Process-driven Management Information Systems - ing and work distribution tasks [6].

Recently, workflow management sys-

Combining Data Warehouses and Workflow Technology tems have spread beyond the administrative environment and can also be
found as embedded software components, that enhance existing application
packages (e. g. ERP systems) as well as infrastructure components (such as
Michael zur Muehlen
application servers) with process management functionality [31].
Department of Information Systems
University of Muenster Although workflow management systems can increase the process effi-
ismizu@wi.uni-muenster.de ciency of an enterprise by as much as 150% [17], they do not necessarily
lead to a more flexible organization. Since the introduction and deployment
Abstract of a workflow-based information system architecture is very often a com-
The use of workflow technology promises efficiency gains through the auto- plex and time-consuming endeavor, it can be observed, that once this kind of
mation of manual routing, coordination and work distribution tasks. During architecture has been successfully deployed, many companies resist the urge
the execution of workflows, state-changes of the workflow engine are to apply changes to the new system (an effect that can be observed at compa-
recorded in a log file or database, the so-called audit trail. Combined with nies introducing ERP packages as well).
business object data, the audit trail provides exact and timely information However, the complexity of workflow projects is only one cause for the
about the operational behavior of an enterprise. In this paper we discuss the reluctant attitude observed towards change management. An even more
design on data warehouse applications that take advantage of workflow severe cause is the non-transparency of those relationships, that describe the
technology as an information source. We outline evaluation opportunities effects of workflow changes on the organizational, technical, or process
generated by the use of audit trail data and point out potential pitfalls with level. This missing transparency can be attributed to the lack of an integrated
regard to data consistency and integrity. infrastructure for the gathering and presentation of performance indicators,
that describe the behavior of a workflow-enabled organization and give
1 Introduction advice on which parameters to adjust in order to increase the organizational
The use of workflow management systems improves the efficiency of busi- efficiency.
ness processes through the automated coordination of the data and resources During the run-time phase of a workflow application, the workflow
needed for the execution of the single activities. Workflow management sys- engine generates information about the different state-changes on the pro-
tems rely on a formal representation of the process logic that is designed as a cess and activity level that is recorded in the so-called audit trail, which is
workflow model during the development phase of a workflow application either file-based or a database structure (cf e. g. [19]). Originally designed as
(also called build-time). During the execution phase of the workflow appli- a technical protocol for debugging purposes, the audit trail provides informa-
cation (also called run-time), the workflow engine derives workflow tion about the execution of business processes at the operational level.
instances from the generic workflow model and notifies workflow partici-
pants about pending activities through their work-lists (see e. g. [14]). In theory, the use of audit trail information could enhance traditional
enterprise controlling and management information systems by presenting
Workflow management systems have found widespread acceptance since key indicators about process performance as well as drill-down capabilities
the advent of this technology in the late 1980s. The Association for Informa- from single process instances to the business data connected with these
tion and Image Management (AIIM) estimates the worldwide revenue for instances. In practice, the reporting components of many workflow manage-
workflow technologies to grow from $4.3bn in 2000 to $8.3bn in 2003 at a ment systems offer only limited evaluation functionality. Especially the
compound annual growth rate of 31% [6]. Especially in conjunction with combination of workflow audit trail data with data warehousing technology
document management technology, workflow systems are perceived as the is not supported by most workflow vendors, but left to extensions by the
enablers of office productivity gains through the elimination of manual rout- user.
Published in: Gavish, B. (Ed.): Proceedings of the Fourth International Conference on Electronic Commerce Research (ICECR-4), Dallas, TX, November 8-11, 2001., pp. 550-566
 This paper aims at providing a framework for the design of data ware- the fact that there is no isolated cause-effect relationship between workflow
house applications that are specifically targeted at the integration of work- design and enactment. Changes that affect the performance of a workflow
flow audit trail data. In the following section we develop a taxonomy of can be applied not only to the workflow model itself, but also to the invoked
different analysis purposes for workflow audit trail data. We distinguish application systems as well as the organizational surroundings of the work-
between workflow monitoring and controlling, and show, how both concepts flow application. If, for example, the organizational signature power of cer-
relate to the workflow life cycle. In section 3 we focus on the design issues tain workflow participants is increased, these participants could
for a process information system based on workflow audit trail data. Follow- autonomously approve an increased number of cases, lowering the number
ing a phases model that spans data extraction, transformation, evaluation and of managerial approvals necessary and thus reducing the workload of certain
presentation, we point out benefits and potential problems associated with (potentially limited) resources. Even though this measure has no direct effect
the use of audit trail data for data warehouse applications. In section 4 we on the workflow model or the workflow management system, its effects are
show the benefits of workflow-driven process information systems using a a noticeable change of the process performance (e. g. lower average through-
case study from an insurance company. A discussion of related work is put time).
given in section 5. The final section of this paper provides a brief summary
In our experience, the introduction of a workflow management system is
and outlook.
a sequential process, very similar to the development of a complex applica-
tion system. Workflow monitoring and controlling are cyclic activities that
2 A Taxonomy for Workflow Monitoring and Controlling follow the introduction of the system and run in parallel to system adminis-
tration and maintenance tasks. Figure 1 shows a procedure model for work-
2.1 Monitoring and Controlling in the Workflow Life Cycle
flow application development and deployment (cf. [33]). The organizational
Companies face the need to quickly adapt to changing market conditions and
loop on the right reflects the cyclic auditing of the workflow application. Its
customer wishes. Having a transparent overview about ongoing and histori-
feedback is applied in the technical loop through incremental changes to the
cal business processes gives companies the flexibility to adjust the treatment
workflow application. This way, a continuous improvement process can be
of individual cases as well as the opportunity to make structural changes to
maintained, without facing the risks associated with a fundamental redesign
business processes. Workflow management systems separate application
of existing processes and applications.
logic from business process logic, enabling users to modify the business pro-
cesses based on intelligence gathered from the use of a workflow applica-
tion.
The life cycle of a workflow application has been described by several
authors as a closed loop (see e. g. [7]), starting with the definition of the
business process to be implemented (1), followed by the transformation of
the process model into a workflow model (2). The enactment of the work-
flow model (run-time) makes up the third phase of the life-cycle (3), whereas
the ex-post analysis of executed workflows in the sense of process control-
ling (4) generates information that is fed back into the process design phase
