There is a variety of applications that can benefit from the ability to find optimal or good solutions to a proposed problem, automatically. The artificial intelligent (AI) community has been actively involved in efficient problem-solving in complex domains such as military or spacecraft problems with successful results. In this paper, we describe the integration of AI planning techniques with an existing workflow management system. We show how these techniques can improve the overall system functionality and help automate the definition of business processes. The work is based on a short study carried out at BT research laboratories as part of a larger programme that aims to provide technologies for a new generation of business support systems.
Every organisation tries to shape its processes to optimally suit the market and offer the best service to the customer. When an organisation is analysed with the purpose of identifying possibilities for optimising its routines and procedures, three basic facets are outlined:
•
A task or activity describes what should be done.
•
An organisation model describes who should do something.
•
An information model describes which information is needed to perform an activity.
From a historical perspective, the first issue that companies focused on was the design of organisational units. In the years to come, control logic (when should something be done) is set to play a central role in connection with optimisation of business processes.
Numerous issues need to be considered when designing business processes [6] and [18] and implementing them in IT systems. These include: reusability of past processes, accessibility from the different agents, consistency of usage, and selection of the right model.
In recent years, a new class of software infrastructure product to support business processes has emerged: workflow management systems (WfMS) [4], [17] and [25]. A WfMS can provide active support to a business process by controlling the routing of work around the organisation automatically. This is done based on input describing the flow, the decisions, the exceptions, the resource to be used, etc. It co-ordinates user and system participants, together with the appropriate data resources, which may be accessible directly by the system or off-line to achieve defined goals by set deadlines. The co-ordination involves passing tasks to participants' agents in correct sequence, and ensuring that all complete their tasks successfully. In case of exceptions, actions to resolve the problem can be triggered, or human operators alerted.
Prior to WfMS, many enterprises created special-purpose bespoke applications to support their processes. The advantage of WfMS-based solutions is that the workflow representation is explicit, and separate from the application code. This means that a WfMS can be customised quickly to support a new business or process, and that workflows are relatively easy to modify, should a process change. However, current WfMS do not address all aspects of the problem. Specifically, they do not deal with scheduling or resource management/allocation. Similarly, while they provide means of generating exception events when things go wrong they do not have a built-in re-planning function. They do, however, provide interfaces so that application-specific modules performing these functions can be integrated.
Recently, there has been considerable interest in the application of artificial intelligence (AI) techniques to WfMS. The lack of maturity that the area of workflow management presents due to its short history can be addressed by introducing techniques from other fields. Some researchers have seen the advantages of the integration of this approach, as shown by the existence of a technical co-ordination unit of the European research network on planning and scheduling, PLANET [28], on applications of planning and scheduling to workflow. This has lead to some exploratory work reflected in a roadmap and some published papers [16], [21], [23] and [26]. Although the MILOS project [7] of the AI Group at the University of Kaiserslautern and the software process support group at the University of Calgary or the AI group at Edinburgh University in the TBPM project [15] and [31] have addressed the problem, to date very few tools have been developed using these ideas [24].
In this paper we highlight the improvements that a legacy system can gain by incorporating AI planning techniques into its day-to-day operation. We first introduce the phases that both systems have in common. After this, Customer Orientated System for the Management Of Special Services (COSMOSS), a purpose-built legacy workflow application in use at BT is described. Then we review contingent planners, an AI technology that addresses issues found in the COSMOSS application. After this, the similarities between both workflow management and planning are presented. We conclude with an example, based on a COSMOSS scenario that illustrates how ideas from the two fields may be merged.
We have shown the potential of applying AI planning techniques within WfMS. This benefit will be realised as much by introducing workflow specialists and software engineers to planning concepts and representations as by direct application of planning software.
We also want to outline the issues that AI planners can gain with this approach. Generally, to specify the domain theory, a deep understanding of the way AI planners work and its terminology is needed. However, if we use an existing system, the description language is closer to the user (at least quite familiar for the COSMOSS user). Therefore, we try here to solve the planning domain modelling task by using BPR representation models and technology. In fact, it is really a symbiosis in the sense that, once we have defined the domain using a WfMS, AI planning and scheduling technology can help in the automatic generation of process models.
In this paper, we have focussed on how the contingent planner, Cassandra, can help to automate the design of appropriate templates in a legacy system used to support the business processes at BT
