Ensuring business value from Generative AI in EBS

In blog one of this series, I introduced Generative AI in E-Business Suite, and in blog two, I discussed how it differs from AI in Oracle Fusion Cloud. In this blog, we’ll dig deeper into this incredibly powerful technology, and evaluate how to best to use it.

How Generative AI in EBS works

AI natural language queries, the Generative AI being made available to Oracle E-Business Suite (EBS), uses a Large Language Model  (LLM) to convert a text or speech question into a query against the EBS data source. This AI service is available within Oracle Cloud Infrastructure (OCI).

Once it’s set up in EBS, it's lets you ask a question of your data, and then revise the question in a conversational way. You can see the generated SQL as you have the conversation, and then as you revise the question.

I ask: ‘what were my 10 best products sold last year’ and I can see the SQL and the data returned. As I look at that data, I can see that the system is showing me top seller by quantity, which isn’t what I'm looking for. I want it by revenue, so I ask ‘rank by revenue’ and I get a different answer and the SQL is amended.

This is so powerful. Compare this to when a report developer writes a query (and yes, this is an over-simplified example):

• Developer asks business user for a specification
• The specification is: ‘show me the top 10 products sold last year’
• Developer decides how to interpret the request
• Developer writes the code based on that interpretation
• Developer delivers the report for the business user to test
• Business user says: ‘that's not quite what I meant’
• And this may iterate several times before a complex report is written

With AI natural language queries, this is simply a conversation with the data. It’s seamless and fast and you appraise what it returns very quickly.

But here’s the issue…

The hitch is that LLM uses OCI resources every time you make a query. So if you repeatedly run the query from the AI service, it has an overhead cost, and the LLM can use considerable OCI resources. While not a significant amount for one query, it adds up to more than running SQL against your database. This means the AI natural language queries in EBS isn’t  a replacement reporting tool (and Oracle have included some features to ensure people don’t accidentally run up huge bills).

…And a solution

However, the SQL generated by those first queries can be harvested and used in a report, or by a third party tool. This could be a great accelerator tool for your DBA or developers to model requirements.

It's also really powerful in the boardroom. Imagine the CFO, who understands the business, being able to get reasonable and immediate answers to questions.

At Inoapps, we have added AI natural language queries to our award winning Inoapps EBS Archiving product. Since ad hoc queries are why you retain the data in the first place,  it makes this a great use case for AI, versus reports to be run on a daily basis.

But this could quickly go from being a useful tool and amazing demo to consuming a lot of powerful resources. As Larry Ellison says, Generative AI is a fuel and fuel must be used with caution. In the wrong hands, it could be dangerous.

AI natural language queries in EBS is not the most efficient way to acquire data, and it needs to be in the hands of people who can identify that what's being returned has been correctly interpreted by the LLM. So you need controls around who and how Generative AI will be used in your EBS, and there’s a cost to implementation and setting up that governance. Yes, there’s no license cost, but you will be consuming OCI resources, and how much you consume depends on how much it's used and the controls you have around it.

And this brings to mind another saying that I often use in relation to extending cloud applications: just because you can doesn't mean you should.

If you’d like advice on how to get the most out of Oracle AI in your business, speak to the AI experts at Inoapps.

Want to know more about Oracle AI? Explore our blogs