Advancing BPM by adding ‘Smart’ or ‘Intelligent’?
I have been proposing a new direction for process management with ACM Adaptive Case Management for a long time. My experience shows that businesses need a combination of goal-oriented case management with embedded process flows, rule functionality, a machine learning agent and embedded business content management to satisfy their customers and meet strategic objectives. The answer of process management experts: We have all we need. And analysts? There is no market (meaning other vendors with similar products) to observe. Yes, one can focus on short-term rather than long-term solutions to business problems.
Recently however, two new market fragments were invented by analysts: Smart Process Apps (SPA by Forrester) and Intelligent Business Process Management (iBPMS by Gartner). I am pretty happy with that because they document the business need for ACM that I have been proposing. Other vendors try to fill with a mix and match of old components. Smart and Intelligent are cute marketing terms but multiple definitions of what constitutes a component or system to be smart or intelligent are being proposed, all of them rather fuzzy. Others promote the use of ‚intelligent agents‘ within a BPMS to automate some of the process discovery. There is no explanation of how that would work in a rigid-flow process environment with heterogenous system landscapes. These intelligent BPM agents do not exist today and are derived from an approach taken in Artificial Intelligence where they remain esoteric research. The only proper intelligent agent used in BPM to this date is the Papyrus User-Trained Agent that autonomously learns and recommends user actions in ACM.
In general, intelligence or smartness is an inner capability of a being or system describing its ability to observe and learn autonomously facts about its environment to derive decisions that achieve goals. It describes the ability to generalize a learning experience and use it for different problems too. None of that exists in SPA or iBPMS. Experts have to encode logic in rules or flows which requires fairly complex work such as manual process or data mining to identify logic. Rules as used today are not generalized problem solving knowledge, but just prescriptions or restrictions. It is not the system that is smart but the knowledge engineer who defines rules or flows. The main problem is that the complex result is as static and rigid as other forms of programming.
I have discussed SPA some time ago and suggested that they are no more than the emperors new clothes on old software. They are there when you want to see them. When you look closer at iBPMS you will likewise find not much of substance that would justify the term intelligent. Are these two acronyms even on the same page? While the various incarnations can be VERY different, iBPMS just adds predictive analytics into the game. The common aspect is that they represent pre-configured applications using a variety of products that still require substantial implementation and configuration at the customer site. Ad-hoc tasks are supposed to reduce the need for predefined processes. Maybe, but it is not ACM and it does not support knowledge workers. Event, rules and content functionality through linked systems should provide agility. I don’t think so. As a matter of fact, this ‚advance’ of smart or intelligent BPM creates a higher dependency on a software stack of product versions and custom integration code. They are promoted as being feasible as a preconfigured package for one particular application.
The ‚intelligence‘ in SPA/iBPMS does not provide real-time business improvement as suggested. The main problem is the lack of defined goals and machine-learning functionality that will drive towards goal achievement. There is no autonomous or automated learning mechanism. As of now they consist of multiple systems (BI, BRE, CRM, BPM, ECM, …) without a common deployment functionality that are connected through an ESB (Enterprise Service Bus). These systems typically DO NOT have a native change management functionality across the various components. Any change is a project and thus it is NOT AGILE! The cost of creating and maintaining the ‘intelligence’ is most certainly immense and I do not know businesses that actually have staff with that kind of skill. It will be outsourced thus creating more dependency.
Why does ACM meet business needs better than SPA or iBPMS?
- ACM provides homogenous process creation and execution that can be improved by the business. IT involvement is needed for the one-time setup of data interfaces so that processes can orchestrate multiple silos. In ACM the process is not defined by experts in multiple systems and executed by transferring information through an ESB.
- In ACM processes and rules can be defined by non-technical people as part of the process execution using Natural Language. They are not externalized by experts into DMS (Decision Management System) or BRE (Business Rule Engine). Externalizing decisions increases complexity and not agility. Business knowledge is hidden and not made transparent.
- ACM supports human knowledge work with free-flow collaboration towards well defined goals. Enabling a few ad-hoc tasks in SPA and iBPMS in an otherwise rigid flow does not support knowledge work and is not case management.
- ACM detects exceptions or a lack of goal-achievement in both external systems and inside the process both through rules and pattern matching. Success is reaching a goal and not completing a predefined flow.
- Rather than the fuzzy and unproven claim of automated problem solving, ACM empowers the knowledge worker with more transparency and free-flow collaboration to solve problems. Problems rarely happen in a predictable way because then they would not be problems, which means it is impossible to perform fully-automated corrective action on unexpected problems.
- ACM uses the User-Trained Agent to suggest corrective user actions when similar problem patterns are discovered. The performer solves a problem and the UTA observes the solution packaged into a goal. In drastic difference, the predictive analytics in iBPMS are applied to what-if scenarios and require a well-defined limited state space of information within a rigid process and are utterly useless in unpredictable human interaction. Predicting a process outcome or a process problem is only possibly with rigidly defined processes. Analyzing completed rigid processes is quite useless because there is no information about future possible problems outside the predefined execution.
Knowledge (and thus intelligence) is between two ears only. (Peter Drucker)
Let’s face it: Intelligent Business Operations are performed by intelligent people and not by dumb, hardcoded agents or rigid processes written by some BPM or AI expert. ACM supports the knowledge worker with better collaboration, more freedom, more transparency, well-defined goals, better real-time information in the context of the process. The focus has to be the performer and the customer and not automation.
What an intelligent piece of software can do to improve a process is to LEARN FROM A PERFORMER. This is what the Papyrus User-Trained Agent does. It observes the complete process state-space including all data, content, and process information and autonomously discovers patterns linking events to user actions. As all actions (tasks) are organized into goals it is easy to identify the why and also how well a goal is reached. That is only possible in a homogenous system and not with information being stuck in different silos or databases. The UTA learns autonomously what intelligent people do in situations to solve problems in reaching a goal. When such problem-solving actions reappear in similar scenarios then they are recommended to other performers also. The pattern-matching, machine-learning agent does not need to understand why and encode that into rules. The tasks performed to achieve a goal are transparent and can be reused as a new process template.
This means that a problem can be assigned to an expert who will perform corrective actions. These actions are learned by the UTA and linked to the process pattern and a goal definition. In future when such patterns reappear the past expert actions are now recommended by the UTA to the current non-expert performers. All previous ad-hoc tasks of the expert are visible in the template. Goal achievement or frequent use give the actions a higher probability. Process owners, experts or coaches can remove actions or goal definitions from the libraries.
A final remark. What businesses need is a work environment that is quite different from today’s BPM or case management functionality and user interfaces. It must be so simple to use in unpredictable situations that it is adopted without change programs and culture shocks. Improving user interface and interaction is a continuous learning experience and can be quite different for various business needs. There is no boilerplate solution like suggested in SPA or iBMPS that solves all needs.
The biggest change is for management to let go of the control and automation attitude. It is expensive up-front and it is expensve in reduced customer service quality. ROI is a short-term illusion. It endangers the business due to a loss of knowledge in its employees. Insurance companies around the world struggle with the fact that no one knows anymore what is encoded into their CICS/IMS insurance application silos. The hardcoded software has become their business. That’s where ACM can return the business departments to be in control.
That is really the revolution that is needed: that the machine learns from humans rather than the machine (BPM or DMS or BRE) telling humans what to do when without any explanation why!