By Swathi Ambati, HEXstream solutions engineering manager

Imagine asking your business data a question the same way you would ask a colleague and getting an instant answer, complete with a chart. No waiting on the data team. No report request. Just type what you want to know and see the result.

That is exactly what the Oracle Analytics AI Assistant delivers. Built directly into Oracle Analytics Cloud (OAC), it is available as a default feature across all OAC shapes—no special configuration or minimum CPU requirements needed to get started.

In this series, we have covered what AI assistants are, why they matter for utilities, and how to make sure your data is ready for them. Now we want to show you what this tool actually looks like and more importantly, what it feels like in a real utility environment.

So what does the Oracle Analytics AI Assistant actually do? 

Good question. 

At its core, the AI Assistant lets anyone ask questions about data using plain, everyday language. You do not need to know how to build reports or navigate complex dashboards. You just type what you want to know.

The screenshot below shows exactly what that looks like in practice. It is a storm-outage dashboard with metrics on customers out, outage events, restoration trends, and an outage map with the AI Assistant panel open on the right. A user has typed: “Which storm had the most customers out?” The Assistant generates a bar chart instantly, showing Storm Alpha with 18,039 customers out. No filters touched. No report built.

Figure 1: A utility end user asking the AI Assistant a question alongside a storm outage dashboard the answer, Storm Alpha with 18,039 customers out, is generated instantly.

This is the end-user experience. The dashboard stays visible on the left. The Assistant works alongside it on the right. Users can keep asking follow-up questions without ever needing an analyst in the room.

Beyond answering questions, the Assistant can also:

•       Change chart types on command just say “change this to a line chart”

•       Add or remove filters and data fields using natural language

•       Save any visualization to a Watchlist for quick reference later

•       Work in multiple languages including English, Spanish, French, German, Italian, Portuguese and Thai

•       Support both dashboard builders and everyday readers

One note worth keeping in mind: the AI Assistant uses a large language model that works by recognizing patterns in data. That means it can occasionally produce answers that look right but are not perfectly accurate. Oracle’s own guidance says to verify important answers against your source data before making critical decisions...a sensible standard in any regulated industry.

What It looks like behind the scenes

The Assistant does not automatically know about all the data in your OAC environment. Someone needs to tell it which datasets or subject areas it can use, and which columns it should pay attention to. This process is called indexing and it is where most of the real configuration work happens.

Think of it like giving the Assistant a study guide before an exam; the better and more complete the study guide, the better the answers.

The screenshot below shows the indexing configuration for the storm-outage dataset used in this example. Each attribute is indexed with name and values, meaning users can ask questions that reference specific values like a storm name or region. Synonyms are also defined: Event_ID carries synonyms “Event” and “Event ID.” Storm_Name carries the synonym “Storm.” That is why a user typing “Which storm was worst?” gets the right result instead of a blank chart.

Figure 2: The dataset Search configuration screen, showing indexing enabled with name and values selected and synonyms defined for key columns.

For subject areas, the structured data models used in Oracle Utilities Analytics and similar environments the same principle applies. The screenshot below shows a live environment with multiple utility subject areas configured, along with an important Oracle warning: “The Oracle Analytics AI Assistants rely on the quality of metadata. Simply using existing analytics semantics may lead to subpar results.”

That warning is worth taking seriously. The quality of what the Assistant returns is a direct reflection of the quality of what it was given to work with.

Figure 3: The Console-Search Index screen showing multiple utility subject areas configured for the AI Assistant, with Oracle’s metadata quality warning visible.

Making the Assistant work well: start with the foundations

If you have been following this series, you already know the answer to why some AI-assistant deployments work and others disappoint. We covered it in depth in our previous blog Empowering Your AI Assistants With Data That Actually Works. Clean column names, clear metric definitions, consistent labels, proper governance over what data gets exposed all of that determines how well the Oracle Analytics AI Assistant performs.

There are also a few OAC-specific things worth calling out, because they are unique to how the AI Assistant reads and uses your data:

•       Choose the right index type for each column. OAC gives you two options: index the column name only or index the column name and its values. If users will ask questions that mention specific values like a storm name, a region, or a customer type you need to index the values too.

•       Disambiguate when you have multiple date columns. If your dataset has several date fields outage start date, restoration date, report date only indexes the one you want as the default. Otherwise, the Assistant may pull out the wrong date entirely.

•       Watch out for special characters and reserved words in column names. Symbols like @ and words like “count” or “total” can confuse the AI. Simple, clean column names always work better.

The storm-outage dataset in this example follows these principles throughout. That is why the Assistant could answer “Which storm had the most customers out?” correctly on the first try.

The bottom line

The Oracle Analytics AI Assistant is a ready-to-use tool. The end user does not need to know anything about OAC to use it. They open their published dashboard, see the Assistant panel on the right, type a question in plain English, and get an answer in seconds. No analyst needed. No report on a two-day turnaround. The data speaks directly to the person who needs it.

But like every AI assistant we have discussed in this series, it works well when the data behind it is set up thoughtfully. The broader data foundations covered in our previous blog matter enormously they are what make the answers users get actually trustworthy.

If you are already on OAC, the question is not whether you should use the AI Assistant. The question is: is your data ready for it?

Want to know if your Oracle Analytics environment is ready for the AI Assistant? HEXstream can walk through your setup and identify exactly where to focus. CLICK HERE TO CONNECT WITH US.