Other workflow cycles, such as the one described by HEILMANN (see. e. g.
[9]), contain different phases, but they all depict the development, deploy-
ment and analysis of a workflow application as a closed loop.
Figure 1: Procedure Model for Workflow Development Projects
In practice, the closed loop assumption does not reflect the development and
deployment of workflow applications as it actually happens. This is due to
2.2 A Taxonomy for Workflow Analysis Evaluation Focus
The analysis of workflow data can be based on either short-term or long- Technical Business- oriented
term observation of the audit trail. This differentiation is typically used to Data Scope Current Data (live) Single Instance Single Instance
distinguish between workflow monitoring (discussed in section 2.3), which • exception handling • workflow monitoring
describes the analysis of active workflows, and workflow controlling (dis- (e. g. manual termina- (e. g. tracking of the
tion of workflow individual process
cussed in section 2.4), which deals with the ex-post analysis of (potentially instances, manual re- state)
finished) workflow instances. While short- and long-term observation pro- assignment of orphan
activities)
vide a suitable differentiation for the data to be analyzed, the purpose of the
Multiple Instances Multiple Instances
analysis can serve as another point of differentiation. With regard to this,
• system monitoring • workforce manage-
technical and business-oriented analysis goals can be distinguished. (e. g. number of data- ment (e. g. scheduling
base connections, of resources based on
From the end user point of view, another separation can be made between response time) predicted workload)
the analysis of data for the purposes of an individual user as opposed to the • license monitoring • workload manage-
(e. g. utilized invoked ment (e. g. assign-
purposes of an organization (for reasons of clarity, this view has been applications) ment of work items
ommitted from the table). These different perspectives determine both the based on individual
capacity restrictions)
relevant objects for the analysis (activities, resources, data, applications), as
Historic Data (ex- post) Single Instance Single Instance
well as their cumulation (e. g. the evaluation of single workflow instances as
• workflow debugging • audit purposes (e. g.
opposed to the evaluation of an aggregate of multiple workflow instances). • workflow recovery (in proof of fulfilment in
case of a system fail- case of customer
Table 1 shows a taxonomy for the analysis of workflow protocol data. ure) complaint)
For reasons of clarity, the table contains different evaluation opportunities Multiple Instances Multiple Instances
for single workflow instances in contrast to aggregate workflow instances. • license management • workflow controlling
(e. g. utilized invoked (e. g. activity based
The analysis of workflow audit trail data for technical purposes is per- applications) costing, resource uti-
formed mainly by system administrators and workflow designers, that are lization etc.)
• workflow planning
debugging workflow models that are not executed correctly (since some (e. g. development of
workflow management systems use relaxed formalisms for their modeling new processes and
procedures)
languages in order to accommodate less formal business process modeling
techniques, the resulting processes may contain potential deadlocks or simi- Table 1: Taxonomy for workflow audit data evaluation
lar problems). Another purpose could be the monitoring of system loads and dabase-related transaction processing and its impact on workflow technol-
response times in order to determine the proper scale of the underlying hard- ogy see e. g. [2]).
and software-components (mostly in terms of databases). In order to identify
workflow instances that are “stuck” on the worklist of an absent user (and The business-oriented analysis of workflow audit trail data can be
the respective activity does not have a “timeout” attribute), the system divided into the monitoring of current workflow instances, which is per-
administrator may query the audit trail for activities that are assigned to formed by workflow users and process managers (or process administrators),
absent users. But not only human users take advantage of the logged work- whereas the controlling of past workflow instances is mainly conducted by
flow history. In case of a system failure many workflow systems use the enterprise controllers. Since these two analysis purposes promise significant
audit trail file to recover the system state to the last committed transaction, business value, we will discuss them in detail in the following sections.
much like a database system uses a transaction log file (for a discussion of
2.3 Process Monitoring for a forecasting of open processes at a specific point in time, the number of
Process monitoring deals with the analysis and overview of process active process instances as well as the current activities of these instances
instances at run-time. Using monitoring information, workflow administra- allow the short-term prediction of staff requirements. Combined with the
tors and process managers can adjust the behavior of current workflow ability of workflow management systems to prioritize work items according
instances and react to problems that arise during process enactment. Further- to the case attributes and the age of the case (i. e. the idle time of a pending
more, process monitoring is used to improve the responsiveness of an orga- case), process monitoring helps companies maintain a consistent level of
nizations to customer inquiries. When the current state of a process instance cycle times even during seasons with high workloads.
can be determined easily, questions such as “Who is handling the customer The importance of workload transparency is illustrated by an example of
order 4711?” can be answered in an efficient manner. For the individual a German insurance company. Due to a proposed (and later cancelled)
workflow participant, monitoring provides the ability to identify those col- change of the tax legislature, life insurances were subject to additional taxa-
leagues that worked on a particular case earlier, in case of open issues that tion, if the contract was signed on or after January 1st, 2000. This announce-
need to be resolved. Figure 2 shows the monitoring of an online order while ment led to a fourfold increase in life insurance applications during the
it is being processed by the vendor. second half of 1999. The staff at the life insurance department worked over-
time to handle the unusual amount of applications, neglecting all other cases
that were not new applications. As a result, the structure and age of the
remaining cases was unknown and customers complained about the long
time it took the insurance to get back to them regarding their inquiries. This
situation could have been avoided easily, if a workflow management system
had kept track of all the cases and prioritized those, that were older than a
certain threshold.
Under certain conditions it is desirable not to expose the detailed process
structure to the party monitoring a certain workflow. An example is the pre-
sentation of workflow data to workflow participants outside of the company
the workflow is executed in, e. g. customers or suppliers. Figure 2 shows the
web display of an ordering process at an e-commerce web site. Even though
the internal processes are much more complex, only four steps are displayed
to the user. This business state differs from the actual process state in the
way that it is a simplified state model of the underlying process state model
(NAEF et al call these state models shadow processes [20]). Figure 3 shows
an example for the abstraction level between the actual process state and a
business state. The four activities in section B and the two activities in sec-
tion C are combined into the single status B’ and C’, respectively, whereas
the activities A, D, and E appear at the same level of granularity in the busi-
ness state model.
Figure 2: Monitoring of an E-Commerce Order In the (simplified example, the business state model can only contain the
same number or fewer states than the process state model, since it is derived
Process monitoring beyond single process instances can be used to pre- from the workflow states exclusively. If context data such as the values of
dict staffing requirements. If the average processing times of activities allow
 Process Business
State State

