What could AI Planning and Scheduling do for BPM?

After coming back from my presentation at BPM 2010 Demo Track (and my very first time in a conference of Business Process Management) about Smart Process Management, i would like to point out some thoughts I have in mind, mostly regarding at the future possibilities of using Artificial Intelligence Planning and Scheduling techniques in order to cover some of the gaps that still exists in BPM.

Even if using P&S for BPM can be very promising, I think that clearly the main research problem that have to be focused within this area is the dynamism that a Process can suffer during its life-cycle. When I say dynamism, I refer to the changes in the environment and/or the process model itself, that could happen once that the process has been started. These changes in the environment are the main reason for the emergence of a new area called “Adaptive Case Management”, and a very interesting panel discussion took place in the conference, driven by Keith Swenson, Dana Khoyi, Dermot McCauley and Jacob Ukelson. I was lucky to talk with Dermot McCauley, and what he told me is basically exposed in this excellent blog post at Column2, which briefly overview ACM as well.

In this panel discussion, K. Swenson gave a good definition of what ACM is, by using the “House” serial as example, and where decisions are made “on the fly” by the doctor, where the execution path cannot be predefined. Then, I came out with an issue: even if this is inevitable, doctors also try to do their own processes repeatable, capturing their models into what they call Clinical Guideline Protocols. So, as it is said in Column 2 blog, they look for a “happy path” showing what the patient treatment should be like. Hence, “this is not purely unstructured process, where there is no predefined model, but dynamic BPM where the model is predefined but can be readily changed while in flight.“. We can see that BPM and ACM can coexist.

So, if P&S techniques are going to be used in order to help to control and manage the technical problems associated with this dynamism, we must have clear in mind that P&S should be able to manage *at least*:

* exogenous events that can occur, even without having previously been foreseen in the process model, that changes either the model or the execution path.
* changes in the process model itself (skip tasks, add new tasks).
* changes in the enviroment (conditions of the knowledge workers, new knowledge workers, controlling the busy or available states).
* different (and possibly concurrent) execution paths, i.e. cases can be interrelated.

From my point of view, I see that P&S could be able to offer these features, by interleaving different planning stages, where the start and end points of these planning stages are delimited by the time at which the changes occur. Furthermore, there is already strong research open in continual planning, plan monitoring, re-planning and plan repair, which is similar and can be used for the problems arisen in this post.

Update: More insights into our very first approach in this direction can also be explored in the paper “Integrating plans into BPM technologies for Human-Centric Process Execution” presented at KEPS 2010.

3 comentarios en “What could AI Planning and Scheduling do for BPM?

  1. Great post!

    I would like to clarify something about ACM. I avoid the term “unstructured process”. I know it is used a lot, but it is misleading. In the case management scenarios, the work *is* structured — it simply is structured at the time that you do the work, and not before hand. One should not think that the work is completely unstructured, and that everyone just does what they feel at the time. Instead, the case manager is active in planning the work, coordinating the tasks, as the work proceeds.

    I use the House example to illustrate two things: one is that you can never tell in advance what you are going to do. The second thing is how when a conclusion is made about the situation, there are plenty of pre-defined processes to use: particular lab tests, particular treatment plans. They have to make those up on the fly, but rather seem to exist as “snippets” of processes that can be used at any point in time. These snippets do not need to be made from scratch every time.

    You are absolutely correct that doctors will be develop and elaborate procedures over time. These snippets are used in a slightly different way than BPM: the snippet may be changed and adapted to the particular situation. Actually, sometimes they are used exactly like BPM, that is executed as if it was a fixed program. This leads to what you aptly call a coexistence of ACM and BPM.

    I am interested in hearing more about AI planning and scheduling. I have some experience in this, just enough to know scheduling is much more challenging that it intuitively seems it should be.


    1. @Keith,

      Thanks for the comments!

      I like very much that vision of having “snippets” that you can reuse as part of your “human knowledge base” every time that “the doctor” makes a decision. I think it can fit quite nicely with my proposal of interleaving planning stages through the decision process, supported by a well-designed monitoring unit and maybe using continual and reactive planning techniques.

      Concretely, every snippet could be seen as a piece of the whole HTN (Hierarchical Task Network) planning domain, and of course this planning domain could “evolve” as the decisions are made through the process.

      I will try to give further details in another blog post, but it is mainly one of our research directions.



Introduce tus datos o haz clic en un icono para iniciar sesión:

Logo de WordPress.com

Estás comentando usando tu cuenta de WordPress.com. Cerrar sesión / Cambiar )

Imagen de Twitter

Estás comentando usando tu cuenta de Twitter. Cerrar sesión / Cambiar )

Foto de Facebook

Estás comentando usando tu cuenta de Facebook. Cerrar sesión / Cambiar )

Google+ photo

Estás comentando usando tu cuenta de Google+. Cerrar sesión / Cambiar )

Conectando a %s