With one dimension, the entities in that dimension form a group. For example, a group of categories.
When you add in another dimension, you could think of it as a group of groups. For example, each category in the group has its own group of sub-categories. You might visualize this like so:
The narrative that explains a more complex, two-dimensional dataset like this can be broken down into simpler components.
The first section aggregates the secondary dimension up so that it's left with a simpler, single dimension, single measure narrative. In this example, it would write about Sales Revenue by Category first. The narrative content for this section follows the same structure and logic as the discrete single dimension, single measure narrative.
The subsequent sections (drill-down sections) each focus on an individual primary dimension entities, e.g. Furniture. Since each primary dimension entity is actually a group of secondary dimension entities (e.g. subcategories), the drill-down section can be viewed as another discrete single dimension, single measure narrative.
The narrative will always choose the top three primary dimension entities (e.g. categories) by totaling the metric for all its subcategories. In this way, it's always drilling down into the most important information. Note, if the measure characterization is set such that larger values are considered bad, the drill-downs will be about the three primary entities with the lowest totals.
In some cases, the two dimensions exist on the same hierarchy (e.g. Sub-categories roll-up to Categories). But it's also possible, and common even, to analyze data by crossing dimensions from independent hierarchies (e.g. Category and Region).
In the latter scenario, one could reasonably care about Categories grouped into Regions, and also be interested in the opposite: Regions broken down by Category. For this reason, the narrative will produce a Support Story, which basically reverses the primary and secondary dimensions and writes another two dimension, single measure story from this other perspective. This alternative view can be very insightful.
The tool knows whether or not to produce a support story by detected how many secondary dimension entities "overlap" across all the primary dimension entities. For example, a Cub-category would only ever exist in a single Category, so there's no "overlap". But a Category could exist in every Region, and vice versa, so there is "overlap".