A A'

B B'

C C'

D E D' E' Figure 4: Technical Monitoring Facility

those processes and activities that have been accepted by a user but that have
Figure 3: Process State and Business State not been completely processed).
certain process relevant variables is taken into account, the coarse states of 2.4 Process Controlling
the business state model may be refined into sub-states.
Process controlling deals with the ex-post analysis of workflow audit trail
Besides the organizational process monitoring, workflow management data. Here the single workflow instances are aggregated according to differ-
systems typically provide facilities for technical monitoring, which deals ent evaluation dimensions schemes. Workflow controlling is useful for the
with parameters such as response times, system load and the like. With detection of long-term developments in workflow enactment and the review
regard to technical monitoring workflow management systems do not differ of already existing workflow implementations. In order to identify devia-
from complex application systems that are managed through commercial tions in process execution, the audit trail data is often compared to target
packages such as Tivoli [25] or Candle [4]. Figure 4 shows a screenshot of data derived from corresponding business process models. The goal of work-
the technical monitoring facility of a commercial workflow management flow-based controlling is the improvement of future process enactment, thus
system. Besides the current numbers of active users, processes and activities its effects are more long-lasting than the results of workflow monitoring.
the system also displays the number of pending processes and activities (i. e.
 Whereas the target audience for workflow monitoring data is administra- specifications. Nevertheless, these systems rarely store this kind of data in
tive IT personnel on the one hand (for technical information) and workflow the audit trail file.
participants on the other hand (for organizational information), workflow
From a controlling perspective, workflow audit trail data is another
controlling data is mainly used for enterprise controlling purposes. Analyzed
source of information, just like financial statements from an accounting sys-
autonomously, audit trail data provides information about the temporal
tem or log files from a transaction processing systems. In order to enhance
aspects of process execution as well as information about resource utiliza-
the business value of audit trail data, it needs to be combined with applica-
tion on the process and activity level. However, information about the busi-
tion data. Data warehouses are a common repository for this kind of data,
ness context of a particular process in most cases cannot be answered by
and elaborate on-line analytic processing tools exist that support the control-
looking at audit trail data alone. This is due to the fact that most workflow
ling recipients during the evaluation of information stored in a data ware-
management systems do not store data that is processed in the applications
house. Therefore, the integration of workflow audit trail data into the
invoked during workflow enactment. The WORKFLOW MANAGEMENT COA-
existing data warehouses provides the opportunity of enhancing existing
LITION has published a definition of three classes of data associated with
controlling infrastructures with the ability to analyze the business process
workflow systems [30].
perspective of an enterprise. This task is non-trivial and we discuss the
• Application data is data beyond the control sphere of the workflow implications in section 3.
management system. It is managed and stored by the applications
invoked during the enactment of workflows, e. g. a letter to a customer 3 Data Warehouse Design for Process Information Sys-
that is managed by the word processing system. tems
• Workflow-relevant data is managed by applications and has an impact
3.1 Transforming Audit Trail Data Into A Data Warehouse Schema
on the control flow of the current process. Typically this type of data is
queried at decision nodes during the process, when the workflow man- During the extraction, transformation and loading phase of data warehouse
agement system has to decide which of several alternative paths to fol- development, source data is converted into a format that fits the data ware-
low. Workflow-relevant data can also be used to increase the flexibility house schema. Depending on the format of the existing audit trail logs, that
of staff assignment rules (e. g. “IF claim.value()<50,000 THEN per- could be database records or flat file structures, an appropriate import mech-
former.role() = accountant ELSE performer.role() = manager”). This type anism needs to be deployed. Using a transformation algorithm, the propri-
of data is read (but not updated) by the workflow management system, etary schema of the audit trail is converted into the data base schema of the
but only few systems store this information in their audit trail records. data warehouse. Since the audit trail data is machine-generated, the removal
of missing or incorrect values is of little importance during this step.
• Workflow-internal data is managed by the workflow management sys-
tem itself and contains information about the current process instance, The audit trail schema used by the workflow management system has to
e. g. the ID of the process starter or the name of the performer of the last be converted to a schema that is suited for the analytical purposes of the data
activity. This information is used to realize run-time specific semantics warehouse. If audit trails from different workflow management systems
in the process flow, such as the assignment of an activity to the manager have to be integrated, a common schema for all workflow audit trail formats
of the process starter. This is the kind of information found in most audit involved has to be developed. The WORKFLOW MANAGEMENT COALITION
trail formats. has published a common audit trail format, that could be used as a common
denominator by different workflow vendors (WfMC Interface 5 specifica-
Many commercial workflow management systems adhere to the WfMC tion [29]). In practice, however, most workflow vendors have defined a pro-
separation of application data, workflow-relevant data and workflow-inter- prietary format for their audit data. Out of 20 vendors claiming conformance
nal data. Some systems allow the workflow designer to specify complex data with WfMC standards, 9 have expressed interest in supporting the Interface
structures and provide facilities for explicit data flow and transformation 5 standard, but only 2 have actually implemented the standard. It should be
noted, that currently no official conformance testing facility for the WfMC While the events (such as state changes) during workflow enactment are
standard exists, so even the two implementation have not been evaluated by the point of reference for the recording of the audit trail information, most
an independent organization yet. users of the audit trail will analyze the data at a higher level of integration.
Evaluations at the process or activity level require the integration of several
Figure 5 shows a section of the WfMC Interface 5 data format. Each
audit records. If a workflow system supports multiple state changes between
audit trail entry consists of a CWAD (Common Workflow Audit Data) prefix
the suspend and resume of an activity or a process instance, the number of
and suffix. While the prefix is defined by the WfMC, the suffix can be used
audit trail entries for an activity is not predefined. Integrating these entries
for vendor-specific extensions. Depending on the type of event that occurred
can resulting in complex join operations on large data volumes. For the
in the system, a different set of attributes is recorded in the body of the log
design of a process-oriented data warehouse this implies, that during the
entry. The example shown in the figure depicts an audit trail entry for the
extraction, transformation and loading phase, summarized datasets have to
creation or start of a process or subprocess. Accordingly, e. g. the termina-
be generated for activities and processes, to ease subsequent evaluations.
tion of an activity would result in a different set of data recorded in the body
of the log entry. For the transformation of audit trail records into a common 3.2 Integration of Business Object Data
data warehouse schema, this results in the filtering of excess data and the
As pointed out in section 2.4, most audit trail formats do not contain a
replacement of missing values, that are not recorded for certain event types.
link to the application data that has been processed in the workflow instance.
However, meaningful evaluations in a business context almost always
require the workflow audit data to be linked to some business object infor-
mation, such as the customer account involved or the merchandise handled
during the process. In our previous research prototype PISA (cf. [32]), we
manually added an activity that created a database entry containing the ID of
the process instance and the OID of the business object treated in the pro-
cess. This artificial linking of the audit trail data with (semantic) business
data provides the opportunity to navigate from a particular process instance
to the business context and allows for the application of numerous evalua-
tion methods.
The linking of audit data with business object information can create a
number of problems for the data warehouse designer:
• Not every workflow context is a business object. If, during the course of
a workflow instance, data in different application systems is used, an
artificial wrapper for this particular data set has to be created for every
workflow instance in order to make this data set accessible through a
unified OID. While object-oriented workflow management systems
(e. g. those running inside a java-based application server) provide a
native support for this concept, most stand-alone workflow systems are
designed to work with different applications without an additional wrap-
per.

