> ## Documentation Index
> Fetch the complete documentation index at: https://docs.activeviam.com/llms.txt
> Use this file to discover all available pages before exploring further.

# What Auto-Explain can analyze

> The scope of a single Auto-Explain analysis: what one run compares, the requirements a measure and starting cell must meet to succeed, and how the drill-down search prioritizes hierarchies.

Auto-Explain analyzes data variations and identifies root causes.
This page describes what a single analysis can currently handle, so an analysis can be set up to succeed on the first attempt. It assumes familiarity with the [Auto-Explain user guide](./auto-explain).

<Info>
  ### Atoti Intelligence Essentials

  This is part of the Atoti Intelligence Essentials offer.
</Info>

## What does a single analysis cover?

Each Auto-Explain run is scoped to one focused comparison:

* Compares one measure between two specific members (for example, two dates) on a single comparison level
* Explains one measure at a time; comparing several measures requires several separate runs
* Requires the two compared cells to be identical except for the level being compared (for example, the date)
* Uses individual cells, not ranges of dates or values
* Stays within a single cube; a cause in a different cube is out of scope for the run
* Context values (settings that affect how a query is computed, such as a currency conversion) can be included in the comparison, alongside the cube's regular hierarchies
* Searches the cube's regular hierarchies only, not virtual hierarchies (lightweight hierarchies whose members are not stored in the cube)

## What does an Auto-Explain run require to succeed?

An analysis needs a complete, valid starting point:

* The measure is numeric and has a value at both compared cells
* The measure and the cube exist and are valid
* The measure works with common aggregations, such as SUM, COUNT, AVERAGE, and MULTIPLY. Percentile and median measures are not yet supported

## How does Auto-Explain search for causes?

Auto-Explain drills down automatically through the cube's regular hierarchies to find the members that are most likely to explain a variation. A few behaviors shape how that search proceeds:

* Hierarchies are ranked by their number of members at each step, and only a limited number of the smallest ones are explored. A very large hierarchy that holds the true driver can occasionally be skipped as a result. Apply filters to narrow the search if this seems to be happening.
* Whoever sets up Auto-Explain for the project can narrow or exclude specific hierarchies from the search, to focus the analysis on relevant dimensions
* Virtual hierarchies (lightweight hierarchies whose members are not stored in the cube) are never explored
* Sometimes, the search encounters a hierarchy that is correlated with the variation but is not the true cause. In this case, Auto-Explain continues to drill down. The real driver is surfaced alongside the correlated cause.

<Tip>
  If certain hierarchies are consistently irrelevant, or if it seems that the true driver lies in a large hierarchy that got skipped, the search can be narrowed or adjusted through configuration. See [Set up Auto-Explain](../developer-guide/enable-ai-tools/configure-and-start/set-up-auto-explain) for details. Please talk to the person who manages Atoti in your organisation for more details if necessary.
</Tip>

## Related reading

* [Auto-Explain user guide](./auto-explain) to learn how to run an analysis and interpret its results
* [Set up Auto-Explain](../developer-guide/enable-ai-tools/configure-and-start/set-up-auto-explain) to add the feature to a project and configure the hierarchies it searches