Last post we covered what the EDM AI Change Assistant is and how it can help less technical administrators. Let’s face it, most people who work in EDM day-to-day aren’t doing that as their sole function in the business. Most likely, they are administrators for multiple financial applications and have many responsibilities beyond just getting the metadata correct each month. Join me while we see what this assistant, who I lovingly referred to as HAL, can do with some practical examples.
One thing to keep in mind when you are starting your conversation, try and keep your context clean each session. If you hide the chat window, it will retain where you left off from your conversation. If you pick back up and need to do something different, use the little eraser icon to reset the context.
As I mentioned last time, if you haven’t already enabled the Generative AI features, you will need to do that first before you can use my pal HAL. 
To do that, log into your EDM instance and navigate to Tools>Settings. Here you will want to check the box under Generative AI. Oracle doesn’t have this turned on by default, you must opt in. Before you do this, you may want to make sure that your IT security team has blessed using these tools with your company’s data.
Navigate to your choice of Viewpoints. Here I am using the wonderful Oracle sample EDM application. By the way, I want to give a huge THANK YOU to the Oracle team for allowing us to create an instance that has stuff pre-built. It’s a huge help when I am trying to do this kind of stuff or show a new customer how EDM works. But I digress, the sample application has plenty of viewpoints to choose from and more applications built than you can shake a stick at.
I started out in the Account Maintenance view because it has ERP and FCC viewpoints. I thought this might be a typical place to start for most administrators. I figured we may as well start with the most basic question, “What can you do?”
Here are some practical hands-on examples that an administrator might run:
- Show me details and history for Entity C_305
- Set Account Type = Expense for all accounts containing 6
- list members with “x” in the name
As I played around, I took some screenshots to try and show the types of responses you can expect.
One of the above examples, I purposefully tried to give the system a prompt that I knew would result in a validation error. 
Old HAL went along with my request and sure enough, reported the error to me. I also tried to feed it a misspelled prompt to see if that would throw it off, but it knew what I was trying to do.
Each of my requests was done in the same viewpoint and the request items built upon each other. Each query built a new viewpoint query on the side and then created a spreadsheet of request items to add to the open request.
If the assistant can’t understand your intent from the original prompt, it will ask additional questions to refine the selection.
As I played with the assistant for a couple of hours, I learned a few things that I can pass along:
- The assistant can’t create new requests on its own. A user must open a new request, and then the agent can add request items to the open request.
- Security and governance are still enforced. If a delete action can’t be performed on a viewpoint, the AI assistant can’t add a delete request item.
- As the AI thinks, it sends back a little log that gets collapsed. You can use the carat icon to expand the details and see the steps the AI took to get your result.
- The assistant is helping YOU make the request, all of the changes are attributed to your user ID so be careful.
- While this assistant has guardrails, it did give me some information when I posed a design question which was interesting. The Oracle FAQ on the assistant notes that it’s not intended for creating node types or other admin functions that aren’t currently documented.
- It can’t do comparisons yet. You’re pretty much limited to building a node list with a query and then doing request actions against the query items.
- You can take a hybrid approach. Maybe you open a request, tell the bot to do a few things and then finish up manually.
- I had better luck getting the AI to recognize the property names when I put them in quotes. Maybe that was just me, but it wouldn’t understand Alias: Default until I told it ‘Alias: Default’.
- When you’re querying nodes with the assistant, there are specific properties that are and are not indexed for searches. You can find those indexed properties with the little “one to many” symbol next to them in a viewpoint query as below.
To wrap things up, if you’ve mastered the UI and are an EDM expert this probably isn’t the tool for you, yet. I can whip up a Excel sheet with request items a lot faster than I can refine my prompts well enough to do any bulk updates. The power of being able to search nodes a little bit better is promising so I may keep playing a little bit.
The Oracle EDM product team didn’t just come up with a chatbot for EDM, they really built a different way to enable users to query nodes and create request items to work with the properties. All of the governance and security remains intact, the assistant is there to help users do those actions based upon the instructions the user provides.
Just because the assistant helps to build a technically correct request, that doesn’t mean that it’s an appropriate request. My buddy HAL doesn’t know your organization’s politics or policies or data ownership boundaries. It just knows the instructions that you gave it and checks that against your security and validations. I think the most important skill isn’t how to engineer your prompts better, it’s knowing that you are responsible for reviewing what you’ve just told the system to do. We will explore that topic more in the next post.
Discover more from The EPM Garage
Subscribe to get the latest posts sent to your email.