Figure 5: WfMC Audit Trail Format (excerpt)

• Business data is subject to side-effects. The notion of application data • Processes evolve. The opportunity to model and change processes with
was introduced by the WfMC to make sure, that mission critical data a workflow management system is significantly easier than the adjust-
was not locked from use while being used by a (potentially long-run- ment of hard-coded process logic in application systems. Therefore,
ning) workflow instance. Reverting this view, application data can be over time, enterprises using workflow applications will adjust these to
manipulated by applications outside of the workflow context at any match their changing environment. The change of the process structure,
time. If audit trail data is transferred through a batch procedure once a e. g. the replacement of an activity with another results in a change of
day, intermediate changes to the business data are invisible to the data the audit trail data during process enactment. If an analysis is to be per-
warehouse. To avoid this problem, business data would need to be trans- formed across a collection of process instances, in many cases only
ferred to the data warehouse as soon as the workflow instance treating those instances will be relevant, whose execution path is identical, i. e.
this business object is finished. For performance reasons, this will only whose set of activity instances is identical at the type level. Therefore, it
be feasible in few cases. is vital for the validity of the data warehouse information, that for every
workflow instance the underlying workflow model variant is recorded.
• Synchronicity is not always desired. From a controlling point of view, it
is possible that the most recent version of a business object is of more 3.3 Evaluation Dimensions
value than the version that was current at the time of workflow enact-
During the evaluation of workflow audit trail data, different evaluation
ment. When, for example, a query is issued “Show all order fulfilment
dimensions are used and various evaluation methods can be executed on the
processes from Q1/2000 for customers residing in the U.K.” the result
data repository. While not all evaluation requirements of the data warehouse
set may or may not reflect the customers domicile during process enact-
users may be known in advance [8], some basic evaluation perspectives can
ment. The correctness of the result set depends very much on the objec-
be predefined. In its most basic format, workflow audit trail data supports
tive of the person analyzing the data. If the purpose of the query was the
analyses from a process perspective as well as from a resource perspective.
identification of buying patterns for a direct marketing campaign, subse-
quent relocations of customers should be included. If, however, the goal 3.3.1 Process perspective
of the query is the identification of demographic profiles of specific
The process evaluation perspective focuses on information that can be
geographic regions, the customer addresses as of the time when the pro-
derived from an analysis of the temporal and logical information associated
cesses were executed is relevant. This example shows, that the purpose
with the activities and transitions recorded in the audit trail. Without any
of the data recipient determines the synchronicity requirements for audit
modification, this information is derived from the activity identifiers as well
trail and business data during the import phase.
as the timestamps and status recorded in the audit trail.
• Recording audit trails creates potentially large amounts of data. LEY-
Typically, a workflow audit trail contains the following timestamps:
MANN and ROLLER have pointed out the risk of growing audit trails [14].
They estimate the typical number of log entries per activity to be 5 (or 1 • Instantiation time of the process
KB for a fully featured audit trail entry). For 10,000 processes with 10
• Starting time of the process
activities per day this creates an audit trail of 500 MB every day, not
including the associated business object information. Numbers like • For each activity
these are not unusual. The average number of pages received in the mail
• creation time (the time when the activity becomes ready for execu-
room of the insurance company described in section 4 is on average
tion)
29,000 per day, resulting in 8,900 different cases that are instantiated
every day. • activation time (the time when the activity was activated by a work-
flow participant, e. g. by selection from the participant’s work list)
 • completion time (the time when the activity was marked as com-
pleted either by the system or the participant in case of a manual
activity)
• Completion time of the process
Using these timestamps and the differences between them the following Activity
information can be derived easily: Level
Analysis
• Cycle time of the process (the time elapsed between the instantiation or Segment
starting time and the completion of the process) Level
Analysis
• Net processing time (the sum of the time elapsed between the activation
and completion of each activity of the process) Process
Level
• Idle time (the time elapsed between the creation time and activation Analysis
time of each activity).
• Frequency of process occurrences (which processes were executed in
which frequency at a certain hour/day/week/month etc.)
Figure 6 shows different analysis scopes, namely at the activity, segment
and process level. The resulting analyses can be performed at the instance
level (for the analysis of the behavior of single activities and processes) as
well as at an aggregated level (for the analysis of the behavior of activities
and processes over time or the computation of average values).
A frequency analysis can provide information about the likelihood of the
execution of certain activities and/or process paths. This information is of Figure 6: Analysis at the activity, segment and process level
purely statistical value, if no connection can be made to the business objects
that were treated in the different process instances (i. e. the reason, why a slow) processes may yield some information about the reason for this non-
process instance was executed in a certain fashion). This does not mean, homogeneous behavior.
however, that these results are not meaningful at all. For example, probabili-
ties about the execution of certain process paths could be inserted into the 3.3.2 Resource perspective
workflow model for simulation purposes, enabling the process designer to Most workflow management systems allow for the recording of the per-
experiment with different workloads or staff assignments before actual former ID in the audit trail. In our experience, during the introduction of a
changes are made to the organization. workflow management system, this technical option often creates mistrust
and resistance among the future users of the system. The measurement of
In combination with the semantics of the process and the business individual performance through an information system is subject to privacy
objects involved, the results of a frequency analysis can be used to identify laws in some countries. In Germany, for example, the measurement of
potential weaknesses or bottlenecks along the process. For example, if the employee performance through information technology requires the agree-
variance of the cycle time for an order fulfilment process is particularly ment of the employee’s union. If no employee’s representatives exist at the
large, the analysis of the business objects associated with extremely fast (or company in question, a visible notice has to be posted at the workplace under
surveillance. Therefore, a sensible measure during the initial setup of a object. For a supply chain process, this information could be used to classify
workflow management system is the creation of a formal charter that con- suppliers into A, B and C suppliers, according to the frequency of exceptions
tains the level of detail for the audit trail as well as the evaluation purposes, generated within the processes triggered by the different companies. Using a
this data is collected for. combination of business object and audit trail data, an analyst can navigate
along the process structure as well as the data structure of the associated
Combining temporal information about process performance with infor-
business object. Starting from the process perspective (“what is the average
mation about the resources involved provides opportunities to create mean-
cycle time of an invoice auditing process?”) the analyst could single out pro-
ingful financial indicators. If the cost factor for the utilization of a resource is
cesses according to their temporal behavior (“display those process
known, the (personnel) cost for the execution of a process instance an be
instances, whose cycle time exceeded the average cycle time by more than a
computed by multiplying the processing times of the activity instances with
week”). Using this information, the analyst could traverse to either the
the cost factor of the workflow participants. This simplified approach does
resource or the business object perspective (“out of this set, display the name
not take into account work that is performed outside the scope of the work-
of the supplier, whose invoices exceeded the cycle time the most”). From an
flow management system, e. g. answering to a customer call while another
enterprise controlling perspective, these are the most valuable insights, the
customer order is being processed inside the workflow system. However, for
use of workflow audit trail data can generate.
organizations that use workflow systems for administrative processes (where
personnel cost account for the majority of process cost), the use of workflow The number and kind of evaluations available at the business object per-
audit trail data can lead to a better accountability of overhead cost. spective is determined mainly by the semantics of the business objects. Ide-
ally, the evaluations are defined at the time, the workflow application is
Another evaluation using the resource perspective is the analysis of
being implemented, to ensure that the data required for the analyses is
resource utilization at the activity or process level. This analysis answers
recorded (either through the workflow management system directly, or
questions such as “How many different participants were involved in an
through external programs).
ordering process?” or “How many work items typically reside in the work
list of an accounting clerk on average?” 3.4 Designing a Process-Oriented Data Warehouse
In [32] we have discussed the application of SASSONE’S hedonic wage Figure 7 shows the schematic design of a data warehouse that integrates
model to workflow audit trail data. By classifying workflow participants into workflow audit trail data with business object information. The extraction,
groups such as executives, experts and administrative staff, and by assigning transformation and loading layer imports workflow audit trail data as well as
categories to workflow activities that reflect these groups (for example exec- business object data into a common repository. The data is formatted accord-
utive tasks, administrative tasks, expert work), the analysis of audit trail data ing to the respective metadata sets that describe the format and semantics of
provides information about the accuracy of work assignments. Using the the underlying raw data structures.
hedonic wage model, the financial effects of organizational restructuring on
The evaluation engine performs various evaluation methods on the pro-
process performance can be visualized.
cess-oriented data warehouse and delivers the results as an evaluation result
3.3.3 Adding Perspectives through Business Objects set. For performance and warehouse size reasons it may be feasible to create
data marts for specific processes or analysis purposes (e. g. a specific data
The combination of audit trail data with semantic information about the
mart for activity-based costing analyses). Finally, the visualization layer
business objects allows for a multitude of evaluations. Depending on the
reads the evaluation results and presents them to the user in a usable format
process semantics as well as the information available about the business
(e. g. as a table or a graphical chart, depending on the type of result data).
objects, numerous evaluation dimensions can be derived.
The arrows on the left and right of the data warehouse diagram indicate
An example for an analysis originating at the business object perspective
the steps taken during design and use of the process-oriented data ware-
is the analysis of processing times grouped by the vendor of the process
 information necessary for the fulfilment of these requirements can be identi-
fied (what does the user need to know in order to solve the problem at
hand?). The information requirements are the major guideline for the selec-
tion of an appropriate evaluation method (which evaluation delivers the
results needed?). Finally, the choice of an evaluation method determines the
required entries in the process-oriented data warehouse (what source data is
required for the execution of this evaluation method?).
This top-down approach at the design of a process-oriented data ware-
house illustrates, that the managerial controlling requirements need to be
stated already at the time the workflow is designed. Subsequent changes to
the evaluations required may result in necessary adjustments of the underly-
ing workflow application, which can be both time-consuming and expen-
sive.

