Key Drivers: discover tested impact

Methods and settings

Discover, then inspect

Select a meaningful target, such as forecast EBITDA. Discovery starts automatically for the current eligible sources; use Refresh to repeat it. Read the range and unit behind each entry before interpreting its rank. The Layer can organize sources into a hierarchy, including Business Lines, and show supported individual sources. Group tests move the group’s assumptions together; they are not the sum of independently ranked bars.

What the ranking measures

Each supported test compares low and high input outcomes with the current Base result. Ranking uses the largest absolute tested deviation from Base. Positive, negative, mixed and flat describe the tested direction. Inputs without a material detected effect are not evidence that the business relationship is universally irrelevant.

The ranges are analytical assumptions. A larger range can produce a higher rank even for a less sensitive relationship. Use comparable, defensible ranges before making a business decision.

Interpret a Business Line result

Suppose a synthetic revenue line has a supported price assumption. With volume held constant, the low-price and high-price cases produce different revenue and EBITDA. Downstream costs move only through relationships actually modeled. A high driver rank means a large tested impact on that target, not that the line represents the largest share of total revenue. Inspect composition separately in Inspector charts.

Choose the next test

Use Sensitivity to examine a selected relationship across several values or two axes. Use Tornado to compare an explicit shortlist of ranges. Use Stochastic only after assigning defensible uncertainty distributions. Key Drivers is not a statistical attribution model, a causal discovery engine or an accounting contribution chart.

Public models do not store this analysis session

Key Drivers rankings and temporary probe results are not persisted in public model snapshots. Publication preserves operator inputs and deterministic operator outputs. After Fork, run this analysis again locally against the restored editable inputs. A reviewed change explicitly applied to an underlying operator input can be published as that input; the analysis session that produced it is still excluded.