Page 351 -

13.2 Process Mining: TomTom for Business Processes? 333

based on these maps, for instance, a field service engineer can see the work items
closest or most urgent.

From Business Process Maps to Business Process Movies
Once events in the log can be related to activities in the process model, it is
possible to replay history on a case-by-case basis. This was used for confor-
mance checking and model extension. Now we go one step further; we do not
consider an individual case but all relevant cases at the same time. Assuming
that events have a timestamp, all events in the log can be globally ordered, i.e.,
also events belonging to different cases can be sorted. After each event, the
process is in a particular global state. One can think of this state as a photo-
graph of the process. The state can be projected onto a business process map,
a geographic map, or an organizational map. Since such a photograph is avail-
able after each event, it is also possible to create a movie by simply showing
one photograph after another. Hence, it is possible to use event logs to create
a “business process movie”. Figure 13.10 shows an example using the ProM’s
Fuzzy Miner [49, 50]. The event log and the fuzzy model are converted into
an animation. The dots visible in Fig. 13.10 are moving along the arcs and
refer to real cases. Such a business process movie provides a very convincing
means to show problems in the as-is process. Unlike simulation, the anima-
tion shows reality and people cannot dismiss the outcomes by questioning the
model. Therefore, business process movies help to expose the real problems
in an organization.

13.2.2 Arrival Time Prediction

Whereas a TomTom device is continuously showing the expected arrival time,users
of today’s information systems are often left clueless about likely outcomes of the
cases they are working on. This is surprising as many information systems gather a
lot of historic information, thus providing an excellent basis for all kinds of predic-
tions (expected completion time, likelihood of some undesirable outcome, estimated
costs, etc.). Fortunately, as shown in Sect. 9.4, event logs can be used build predic-
tive models.
The annotated transition system [110, 113] described in Sect. 9.4 can be used
to predict the remaining flow time of a running case. The transition system is con-
structed using an event log L and a state representation function l state () or obtained
by computing the state-space of a (discovered) process model. By systematically
replaying the event log, the states are annotated with historic measurements. The
mean or median of these historic measurements can be used to make predictions for
running cases in a particular state. Each time the state of a case changes, a new pre-
diction is made for the remaining flow time. Clearly, this functionality is similar to