A Goal-Based Approach for Learning in Business Processes - Intentional Perspectives on Information Systems Engineering - page 246

One situation would be a context group for which a path that uses new
material and quality inspection level 3 would yield zero cases of unaccept-
able quality level (no customer rejects, cancellations, or 100% inspections
performed). However, when a certain worker operates the machine, an aver-
age 5% defects (which is still acceptable by the customer) is obtained at the
sample inspection, while other workers normally have 2% defects. As a result,
the specific worker can be trained, or not be assigned to orders of that context
group.
As a second example, the current process model includes a decision point
where a 100% inspection can be performed (or skipped) after a problem has been
identified and solved during production. It could be identified that avoiding 100%
inspection at this point significantly increases the probability that the customer
would reject the delivered goods. Hence, the process model would be changed
so performing such inspection becomes mandatory.
A third example might identify that for a certain group of machines, when
the product is for the medical market, the frequency of problems identified dur-
ing production increases about four months after maintenance activities, while
for other machines and other context groups it happens only after six months.
This could indicate the need to perform maintenance more frequently for these
machines, or to avoid using them for medical products if four months have passed
since their maintenance.
2. Addressing specific questions that might be asked. Management may have
“hunches” about possible causes of poor performance. These can be specifically
addressed by analyzing path and performance for all the context groups. For
example, in the bottle manufacturing process there are cases of very urgent orders
for which the machine set up is done in an accelerated manner. Management
wishes to check whether this accelerated set up results in decreased product
quality in general or for specific context groups. Specific analysis will try to
correlate the time spent for the set up activity with the outcome for different
context groups.
3. Identifying successful deviations from the existing process model as a basis for
managerial decision making. The process instances in the repository may include
instances where specific ad hoc decisions were made to deviate from the “nor-
mal” process at runtime. For example, there might be cases where special quality
inspection instructions were given, not compliant with the existing three levels.
Since this has not been repeated enough times to get statistical significance for
its results, we can only indicate that such deviations were made and the extent
of their success in achieving the process goals. Such indications may be used
by management, which may decide to repeat this course of action so more data
becomes available for future learning cycles.
It should be stressed that we do not suggest any change to be made automatically
to the process model. All the improvement recommendations should be reviewed by
humans (managers, domain experts), and the performance of the improved process
should be monitored.