Insight Discovery

Insight discovery identifies the splits within each measure with the greatest variability. Users can leverage this screen to kickstart their efforts to identify unexpected variability in the underlying data.

Calculation

Insight Discovery is run on each qualifying field for measures that are set to run it. The qualifying fields are every denominator analytic field, minus the risk stratification fields and the Insight Discovery blacklist fields.

Ursa Studio calculates the variability of the split for each qualifying field, and surfaces the fields with the highest variability. You can think of variability simply as a measure of the unevenness of the bars in a bar chart representation of the split.

Fields for which users expect to naturally have significant variability can be blacklisted to keep them from dominating the analysis, and might already be added as risk stratification fields if Opportunity Discovery is already set up for the measure.

Insight Discovery performs its analysis both on one-way and two-way splits. The two-way splits are the cross product of each qualifying field. These two-way splits are considered children of their parent one-way splits, and are only included for display if their variability is significantly higher than the variability of each of their parents. This logic allows Insight Discovery to surface drivers of variability that might get missed on a first-pass analysis.

Setup

Users can turn on Insight Discovery for a measure and fine-tune some of its parameters from the Advanced Analytics panel in Measure Workshop.

The Insight Discovery blacklist enumerates the fields that are ignored in the Insight Discovery calcluation. Typically, these are fields for which their impact on the measure value is already known and appreciated.

The child penalty parameter dictates how much higher a two-way split's variability needs to be above and beyond its parents in order to merit its inclusion in the final presentation. The child max parameter sets a ceiling on the number of two-way splits that are displayed underneath each parent. Taken together these two parameters can be used to alter the mix in the final presentation of one-way and two-way charts.

The significance threshold parameter establishes a lower bound for any chart's inclusion in the presentation. Users can alter this value on a per-measure basis if they find that they're getting too many or too few results surfaced.

Visualization

Insight Discovery is rendered using the same UI as boards. Each parent split is given their own title header, and all the qualifying children splits are shown underneath, sorted by variability. The variability is noted in a comment alongside the bar chart.

Typically, a split will be left off the Insight Discovery presentation if it does not have a high enough variability. The exception is the case in which a parent has low variability but has some children with very high variability. In this circumstance, the parent one-way split will be displayed as a way of contextualizing the children beneath it, but its low variablity will be noted in its comment.

The sections are sorted by the variability of the highest-variability chart within the section, which will typically be the two-way split displayed directly underneath its parent one-way split.