We have been looking for an affordable solution to save some time with data entry - with sales order automation. Currently awaiting a quote for docstar / ecm, but the times we have priced it out previously, it did not fit within our budget. I know there are a mass amount of software out there for ap automation (one being Traild) that we really liked, so I would assume there is something similar for SO. Any advise? Tips, tricks? We have a small office of 4-5 regular users of the software (not very techy), and process approx 400-600 orders monthly. Being able to automate the SO data entry portion - I would assume would save us approx 3-4hours a day of data entry time.
Are you using paste insert?
We haven’t finished it perfectly, but we have proven it works using MS Power Automate.
The PDF orders come into a central mailbox and the flow extracts the PDF. We’ve trained an AI model to read the PDF, and then this is creates the sales order via an api.
Its a little complex and we’ve had to do slightly separate solutions almost per customer, but if its built efficiently its not too difficult to manage. Obviously depends on your companies process for orders and the format of orders you are receiving, but could be worth exploring.
I’m curious to hear more about this, is it something you are hosting locally or a service? Does it cost a lot to run it? Was it one of the pretrained models and you fine tuned it?
Along the lines of not answering the question asked, do you use build order from history? That is a massive time saver if your customers tend to reorder at lot.
What exactly does “automate” mean to you? What are you wanting it to do?
We’re using a pre-trained document processer: Overview of document processing model - AI Builder | Microsoft Learn that you then teach with your own documents. Having it all in one flow in Power Automate does give you access to the Power Apps suite which is helpful - for example its very easy to give the customers shipping address to GPT, and it responds with a formatted address with the right number of fields, correct validation etc that works for Epicor. For us, with a lot of small orders to process every day its a good time saver.
The document processer is a premium app (or whatever they call it), so there is a cost, but Microsoft’s way of billing for this sort of thing I struggle to quite understand. I think you use AI builder credits. You get a certain amount of these with a Power Automate licence, but you’ll probably need more. Depending on the complexity, you could be using a lot per document.
@Mark_Wonsil explains it more here: Kinetic Order Entry Uploading w/SAP - #16 by Mark_Wonsil
EDIT - I should add I’m not the developer who is actually building this, so if there’s any technical questions I may not be best one to answer, but happy to try and help on the basis I know just about enough to be dangerous.
Are you using the browser… Tried doing PA with classic in the early days of it… Was not that successful, but things have progress way much since then
Yes its all in the browser. As we are Epicor on-prem, we have a Power Platform gateway installed on one of our servers for data access.
I thought it would play nicely. Time to revisit it now…
The most pressing question to answer is how is the quality of the documents you need to automate the input of? If these documents are NOT computer generated and NOT (for each customer) following a standard format, stop right there. Not gonna happen.
Yes quality of the document is key - it has to be computer generated and ideally PDF or standard format.
However, in my experience, the standard format is becoming less important. I agree it can’t be completely random, but model I mentioned above was able to pick up the PO numbers in different locations for example. You don’t need to specify the precise location, you just had to say that there was one and it would find it.
I have spent COUNTLESS hours researching this and frankly the complexity of Epicor and every companies different requirements make it quite hard to develop an off the shelf solution.
With that being said I have developed a solution using AI to parse documents out of an Inbox and automatically post them into Epicor. If you are interested you can send me a message and we can discuss it more.
And the best quality document? A JSON blob with all fields tagged. You have the same challenges that EDI would as far as mapping, but that’s with the PDF version too.
Why do we not send a JSON file along with the PDF?
Welcome to the group Nick! This forum is a place to share. Feel free to join the others who have proposed solutions and it may end up in the Expert’s Corner.
It’s an internal API I built that uses multiple AI models to validate documents and extract required fields using OCR. The extracted data is structured with a Pydantic model to ensure consistency and reliability. On average, it costs about $0.20 per document to process.
I use multiple models as a form of validation—if any key fields don’t match across models, the document is flagged as unclear and sent for manual review. This method allows over 90% of orders to be processed automatically without intervention.
The API is integrated with the Epicor API to create orders directly in the system. I currently automate around 4,000 orders per year through this setup.
This is an internally developed system, so I’m not looking to share the code, but the explanation above should be enough for technically savvy folks to replicate something similar.
We have deployed python bot connecting with REST to do exactly this for sales orders and AP. It’s not using LLMs hence almost free to run. We took an approach of strategically automating based on volume of documents where frequent and cookie cutter things are done by machines while humans handle the long tail/infrequent documents. Happy to share all technical details please free to connect.
Could systems not generate an XML/pdf that could be imported into Epicor ECM, I know that many mentioned not wanting to pay for ECM. I do see some of the solutions always coming back to the format of the document. Generally IDC can handle most formats, there are potentially ways to leverage the new AI to capture document fields but again this requires an AI that you can train.
Could EpiUser do a podcast on this topic?
For years people have used PDF format for printing and many systems are built just to format as pdf and make it easier for printing across different types of printers. Software vendors took that easy route and built around it. Even Microsoft attempted an XPS format in past that is in the junk yard somewhere.
JSON format has the ability to offer context and better that PDF but still suffers from the same malady as XML. Using AI while it’s effective, does increase the per unit cost.
Using a local and low cost PDF read and make a best effort to generate a JSON is a decent middle ground where machines and humans can collaborate effectively.