Harnessing Workflow and AI for Business ROI Processing PDF Documents

Hello, I’m Jim from Augusto. Today I’m going to talk about the power of workflow and AI and demonstrate some real-world ROI and how you can use these tools to help your business. So let us start with an overview. We’ll be demonstrating how to eliminate

some manual data entry by using email and extracting content from PDFs. And in addition to that, we’ll be doing some automation. We’ll be moving data from an email system into a backend system. And then lastly, we’ll showcase how we can ensure accuracy and control

with what’s called human in the loop. So let’s take a quick overview. What will happen here is I’m going to demonstrate reading email. We’ll be looking for a sales acknowledgement, it. All right. a document, a PDF that came in, we’ll be using OCR using a tool called Mistral,

which does, AI recognition. And then we’re going to take a look to see if content of that document is there. Does it have a PO number? Is there a ship date? And if so, we’re going to go ahead and store it in the cloud storage. And then in our backend system are either our API or ERP.

In the event there’s not, in the event we’re missing something, we’re going to introduce what’s called human in the loop. We’re going to do a human review to either determine, to take a look at it and change it, or to just accept it and continue going on.

So to do this demonstration, I’ll be using a tool called Natant, which is a workflow engine, and 8n.io. And I’ll also be using Mistral as an API. So I’m going to go ahead and send an email and walk us through this workflow and talk through some of the key events.

Before I begin, let’s take a look at one of the PDFs that we’ll be sending. In this PDF, you can see a sales order acknowledgement, and in it there’s a sales order. In addition to that, there is a purchase order as well. You’ll see that purchase order down here.

We’ll be checking this to make sure those two things are there, as well as a ship date. So let’s go ahead and do that. I’m going to send this along in email, and we’ll watch the workflow run. So I’ve kicked off the workflow, it’s waiting. I’ve sent the email,

it’s sitting in the inbox. And that workflow will be watching it, and it did. It grabbed that email. It’s processing it, and I’ll walk you through the steps here really quick. So the first thing is, we’re watching email, looking for something coming in.

We’re setting some global variables. Those variables could be which system we want it to put into. It could be just extracting some other information. The first thing we do here is we’re looking for attachments, a PDF, and then we’re going to go save it to a cloud drive.

Now let me demonstrate this. This is the actual file. It’s now stored in the cloud. It could be our OneDrive. It could be our image server in the back end, but it’s just a way for us to take what we received in email, move it out of email, and put it into a longer-term storage.

From there, we’re sending this all the way to do, character recognition. We’re using Mistral, so it’s calling an API, and it’s sending those results. What it brings back when it’s done is a complete extraction of what is in that PDF document. We’re then using a tool inside of Natan

that’s called the image extractor. It’s using open AI. In this particular case, I’m using chat GPT 4.1 nano, really small AI work here. And what it’s doing is it’s looking for key elements in there. It’s looking for a ship date and you really distribute that.

Describe that ship date. The shipping date on the document may also be called promise date. same thing. It’s looking for a PO. You’re describing what you’re looking for in that document and AI is helping to review that and bring back the appropriate things. So in this particular document I shared,

you can see there’s a ship date. There’s a PO number. There’s an order number. It has the customer name and it’s got other information. For example, unit price and product price. Now, when this is done, this workflow went through and it said, okay, I’ve received it. I’m going to update our API.

In this case, I’ve set it. So really quick, our ERP system or backend system right here could be anything has extracted that PDF has the PO number. It has the ship date and it also has the as some total information as well as sales order. I’ve made a link right off to that

actual PDF. So that’s going to go ahead and load. We can just bring that up really quick. So there’s that PDF that was sent. And then if I go back to the workflow, you can see that it went into my, document and I put a little logic on here that says,

yep, it’s a valid process. Now let’s look at a document that maybe has a error or something that we want our team to look at. In this case, we’re going to send a sales acknowledgement over that’s missing the PO number. And in this case, what we want to do is we want some human in the action.

It’s called human in the loop. So I’m going to go ahead and execute this workflow and I’ll send that email. We’re checking that inbox to see if the document came in. Here we go. It looks like it’s processing. It’s doing OCR. It’ll do the image extraction and then it’s progressed.

Now I’m going to go over to my inbox for this example of human in a loop. I sent an email out to the people that would be interested. And watching this process and the email states this, please review this order. It says the PO or the customer name is missing

or doesn’t match. You can see here really quick that the PO number shown is empty. And then the customer name is actually a customer name. So you have an option right here to click and view that document. So let’s go ahead and do that just to recap and validate.

They hear that this particular document, the sales order did go to customer name, but there is not a PO number on it. There’s a sales order number, but not a PO. So we can choose to continue. I can approve this and it would work its way through the system or I could decline it.

Declining would stop the process. Approve would let it continue to go. For this video, I’m going to go ahead and hit a break. Approve. So we’ll actually be pushing a document in the system. So I hit approve. And if we went back to the workflow, the workflow received that, you know,

human approval and it progressed. So let’s go take a look at our final system. And you can see here that the document did come through. it was approved. However, it is missing the PO number. If I would have chosen to select the select no and decline it,

it would have declined that message. Just showing the power of how this system with AI can take an email, read what’s in it, process it, and then look for edge cases or any issues to be able to handle those things. AI and workflow here with Natan are very powerful

tools. We do many things to help businesses, both become more efficient, help reduce manual entry, and increase accuracy. If you have any questions, reach out to us at Augusto.