Page 274 -

256 9 Operational Support

Fig. 9.11 A model based on historic data is used to provide recommendations for running cases.
Recommendations are not enforced and may have quality attributes attached, e.g., in 85% of similar
cases, x is the activity that minimizes flow time

9.5 Recommend

The third operational support activity we consider in this chapter is recommenda-
tion.AsFig. 9.11 shows, the setting is similar to prediction, i.e., a partial trace is sent
to the operational support system followed by a response. However, the response is
not a prediction but a recommendation about what to do next. To provide such a
recommendation, a model is learned from “post mortem” event data. Moreover, the
operational support system should know what the decision space is, i.e., what are the
possible actions from which to choose one. Based on the recommendation model,
these actions are ordered. For example, in Fig. 9.11 the operational support system
recommends to do action x with 85% certainty. The other two possible actions have
a “lower” recommendation: y is recommended with 12% certainty and z is recom-
mended with 3% certainty. In most cases it is impossible to give a recommendation
that is guaranteed to be optimal; the best choice for the next step may depend on the
occurrence of unknown external events in the future. For example, in Fig. 9.11 there
may be cases for which z turns out to be the best choice.
A recommendation is always given with respect to a specific goal. Examples of
goals are:
• Minimize the remaining flow time
• Minimize the total costs
• Maximize the fraction of cases handled within 4 weeks
• Maximize the fraction of cases that is accepted
• Minimize resource usage

These goals can also be aggregated and combined, e.g., to balance between
cost reduction and flow time reduction. To operationalize such a goal, a perfor-
mance indicator needs to be defined, e.g., remaining flow time or total costs. This
performance indicator corresponds to the response variable in supervised learn-
ing.
A recommendation makes statements about a set of possible actions, i.e., the
decision space. The decision space may be a set of activities, e.g., {f,g,h}.This
means that in the current state activities f , g, and h are possible candidates and the
question to be answered by the operational support system is “Which candidate is