Extended analysis

Generate a deeper level of analysis that can reveal new insights, enabling you to make more informed decisions.

Extended analysis requires the analytics hidden insight skill from Query Generation to be active. For more information, see Query Generation skills.

To turn on extended analysis for an exploration, select it in the Ask Now Assist a question about data field. You have the same choice when you launch AI Data Explorer in a data visualization or list.

While the results of extended analysis usually enrich the responses in an exploration, the analysis takes more time and uses more system resources than a standard analysis. You have to use your judgment on when to employ it.

Extended analysis of indicators

When you view an indicator, Extended analysis automatically analyzes the following data:

Breakdown distributions

How indicator values are distributed across dimensions such as Priority, Category, and Status

Cross-dimensional correlations

Relationships between breakdowns, such as how Priority correlates with Category

Temporal trends

How the indicator has changed over time

Prior-period comparisons

How current performance compares to a previous period

In the response, Extended analysis provides a formatted summary that includes the following information:

• Overall score for the current view
• Distribution breakdowns showing individual values and their counts (for example, "Critical: 42 incidents") and the percentage of total for each value (for example, "35% of total")
• Cross-dimensional analysis (for example, "Critical + Software: 18 incidents, 43% of Priority=Critical")
• Temporal trend table showing historical scores by period (monthly, daily, and so on) and period-to-period changes

The data that Extended analysis returns also depends on the breakdowns and Group By values that are applied in the data visualization in the response.

Table 1. Data returned by grouping on the data visualization

Current view	Data returned
No breakdown or grouping (overall view of indicator)	All breakdowns for the indicator that have values, along with their distributions and temporal trends
Grouped by one breakdown (for example, grouped by Priority)	The current grouping, other available breakdowns cross-tabulated with it, and their temporal trends
Filtered to a specific element (for example, Priority = High)	The filtered view, the same element across all instances, the overall baseline for comparison, and temporal trend

Extended analysis also automatically expands the time window to provide broader context. For example, if you are viewing the last 3 months, it analyzes 6 months to show a prior-period comparison. The analysis stops expanding at 2 breakdowns to keep results focused.

Breakdowns are returned in the following priority order:

1. Currently applied breakdowns (what you are already viewing)
2. Top 50 distinct values per breakdown, ranked by score (highest values first)
3. Cross-dimensional combinations — breakdowns are paired to show correlations with the current view
4. Other available breakdowns, if the data budget allows

Note:

The default limit is top 50 values per breakdown, sorted by score in descending order. This limit is configurable.

Extended analysis of table data

For table data, Extended analysis involves aggregating the records related to a response in an exploration. It examines the same columns that you see when you view the list of records for the relevant table. It takes a Count of Choice, Reference, Glide List, and Boolean columns. Therefore, you can influence extended analysis by selecting which fields to view in the relevant tables. The relevant tables include any related tables that Query Generation dot-walks to.

The number of columns that extended analysis examines is set in the system property sn_query_gen.hidden_insights.groupby.min_fields. The default value is 5. If the number of eligible columns that are visible on the record list is lower than this value, the system searches for more fields on the table. The search stops when the total number of fields from both the list view and the table search reaches the value of the system property. If the system can’t find that many fields, it uses the fields it does find.

The search for fields on the table follows this logic:

1. The system searches for Choice fields on the table.
2. If the count of located Choice fields and eligible columns on the record list is greater than the value of sn_query_gen.hidden_insights.groupby.min_fields, not all located Choice fields are selected. Instead, the system selects the Choice fields with the highest cardinality, to bring the total count of fields up to the value of the property. The search then stops.
3. Sometimes the system can’t find enough Choice fields to bring the total count of fields up to the value of the property. In this case, the system first selects the ones it finds. Then it looks for Reference fields. The same logic that was used for selecting Choice fields applies.
4. Sometimes the system can’t find enough Reference fields to bring the total count of fields up to the value of the property. In this case, the system first selects the ones it finds. Then it looks for Boolean fields. The same logic that was used for selecting Choice and Reference fields applies.
5. Extended analysis insights are generated for however many Choice, Reference, and Boolean fields are found, up to the value of the property. If no fields are found, extended analysis isn’t performed.

Sometimes different questions or follow-up questions return the same extended analysis findings. For table data, instead of repeating those findings as insights, the system digs deeper into the underlying reasons for the repetition. As a user, you have three benefits from this approach:

• Every insight that is generated is unique.
• You don't need to posit as many follow-up questions to get the insights you need. The system drills down automatically, directing you to the interesting part of the data to focus on.
• Results are faster, because extended analysis focuses on an increasingly specific subset of the data.

Standard vs. extended analysis of table data

Consider the following request made in an exploration: "Analyze the incident creation trend over the past 12 months." Using standard analysis, you get the following insights: "Incident creation peaked in July 2025 with 2,441 incidents, while the lowest monthly count was 1,177 incidents in August 2025. The first half of the period (September 2024 to February 2025) saw monthly incident counts consistently above 1,700, but the final month dropped by more than 50% compared to the peak."

Figure 1. Insights from standard analysis

Making the same request with extended analysis, you get the same insight plus the following information:

• Incident volume dropped 49.2% from 2,441 in July 2025 to 1,177 in August 2025, indicating a major shift.

• The largest monthly increase was 28.3% in April 2025, with incidents rising by 512 to 2,321.

• Incident counts show a consistent downward trend (slope: -24.22), which suggests an ongoing reduction in reported issues.

• Monthly incident counts vary widely (range: 1,264; standard deviation: 369.24), indicating unstable operational demand.

Figure 2. Insights from extended analysis