a high probability of delay or non-conformance, send that to a buyer, have a buyer review that and then send that right to an orchestration and have it update data inside of JD Edwards. So again, just kind of that same cycle of using past data to predict future behavior. That's a very common theme you'll find in procurement. You'll also find that in other areas, but it's very prevalent in procurement.

And kind of the last use case on the similar line that we're going to talk about is using anomaly detection to analyze supplier risk. So same thing, there is a supplier performance module in JD Edwards which calculates on-time performance percentage and quality performance percentage, but it's simplistic if you need to start to take into consideration what is the risk profile of the supplier, are they in a part of the world where there are issues going on that I need to be considering, am I relying too much on this supplier, am I stretching myself too thin with this supplier? You can build your risk criteria into this agent. It can build a risk profile for you. And the nice thing is because we have natural language processing paired with this, you can actually have it build you up a risk mitigation strategy for each supplier.

Again, provide those to an agent. With anything like this that's predictive based on past data and probabilistic based on future outcomes, I always recommend having a human look at things because probability is probability, and there's something to be said for human ingenuity. So this is more of the human reviewing the recommendations and accepting it, based on what the agent can do. So I think just to conclude, with procurement what you'll find is there's a lot of great use cases around predictive analytics, predictive understanding, there's a lot of great use cases around agents and a lot of great use cases around document vision and computer vision. So this is just a couple of the examples.

Conclusion and Closing Discussion

Thank you all for spending some time with me as we walk through the use cases. I will be talking through sales order and inventory use cases tomorrow. I will also be talking through manufacturing use cases today. So please join in for those. I love talking about these. And if you've got some time, please drop me some examples of use cases that you'd like to hear more about in the chat and happy AI week.

Yeah, thanks Mo. Last session of the day. Oh, I was on the wrong chat. There are some… okay never mind, those are all just people talking about how they can hear you. Give it one more minute, see if anybody has anything? Yeah. What did you guys think of the use cases? Did you guys think we were there any use cases you were hoping to see that you didn't, or any ones that you weren't really expecting to see and you were like, “Whoa, I didn't even think that would be a use case.” Or you're just like, “Mo, it's 5:00 pm on a Tuesday and I just want to go hang out with my family and go to the gym.” Which is a totally acceptable answer, by the way.

I did put a poll out there. I don't know if you saw that, Mill. It was… oh, now it's not loading for me. Oh, nice. No use cases. Mine's not loading. I guess you're able to see it. Yeah, I see. I see it. That's pretty funny. I like the last option. “No, that's why I'm here.” Well, I hope you guys, all of you guys who voted “no, that's why I'm here,” you got something out of it, right?

Very cool. Well, thank you everybody. I'm going to stick around for a few minutes in case anybody has questions. Thank you all for joining us. Feel free to drop off. Enjoy the rest of your evening today and we will see you tomorrow for more AI jam-packed sessions.