4 Case Study: Using Workflow Audit Trail Data in an

Insurance Scenario
4.1 Preconditions
In the following section we outline the practical relevance of our approach
through a case study. The study was conducted at a medium-sized German
insurance company with 1,200 employees. The company had conducted a
large business process re-engineering project two years before the case
study. During the course of this project, the major processes affecting the
insurance-specific departments had been documented and grouped into clus-
ters. Based on statements from employees, the average processing time for
activities was documented (with a resolution of minutes). Based on this
information, an activity-based costing system (ABC) had been implemented.
After the new system had been in use for two years, it became apparent
that the underlying business processes had changed. Also, certain informa-
tion that was required for purposes such as capacity planning was not avail-
able from the existing system. A project was set up to investigate the
Figure 7: Design of a Process-oriented Data Warehouse potential benefits of a workflow-driven process information system for the
company.
house. During the operation of the process-oriented data warehouse, a con-
tinuous feed of audit trail data will be integrated (either continuously or via 4.2 Information Availability
batch procedures), evaluated and presented to the user. The design process Figure 8 illustrates the existing activity-based costing system. It is character-
reverses this flow. Starting with the business requirements of the user (what ized by a variety of data feeds, most of which are transferred manually into
kind of business problem or decision needs to be addressed by the user?) the the activity-based costing system. The activity structure of the business pro-
cesses was transferred manually from Excel sheets into the proprietary that resulted in the execution of a (traceable) transaction in the transaction
ABC-system. Transactions from the operational insurance application sys- processing system. This way, the ABC-system documented 20,000 cases of
tem were assigned to activities according to a transformation schema devel- customers cancelling their auto insurance per year. However, the actual num-
oped during the BPR project. Since the number of transactions were ber of cancellations was much higher, since many customers could be con-
recorded in the legacy system, the log files were used to compute the actual vinced to revoke their cancellation request, before it was put into effect using
number of activities performed in a certain period. Using the estimated pro- the transaction processing system. These cases, even though their processing
cessing times per activity, the required personnel capacity per period was consumed time and kept resources, did not appear in the ABC-system. Fur-
computed using a custom Excel application. This data was combined with thermore, changes to the transaction processing system jeopardized the link
the activity structure and the internal accounting information, which was col- between the BPR activity structure and the transactions, since the measured
lected in a data warehouse application and manually transferred into the amount of time necessary to perform a certain activity was tied to the trans-
ABC-system. action used within the activity. When a transaction was changed the process-
ing time for the associated activity changed, but this change was not
documented anywhere.
Internal Internal Internal
Accounting accounting
PC
accounting sheets 4.3 Improvement through Workflow Data Integration
System sheets SAS (manual transfer)
Data Warehouse After the identification of these shortcomings, the concept for a workflow-
based data warehouse was designed, that should be used as a data feed for
the existing ABC-tool. Figure 9 shows the structure of the target scenario.
BPR
Process
Activity structure Activity-based The central component of the future controlling infrastructure is a data ware-
(manual transfer) Costing Tool
documentation house, that serves as the central repository for all operational data the needs
Personnel capacity used to be evaluated. The workflow audit trail is transferred into the data ware-
per activity (manual transfer)
house using two separate import filters. On the one hand, the process and
Insurance
Transaction activity information is extracted for the process perspective evaluations. On
Processing Log files
System
the other hand, resource information (which resource performed which
(legacy)
Number of (specific)
activity) is extracted for resource perspective evaluations. In order to provide
Data
transactions semantic information about the process structure, to-be information about
Data Data per period
Data pool
the processes and activities is transferred into the data warehouse at build-
time. Manual activities (for example phone calls), however, will still not be
Number of (specific)
Transformation Transformation recorded by the new system.
Operational Number of to activity Activities to personnel
transactions
Statistics transactions quantities performed capacity used
per period
(BPR factor) (BPR factor) In order to link the audit trail information to operational business data,
two tables need to be maintained separately. One table contains the relation-
ship between the process instances and the insurance contracts or the cus-
Figure 8: Controlling Data Flow (As-Is) tomers involved. Using this table as a bridge, a drill-down from a single
process to the customer who triggered the process can be performed. The
Besides the fact that the whole evaluation process was time-consuming other custom table contains the relationship between the workflow partici-
and error-prone, due to many media-breaks and manual transfers, the results pants and the internal cost accounting structure of the enterprise. Since
of the ABC-system did not reflect the actual operations at the insurance workflow management systems only record a technical ID of the performers
company. For example, only those processes appeared in the ABC-system,
 infrastructure and work packages for the realization of the proposed infra-
structure have been added to the project plan.
Cost Accounting
Cost Accounting Structure
Manual
Activities*
5 Discussion
Internal Internal
Accounting accounting There are relatively few research projects that aim at combining data ware-
System sheets
Imaging/
DMS housing concepts with workflow management systems. The existing sources
can be grouped into two categories: Those who apply workflow concepts to
