AI Root Causes

The AI Root Causes analysis is an improved version of the Root Causes analysis that leverages the Snowflake's Top Insights feature. This allows for a streamlined and powerful analysis of the factors influencing your processes. The AI Root Causes analysis automatically generates insights based on your data, running natively in the Snowflake AI Data Cloud for scalable analytics.
The AI Root Causes analysis identifies the most influential factors that contribute to specific outcomes in your process. For example, you can analyze which case attributes are most likely to lead to long case durations or high costs.

There are presets that simplify this analysis by providing pre-configured viewpoints. Instead of manually selecting and configuring attributes for analysis, you can use presets to quickly investigate common areas of interest.

How AI Root Cause Analysis Works

AI Root Cause Analysis begins by constructing features based on user-configured settings. These features are then analyzed by the Snowflake Top Insights tool to identify the most impactful combinations of correlating features. Leveraging a decision tree model, the analysis can uncover complex correlations among features. For instance, it can detect the following types of combinations:

• Combination of Different Features: Such as when attribute A value is B and attribute C value is D.
• Combination of Feature Values: Such as when attribute value is either A or B.
• Numeric and Date Value Ranges: This includes identifying conditions where an attribute is greater than or less than a specified value.
• Inverse of Feature Values: For example, cases where the attribute value is not A.

The analysis can evaluate various features as potential root causes, including:

• Values of case attributes
• Values of event attributes
• Specific event types occurring within a case
• Whether the case has a specific flow (transition between two consecutive events)
• Starting and ending events
• Variation within cases
• Duration of cases

This structured approach enables the identification of underlying issues with precision and depth.

AI Root Causes (case attributes) Preset

On top of the AI Root Causes (case attributes) table there are the on-screen settings for choosing the case attributes to analyze.
The following columns are shown when using the AI Root Causes (case attributes) preset:

• Root cause: Shows the combination of the different attribute values that form the root cause.
• Total case: Number of cases having the attribute value combination that the row represents.
• Selected cases: Shows the number of cases in the selected set of cases (i.e. in the selected root cause criteria).
• Selected cases %: Shows the percentage of the cases with the attribute value combination in the selected set of cases.
• Contribution: The number of cases having the attribute value combination which contribute to the deviation from the average percentage.
• Contribution %: Percent of the selected cases having the attribute value combination that contributes to the deviation from the average percentage.

AI Root Causes (all features) Preset

The AI Root Causes (all features) is similar to the AI Root Causes (case attributes), but in addition to the case attributes, all available attributes and other settings are enabled. There is also an additional on-screen settings selector for choosing the event attributes to analyze.

AI Root Causes (weighted) Preset

The AI Root Causes (weighted) is similar to the AI Root Causes (case attributes), but the Metric setting used in the analysis is the alphabetically first numeric case attribute and the numeric case attribute can be selected with the additional on-screen setting. The following columns are shown when using the AI Root Causes (case attributes) preset:

• Root cause: Shows the combination of the different attribute values that form the root cause.
• Total metric: Sum of the test and control metric in this contributor.
• Selected metric: Sum of the test metric cases in this contributor.
• Selected metric %: The percentage the test metric from the total metric in this contributor.
• Contribution metric: How much this contributor contributes to the selected phenomenon in the metric units.
• Contribution %: How much this contributor contributes to the selected phenomenon in percentages.

AI Root Causes Settings

The AI Root Causes Analyses have the following chart settings on the Analyze tab:

• Metric: Expression selection to define the metric for the analysis. Value for the metric is calculated for each analyzed case. Only expressions that may have numeric data type are shown in the list. When Static value 1 is selected, the metric represents the case count.
• Analyze case attributes: Multi-selection list of all case attributes in the model, excluding the attribute mapped to the case ID. Each case attribute is a textual feature column in the analysis.
• Analyze event attributes: Multi-selection list of all event attributes in the model, excluding the attribute mapped to the case ID. Each event attribute value is a boolean feature column in the analysis indicating whether the event attribute value appears in the case (one or more times) (note also the Analyze top event attribute values setting).
• Analyze top event attribute values: Defines how many most common event attribute values for each event attribute are analyzed. When defining zero, event attributes are not analyzed.
• Analyze top flows: Defines how many most common flows are analyzed, ignoring the start and end flows. Each flow is a boolean feature column in the analysis indicating whether the flow appears in the case.
• Analyze first and last event: Checkbox whether the first and last event type are analyzed (i.e., the start and end flow). The starting event type is a textual feature column in the analysis, and the end event type is another textual feature column.
• Analyze variation: Checkbox whether the case variation is analyzed. The variation path is a textual feature column in the analysis.
• Analyze case duration: Checkbox whether to analyze the case duration. The duration in days is a numeric feature column in the analysis.
• Slice into Dimensions / One row per object: See chart settings.
• Sorting: See chart settings.
• Maximum Rows: See chart settings.
• Cases Sample Size: See chart settings.
• Result Filtering Expression: See chart settings.
• Model: See chart settings.
• Event Type Mapping: See chart settings.