Improve Results from Oracle Analytics AI Assistant

You can improve results from your Oracle Analytics AI Assistant with careful indexing and organization of your data.

Your Oracle Analytics AI Assistant is powered by a large language model (LLM). LLMs are better at understanding text than numbers, so change null values in your datasets to meaningful text like "Unknown" or "Not available". You can also optimize your results by binning numerical values, like binning customers into groups based on total sales, or binning order amounts into ranges of small, medium, and large. See Create a Bin Column When You Prepare Data.

Enriching and transforming your data with suggestions from Oracle Analytics can provide greater context to the Assistant. For example, transforming a dataset to include data part extractions where you extract the day of a week from a numeric date means the Assistant can more accurately respond to a prompt like "show sales for all Fridays in 2024". See Enrich and Transform Your Data.

You can reduce ambiguity in your indexed data by having clear column names that can be understood by anyone reading them, avoiding duplicate names, and leveraging synonyms. For example, datasets often have multiple date fields (order date, Ship date, invoice date). This can cause confusion when someone asks a question like “show sales by month”. You can avoid this by creating a synonym called "Date" on the column you want to assign as the default date. See Specify Synonyms for Dataset Columns.

When indexing your data, carefully select which columns you want indexed. Consider which columns users are most likely to ask the Assistant to provide insights and visualizations for, like metrics, regions, and dates. Avoid indexing high-cardinality columns, like customer IDs, SKUs, or addresses as these can add complexity without enhancing Assistant insights. Start by selecting Use Recommended Index Settings when indexing your dataset and refine individual column choices to optimize your search index from there. See About Recommended Index Settings.