the data warehouse domain and those who describe the analysis of workflow
Process
Date/Time Received
Runtime
Quantity
Data
audit trail data.
WfMS ABC-data
Data Warehouse
WfMS Audit Trail Resource
Quantity 5.1 Applying Workflow Concepts to Data Warehouse Design
Workflow
Feed-back
Models Activity-based
Costing
Sources that use workflow technology for the design of data warehouses aim
Process and Process
System mainly at the design phase of a data warehouse. A notable exception is the
Workflow
Modeling
and Activity
Structure
work by LIST ET AL. who aim at designing a generic data warehouse for
Relationship
Process-ID /
Customer business process evaluation [15]. Since the retrieval and transformation of
operational data into the data warehouse is a frequent process, whose steps
Transaction
Processing
Systems and
Customer and Relationship
Workflow-Roles/
are structured, it is a potential candidate for workflow automation.
Contract Data Cost Accounting
Database
Structure
Systems BOUZEGHOUB et al. describe the modeling of the data warehouse refresh-
ment process as a workflow [3]. They use an event-driven approach to work-
Figure 9: Controlling Data Flow (To-Be) flow modeling to trigger various parts of the refreshment process either after
a certain timer has expired or after a certain condition in the data warehouse
who executed activities, this table is required to allocate cost factors to the has occurred. In particular, they distinguish between Client-driven refresh-
workflow participants. ment, when a user activity causes an update to the data warehouse structure,
The realization of the process-oriented data warehouse will enable the Source-driven refreshment, when the changes to the source data of the data
insurance company to evaluate its operational business processes in a timely warehouses trigger a refreshment process, and ODS-driven refreshment,
fashion and with a much higher level of accuracy than currently possible. It PATTERSON discusses workflow and data warehouse concepts and identi-
should be noted, that the existence of a workflow infrastructure is the major fies several areas in the data warehouse design process that could benefit
precondition for the realization of such an information system. At the com- from the use of workflow technology, such as handling update anomaly
pany in question, a project was initiated to lay the foundation for a future problems [21].
data warehouse by selecting and implementing a combined document man-
agement and workflow infrastructure. Even though the financial benefits of 5.2 Analyzing Workflow Audit Trail Data
an improved process controlling infrastructure are hard to determine, the The application of data warehouse concepts to workflow technology can be
financial savings from the introduction of the document management and divided into three groups. The first group of publications in this area deals
workflow infrastructure were sufficient to provide a return on investment with the technical facilities necessary for the logging of the audit trail data
just three years after the deployment of the new system. The ability to per- (e. g. [27]). A second group of papers describes the reverse engineering of
form detailed process analyses is perceived as an added benefit of the new workflow processes from audit trail data [1]. Finally, a number papers can be
identified, that focus on the ex-post analysis of workflow audit trail data with proposal for an API has been submitted to the Coalition, but so far no
methods known from enterprise controlling ([5],[22],[19],[28],[32],[33]). progress has been made to change the interface from a data format specifica-
tion to a functional specification.
5.2.1 Technical Facilities for Audit Trail Generation
MCGREGOR discusses the existing WfMC standard with regard to a pro-
KOKSAL et al.discuss the management of audit trail data in the distributed
totype, that uses this information to generate the process perspective of a bal-
workflow management system Mariflow [11]. The paper focuses on techni-
anced scorecard [16]. Her work aims at the development of a closed-loop
cal issues regarding the storage of audit data and economic queries on dis-
system that takes audit trail information as an input and delivers advice to
tributed data sources, but the authors do not address the business value of the
process designers, which aspects of the workflows analyzed could be opti-
history data.
mized.
An approach for the tracking of history information in a distributed
One of the first sources on workflow based controlling can be found in
workflow management system is presented by MUTH et al. [18]. Within the
[19]. MCLELLAN provides an overview of the analysis of historical process
prototype Mentor-lite, data about current and past workflow instances are
data. He discusses the evaluation of workflow history data as workflow met-
kept in a temporal database that can be queried either at runtime or for ex-
rics. The controlling applications described are statistical evaluations as well
post analyses.
as the run-time detection of late cases and overdue tasks.
5.2.2 Reverse Engineering Using Audit Trail Data SHOHAIEP ET AL. present a methodology for the identification of process
AGRAWAL et al. use data mining techniques to create workflow models from knowledge through the analysis of work practices [24]. Even though their
audit trail data [1]. The purpose of their project is to use data from the ad-hoc method is called “Workflow*BPR”, it does not require the use of a workflow
execution of processes, to subsequently identify common rules and proce- management system. Taking the basic idea of determining the knowledge
dures and to create workflow models using a bottom-up approach. The incorporated in a process, workflow audit trail data can be perceived as a
authors have tested their approach against artificial data sets and audit trail “knowledge store” in its own right. For example, the experience of a work-
data from a live workflow installation. However, the practical use of the flow participant could be computed taking the number of times into account
methodology presented requires the existence of a flexible workflow tool, that this participant has performed a certain activity type. When this infor-
that records processes while they are being created on the fly. Despite the mation is fed back into the workflow management system, new staff alloca-
obvious demand for such a tool, the current workflow market is lacking tion methods could be implemented, such as “assign this activity to the most
products with this kind of execution flexibility. experienced person available”.

