The September 2024 release of OAC introduced the new AI Assistant. The on-line help gives a great introduction to the process of the rollout of the AI Assistant and also some caveats, most importantly “Always verify the results and consult your primary data sources before making critical decisions based on results generated by the Assistant.” Note that Oracle do not publish details about the LLM that is used and it is subject to change.
With that in mind, I generated some sample data (using a Gen AI model) of sales people who are selling different types of cars to a range of customers who live in different locations. It is super easy to load this up into OAC as a dataset and of course see the distribution of data.
If you look at latitude and longitude, I used OAC’s facility to flag these as “location” columns, which gives them a special indicator and OAC knows that these contain location information and are not just “normal” attributes.
I named the data set as SalesData2025. If we inspect it and goto the Search it is here where we set the data set up for the AI Assistant. In the Index Dataset For section we set the Data Elements to Assistant and Homepage Ask – which will index this dataset to be used by both the Ask section of the homepage and also the AI Assistant.
We can then go through the Synonyms for each column and create synonyms for the attributes so that we can ask questions of the data in a natural language.
What is really useful is that OAC will actually suggest synonyms for us.
We can now index the data set. If we press Run Now this will execute this immediately. As this is the first execution of the indexing process we will see the last run is not available. However, once this has been done, we can come back and reference this at any point to see the last time the dataset was indexed. It is worth noting that datasets can be re-indexed when the data set changes.
Once run is pressed, the indexing is initiated as an asynchronous process and runs in the background.
If we wait for it to complete, we can see that the date and time of the process completing is shown.
We can now create a workbook for the dataset and use the AI assistant functionality. This is located under the lightbulb where you also find the Auto-Insights. We can now start to ask some questions of our data using natural language and get to see results in many different ways. Once we have a result we can change/augment the results, but let’s start by being slightly ambitious and asking for a map of sales.
Note that we can flip the chart type, add/remove attributes or refine the query or even just ask something else. We can also add the results to the canvas or watchlists just as we can with Auto-Insights.
If we press add then we get a mini-menu template of additional fields or filters to add.
Let’s just ask another question of the data. This time we will ask for sales over a range of dates – in this case years but we could be more granular.
We can see that the analytic was given a filter and we have a tile that shows the total value of sales over that period. This made me think – OK, so that is the total value of all sales over those years, but I wonder how that was broken down by the sales people. So let’s just ask and see what AI gives us.
We also get an Additional Insights section, where the AI has also found some other interesting things, so if we expand that we can see the top 3 ales values for customers by the sales people who sold to them.
We want to see that as a different chart type, and we could press the Chart Type button but instead let’s ask instead.
If we are interested in exploring relationships then we can ask those types of questions, and start to refine the results. Let’s try an example with sales people and car types – do different sales people sell different types of cars?
Now we can expand on that and restrict to a couple of different sales people – this step is not shown – I just asked to restrict by Alex and Avery – and then in a subsequent step to add sales value and show this as a bar chart. These incremental changes lead us to the following where we can clearly see the sales that Alex and Avery made by different car types. So rather than “getting it right” from the start, I can build up to the analytic I need.
Of course, I can then add this to a canvas and then manually finess as I need. However, as I’m interested in Alex and Avery let’s ask the AI Assistant to show us the sales that Alex and Avery have made to customers and put that on a map – I don’t even need to spell latitude correctly for the AI to know what I mean.
In this overview we have seen how to index a dataset so that the AI Assistant can pick it up, add some synonyms and create some analytics using just natural language queries and after seeing the results finess the analytics. It is important to remember we can add these analytics to watchlists and canvases just like we can with Auto-Insights and use these as a jump off point to get some analytics started.