Oracle Communities

Oracle Analytics and AI

Child Item

OAC AI Agents

OAC AI Agents

We're excited to introduce AI Agents in Oracle Analytics Cloud (OAC)—think of them as domain-specific functional experts for your data. These agents are designed to understand your business context and help answer questions in a more meaningful way. By combining your data, business rules, and internal knowledge, AI Agents can interpret natural language questions and provide insights that are tailored to how your organization actually works—not just generic responses.

The Problem: Analytics Bottlenecks

A common challenge many teams face is the reliance on a few experts to interpret data correctly. This often leads to delays, inconsistent answers, or difficulty scaling analytics across teams. AI Agents help solve this by acting like always-available Subject Matter Experts. They are configured with specific datasets, guided by business instructions (like definitions, rules, or defaults), and enriched with relevant documents. This allows them to deliver more accurate, context-aware answers—whether it's applying the right metric automatically, using the correct business terminology, or filtering data appropriately—helping users get faster and more consistent insights.

Best Practices for Getting the Most Value

To get the most value from AI Agents, a few best practices can go a long way:

Keep each agent focused on a specific business domain — agents perform best when they are specialized.
Define clear and simple instructions (such as default metrics, filters, or naming conventions) so responses stay consistent and aligned with business expectations.
Include relevant knowledge documents like policies or FAQs, which help the agent provide richer, more accurate answers by referencing trusted internal information.

The Future of Conversational Analytics

As organizations move toward more conversational and AI-driven analytics, AI Agents have the potential to transform how users interact with data making insights more accessible to everyone.

How do you see these domain-specific functional expert agents fitting into your workflows?

Could they reduce dependency on traditional experts, or complement them in new ways?

Let's open the discussion and hear your thoughts.

Find more posts tagged with

Agents

OAC

Oracle Analytics Cloud

AI Agents

OAC AI Agents

Comments

Shiva Kumar Oleti-Oracle

Let's open the discussion and hear your thoughts

Aimee Williams-Oracle

Thanks for sharing, @Shiva Kumar Oleti-Oracle! This feels like a natural evolution of how people want to work with data. Less about knowing exactly where to look, and more about being able to ask questions in a way that makes sense. Very neat!

Benjamin Arnulf-Oracle

Very nice!

Bhaskar Konar

Great Overview!

Really like how AI Agents are positioned as domain-specific experts, especially the focus on business context and consistency in insights. The best practices are very practical as well.

This definitely feels like a strong step toward more scalable and accessible analytics.

Thanks for sharing @Shiva Kumar Oleti-Oracle!

Doug Ross

My challenge so far is how to tune prompts and supplemental documents to ensure correct results. I have tried creating agents with basic datasets and what I believe are clear instructions but when I ask the agent a question, in many cases, the response I get is that the question can't be answered or my instructions are ignored.

For example, I have a dataset with invoice records. An invoice has a cost center description.

I have the following instruction:

When a "Cost Center Description" contains any of the following words: "Basketball", "Tennis", "Baseball", "Softball" they should be considered "Athletics" cost centers.

If I ask for the top 5 athletics invoices, instead of using the above prompt, it only returned rows where the actual word "athletics" was in the description.

As a test, I used the same dataset and instructions in Grok and it answered the above question without any additional prompts.

Or, as another example, I included the following KPI definition:

KPI : Time To Process (TTP):
When users asks about TTP (Time To Process),
use the following formula to compute : "Days To Approve" + "Days to Issue"

Asking for the TTP for certain invoices came back with a generic "Can't answer that" response.

So how do I debug these issues that would be trivial to do with a regular semantic model or as a calculation on an OAC analysis?

Doug Ross

Note: I tried to follow the guidance on the prompts for agents in this blog post:

https://community.oracle.com/products/oracleanalytics/discussion/28236/supplemental-instruction-templates-for-oac-ai-agents

mgoswami

Thank you for starting this discussion @Shiva Kumar Oleti-Oracle . I agree keep agents as specific e.g. domain or even offerings level withing domain.