5.2.3 Workflow-based Controlling in the Literature DERSZTELER discusses the design of a controlling tool called WorkFlow
Analyzer in [5]. A partially functional prototype was implemented on the
Whereas the controlling of processes using workflow audit trail data has
basis of Forest&Trees. Based on audit trail data from the workflow manage-
been analyzed in the German literature to some extent (for example
ment system WorkParty by Siemens Nixdorf and to-be data from the busi-
[5],[22],[28]), there are relatively few English sources dealing with this topic
ness process modeling tools ARIS and Bonapart, the prototype provides
(for example [19],[32],[33]).
several quantitative evaluation methods. However, the combination of the
The WORKFLOW MANAGEMENT COALITION Interface 5 specifies the ele- audit trail information with business data was not realized using this
mentary information a workflow management system should record the exe- approach.
cution of workflow instances [29]. The existing standard provides a data
In [22], RAUFER discusses the controlling of workflow-based processes
format for the audit trail data as well as guidelines, which events should be
using a case study from the manufacturing domain. In his example, the
recorded. However, the evaluation of this information is not addressed in the
author focuses on a specific process, which he enhances with cost informa-
WfMC standard. After the publication of the current standard document, a
tion as well as target work- and cycle-times. The presented prototype, which exist, that put the business process into the center of attention. Workflow-
is based on the workflow management system COI, is targeted specifically at based controlling tools that focus entirely on information contained in the
the process analyzed by the author. As a result, the system architecture is not audit trail are limited to the quantitative analysis of the event-driven history
easily generalizeable. of process execution.
A similar prototype is presented by WEISS in [28]. He developed a work-
flow-driven activity-based costing system for the commercial workflow
management system Staffware. The focus of this approach is the realization
of a single evaluation method, therefore, the resulting system is not designed
to be extended by additional evaluation methods.
The COPPA project (Computer-based Process Performance Measure-
ment) conducted at the University of Fribourg, Switzerland, deals with the
design of a performance measurement system ([12], [13]). Using a three
stage approach, the authors first surveyed the market and corporate practice
of performance measurement. During a second stage the architectural and
functional requirements of a performance measurement system were out-
lined, and during a third phase, a prototype of the performance measurement
system was implemented. In relation to workflow-based controlling, the
authors position process performance measurement at a higher level of
abstraction, that includes information about the strategic positioning of an
enterprise, whereas a workflow-based controlling system is mainly focused
on the analysis of operational data.

6 Summary and Outlook

Workflow-based controlling provides companies with the ability to accu-
rately measure the operational performance of business processes. Com-
bined with data warehouse technology and operational business data,
complex evaluations can be performed, that help enterprises to assess their
current situation more precisely than the sole use of traditional key perfor-
mance indicators. The increasing maturity of workflow and data warehouse
Adapted from [12], pp. 425 ff.
products provide an important precondition for the implementation of work-
flow-driven controlling systems. Figure 10: Positioning of Workflow-based Controlling

