Setting Propensity Thresholds
Propensity Thresholds are a means of setting a minimum value for the adaptive model propensity below which actions or treatments will not be considered for arbitration.
Such thresholds can be set in a hierarchical fashion with more granular values taking precedence, the hierarchy is shown below.
- (Action) Name
So, for example, if a threshold of 0.6 is set for the Outbound direction, and 0.7 is set for Outbound and the Email channel, 0.7 would be used for Email, and 0.6 for any other outbound channel that had not been explicitly set.
Propensity Threshold concepts
The thresholds are maintained in the Model Propensity Thresholds DDR containing the properties described below.
|pyLabel||A user-friendly name for a threshold.|
|pyDirection||Propensity Thresholds are defined separately for Outbound or Inbound directions. If this is blank, then the same thresholds are applied to both Outbound and Inbound.|
|pyChannel||The channel to which the Propensity Threshold is to be applied. If this is
blank, then the threshold is applied to all channels, however, if any channel is
specified, then only channel specific thresholds will be applied, and all thresholds
with a blank channel will be ignored.
If channel-specific thresholds are used, then they must be set separately for each channel. Note that this only applies within each direction, so channel could be blank for Inbound, and set individually for Outbound if desired.
|pyIssue||The issue to which the threshold is to be applied. If this is blank, then the threshold is applied across all issues.|
|pyGroup||This is only referenced if pyIssue is not blank and is the
group to which the threshold is to be applied. If this is blank, then the threshold
is applied across all groups within the issue.
If not blank, this will supersede any threshold set at the issue level.
|pyName||This is only referenced if pyGroup is not blank and is the
Action to which the threshold is to be applied. If this is blank, then the threshold
is applied across all actions within the group.
If not blank, this will supersede any threshold set at the issue or group levels.
|pyTreatment||This is only referenced if pyName is not blank and is the
treatment to which the threshold is to be applied. If this is blank, then the
threshold is applied across all treatments for the action.
If not blank, this will supersede any threshold set at the issue, group or action levels.
|pyPropensity||The Propensity Threshold which sets a lower limit for the Final Propensity value of an action or treatment. Only actions and treatments with a Final Propensity greater than this threshold will be considered for arbitration within the NBA Strategy framework.|
- Propensity Threshold Use Cases
The following Propensity Threshold use cases are examples of how to set a minimum threshold for various scenarios.