Process Modelling for Complex Adaptive Systems — #NoBPMN

Complex adaptive systems (CAS) that emerge in the digital world are described as non-linear (having systemic influence, not reduced to linear cause-effect), deeply interconnected, evolving at a fast pace, without any end-state, and having feedback loops between elements to mention a few. In modelling an organizational process which enables interaction and influence on complex adaptive systems, we are going to have to give up BPMN in favor of modelling processes for learning and adaptation.

The basic elements of a business process model.

Building on the work of economics Nobel prize winner Daniel Kahneman and his ideas on slow (mode 1) and fast (mode 2) thinking, Artificial General Intelligence (AGI) researcher and practitioner Monica Anderson talks ( about the shortcomings of traditional (pre-deep learning) Artificial Intelligence in stating that it focused on solving defined, reductionist problems which ended up failing in undefined, complex problems. The reason behind this is the fact that logical reasoning is highly efficient closed, linear systems whereas complex systems and problems require methods which promote intuitive understanding.

Monica Anderson’s take on approaches based on reason and understanding in complex problems and contexts.

To build on these ideas, we can use the Cynefin framework to understand the contexts in which BPMN which is based on reasoning might work and where we should employ different types of process models which focus on understanding, utilizing feedback, learning — and using all this to adapt continuously.

BPMN is useful in domains of reasoning but we need something else in domains of understanding.

The downfall of BPMN in the complex domain is related to the fact that if the system, its actors and its conditions are constantly changing, BPMN modeling becomes a job for itself whereas it is supposed to be a support tool to show how we operate now and how we hope to operate in the future. In process modelling for complex domains, the AS-IS and TO-BE process states are not discarded completely but rather are the ever-changing byproduct of attempts at domain understanding, experimenting, feedback, learning, and adaptation.

So what might a process model which works in the domain of understanding look like? Perhaps a design thinking inspired approach might provide us with the tools to model for learning.

A great practical example if this is the talk by Raphael Grignani from Pinterest titled “Designing with Velocity at Scale” from the Interaction16 conference ( In it he describes the design process which the Pinterest design team uses to define the problem that they are solving, scope the project, use design thinking to experiment and learn, and finally testing the devised solution at scale to find whether the solution is something that should be rolled out for all Pinterest users.

The Pinterest design process is based in the double diamond model but is adapted to fit the Pinterest domain, its engineering processes, and its feedback loops (data sources).

An overview of the Pinterest design process.

The process starts off with the planning phase which is used to define the right problem and scope the project. This phase integrates strategic and business goals to the user and problem research insights to understand whether the idea is sound and what is initially required to get the thing done. Defining the right problem ends with an OKR (Objectives and Key Results) review which aims to connect company, team and personal objectives to measurable results, to promote objective-project alignment. This phase ends with the actual scoping of the project which involves the disciplines involved and results in a creative brief overview.

The planning phase of the Pinterest design process.

The actual project phase is close to the Develop and Deliver phases of the UK Design Council Double Diamond. Options ae explored, prototypes are made and the cross-functional production team is involved to evaluate whether the prototypes are doable. This is followed by what I feel is interesting and often left for little attention in our service design toolkits: setting up the experiment and plan how to learn from it. After QA and launch review the prototype is ready to be launched to a limited audience of 30K conversions/day or 250K over 28 days to gain data on how users interact with the new elements.

If feedback is positive, the project moves into the scale phase which aims to which focuses on using what was learnt, refining, developing, more QA, and scaling the solution.

The project phase of the Pinterest design process.

The design process fails to address how Pinterest maintains the results of the project if they end up going into large scale production. In using DevOps many companies have adopted a principle that whoever builds something is in charge of the maintenance. I wonder whether the production team is in charge of the maintenance and gradual customer-oriented development of the features they create.

My point in writing this post and comparing process modelling for complex domains to process modelling with BPMN is that perhaps BPMN, our turn-to tool in process modelling is a bit too low-level and constrictive to work on higher levels which integrate business, market environment, customer insights, and business opportunities implemented through projects.

High level process models such as the one by Pinterest might help us integrate the different aspects of our organization and beyond to develop a more holistic view of why and what we are doing and how.

If you’re interested in creating models and understanding on how your organization creates value to it’s customer in a holistic way, please do visit my web page at to get in touch with me.

One clap, two clap, three clap, forty?

By clapping more or less, you can signal to us which stories really stand out.