It should be noted that workflow-based controlling does not replace other Our research enhances workflow-based controlling approaches to
controlling mechanisms of the enterprise, but enhances them significantly. include business information that is not stored in the traditional audit trail.
Figure 10 shows the positioning of workflow-based controlling in relation to This way, the scope of potential evaluations is widened to include qualitative
other controlling techniques. While strategic controlling instruments like the information, and the differentiation of business processes through their asso-
Balanced Scorecard focus on qualitative aspects of the enterprise, and tradi- ciated business objects allows for a more detailed segmentation of the analy-
tional controlling methods rely on financial information, few techniques
sis domain. Since this approach relies on the workflow-enabled business ings of the 12th IEEE Intl. Conference on Data Engineering, New
processes of an enterprise, there will be a number of blind spots at the enter- Orleans, LA, February (1996).
prise level (i. e. manual processes, managerial processes without workflow [3] Bouzeghoub, M, Fabret, F. and Matulovic-Broqué, M., “Modeling
support etc.). However, the development of enhanced workflow-based con- Data Warehouse Refreshment Process as a Workflow Application”,
trolling systems can serve as a data-feed to other controlling approaches, like in: Proceedings of the International Workshop on Design and Man-
the Process Performance Measurement Systems discussed by KUENG in [12] agement of Data Warehouses (DMDW ’99), eds. Stella Gatziu, Man-
or the process perspective of the Balanced Scorecard. fred A. Jeusfeld, Martin Staudt, Yannis Vassiliou, Heidelberg, 6.1-
6.12 (1999).
We have discussed the technical and organizational aspects of building a
[4] Candle Corp.: www.candle.com
process-oriented data warehouse and used a case study to illustrate the prac-
[5] Derszteler, G., “Prozessmanagement auf Basis von Workflow-Syste-
tical relevance of this type of application. Nevertheless, many aspects of
men”, Josef Eul Verlag, Lohmar et al. (2000).
workflow-based controlling systems require further research.
[6] Emery, P., “The Workflow Market: A Global Perspective for 2000”,
So far, existing papers focus on the realization of single evaluation meth- in: Excellence in Practice, Volume IV, Innovation and Excellence in
ods, like activity-based costing [28]. An analysis, which existing evaluation Workflow and Knowledge Management, ed. Layna Fischer, Future
methods can be applied in the context of a workflow-based controlling sys- Strategies, Lighthouse Point, 41-47 (2000).
tem based on the attributes recorded in workflow audit trail data would pro- [7] Galler, J. and Scheer, A.-W., “Workflow-Projekte: Vom
vide further insight about the coupling of audit trail data and business object Geschaeftsprozessmodell zur unternehmensspezifischen Workflow-
information. Anwendung”, in: Information Management, Vol. 10 No. 1 (1995), 20-
28.
Also, the handling of changing process models variants over time resem-
[8] Gardner, S. R., “Building the Data Warehouse”, in: Communications
bles the problem of schema evolution in databases. Further research in this
of the ACM, Vol. 41, No. 9, (1998), 52-60.
direction would help to open workflow-based controlling systems to incor-
[9] Heilmann, H.: “Die Integration der Aufbauorganisation in Workflow-
porate audit trails from workflow management systems that support ad-hoc
Management-Systeme” in: Information Engineering, ed.: Heidi Heil-
processes.
mann et al.. Oldenbourg, Munich, Vienna 147-165 (1996).
In the context of the insurance project described in section 4, we are cur- [10] Inmon, W. H., Imhoff, C. and Sousa, R., “Corporate Information Fac-
rently working on a prototypical implementation of the concepts presented tory”, 2nd Edition, Wiley, New York et. al., (2001).
in this paper using a java-based workflow management system which will be [11] Koksal, P., Arpinar, S. N. and Dogac, A., “Workflow History Man-
evaluated using operational data from the business processes of this com- agement”, in: Sigmod Record 27 (1998) 1, 326-348 (1998).
pany. [12] Kueng, P., “Supporting BPR though a Process Performance Measure-
ment System”, in: Business Information Technology Management,
7 References ed. P. Banerjee et al., Har-Anand Publications, New Delhi, 422-434
(1998).
[1] Agrawal, R., Gunopulos, D., Leymann, F., “Mining Process Models
[13] Kueng, P., “Process Performance Measurement System: a tool to sup-
from Workflow Logs”, in: Advances in Database Technology - Pro-
port process-based organizations”, in: Total Quality Management,
ceedings of the 6th International Conference on Extending Database
Vol. 11, No. 1 (January 2000), 67-86 (2000).
Technology, ed. Fèlix Saltor, Isidro Ramos and Gustovo Alonso,
[14] Leymann, F. and Roller, D., “Production Workflow - Concepts and
Valencia, 23-27 (1998).
Techniques”, Prentice Hall, Upper Saddle River (2000).
[2] Alonso, G., Agarwal, D., Abbadi, A. E., Kamath, M. and Gunthor, R.,
[15] List, B., Schiefer, J., Tjoa A. M. and Quirchmayr, G., “The Process
“Advanced transaction models in workflow contexts.” in: Proceed-
Warehouse - A Data Warehouse Approach for Business Process Man-
 agement”, in: e-Business and Intelligent Management - Proceedings neering (CAiSE 2000), eds. Bengt Wangler and Lars Bergman, Stock-
of the International Conference on Management of Information and holm, 446-461 (2000).
Communication Technology (MiCT 1999), Copenhagen, Denmark [27] Weikum, G.: “Workflow Monitoring: Queries on Logs or Temporal
(1999). Databases?” Position Paper at HPTS'95. Pacific Grove, (1995).
[16] McGregor, C., “The Impact of Business Performance Monitoring on [28] Weiss, D., “Prozesskostenrechnung und Workflow Management”,
WfMC Standards”, presentation at the Workflow Management Coali- Deutscher Universitaets-Verlag, Wiesbaden (1998).
tion meeting, New York, NY, May 2001. [29] Workflow Management Coalition, “Audit Data Specification. Version
[17] Moore, C., “Workflow Goes Mainstream”, Giga Information Group 2”, Document Number WFMC-TC-1015, Winchester (1999).
Planning Assumption, Cambridge, MA, April 21, 2000. [30] Workflow Management Coalition, “Terminology and Glossary” 3rd
[18] Muth, P., Weissenfels, J., Gillmann, M. and Weikum, G., “Workflow Edition, Document Number WFMC-TC-1011, Winchester (1999).
History Management in Virtual Enterprises Using a Lightweight [31] zur Muehlen, M. and Allen, R., “Embedded vs. Autonomous Work-
Workflow Management System”, in: Proceedings of the Ninth Inter- flow - Putting Paradigms into Perspective”, in: Excellence in Practice,
national Workshop on Research Issues on Data Engineering, Sydney, Volume IV, Innovation and Excellence in Workflow and Knowledge
148-155 (1999). Management, ed. Layna Fischer, Future Strategies, Lighthouse Point,
[19] McLellan, M.: “Workflow Metrics - One of the great benefits of 49-58 (2000).
workflow management” in: Praxis des Workflow-Management, eds. [32] zur Muehlen, M. and Rosemann, M., “Workflow-based Controlling”,
Hubert Oesterle and Petra Vogler, Braunschweig, 301-318 (1996). in: Proceedings of the 34th Hawai’i International Conference on
[20] Naef, A., Schuler, Ch. and Schuldt. H., “Monitoring komplexer Dien- Social Sciences, ed. Raplph Sprague Jr., Wailea (2000).
ste in unternehmensübergreifenden Prozessen am Beispiel von SAP [33] zur Muehlen, M., “Workflow-based Process Controlling - Or: What
R/3 Business Workflows”, in: Proceedings of the 9th GI-Conference You Can Measure You Can Control”, in: Workflow Handbook 2001,
Datenbanksysteme in Buero, Technik und Wissenschaft (BTW 2001), ed. Layna Fischer, Future Strategies, Lighthouse Point, 61-78 (2000).
85-94 (2001)
[21] Patterson, J. E., “Workflow and Data Warehousing Taxonomy”, Mas-
ters Thesis, University of Texas, Arlington (1996).
[22] Raufer, H., “Dokumentenorientierte Modellierung und Controlling
von Geschäftsprozessen”, Deutscher Universitaets-Verlag, Wiesbaden
(1997).
[23] Sassone, P. G., “Cost-benefit methodology for office systems”, in:
ACM Transactions on Information Systems, Vol. 5, No. 3, (1987),
273-289.
[24] Shohaiep, G., Housel, T. J. and Kanevsky, V.: “Workflow-BPR White
Paper: Business Process Auditing with Workflow-BPR”, White
Paper, University of Southern California, (1997). URL: http://www-
rcf.usc.edu/~housel/IOM499/wp.html [2001-07-15].
[25] Tivoli Corp.: www.tivoli.com
[26] Vassiliadis, P., Quix, C., Vassiliou, Y. and Jarke, M., “A Model for
Data Warehouse Operational Processes”, in: Proceedings of the 12th
International Conference on Advanced Information Systems Engi-