We are all too familiar with the notion that AI (Artificial Intelligence) will be a major game changer in each and every business. In industrial manufacturing robots plow the way. Same development is happening in food production. Self-driving cars are bringing AI out in the open but fundamental changes are taking place mostly hidden inside digital systems in areas like insurance assessments, stock trading, accounting or healthcare, to name a few. Not even the brightest minds in the field of research or any business area are able to tell us the real width and depth of the AI-driven changes that could take place in the near future. The true impact of AI in all its benevolent and malicious capabilities remains largely unclear yet in the years to come.
As a professional who eats and breathes processes and thrives on process mining, I have eagerly asked myself, how would AI change the business process world? I see a relationship between AI and process design, where the latter lies in the very heart of AI being able to work for you and your business.
AI & Current Process Design Methodology
Businesses use quite systematic tools to design and build their business processes nowadays. These process designs are typically complemented with process-specific KPIs. Based on the process designs, companies are interested in understanding their actual process execution and how their as-is processes benchmark against the designed processes. This is where businesses use process mining tools to gain the visibility into their actual operations from the data in their source systems for comparison and checking compliance with their designs.
When we talk about process compliance, be it business-rule based or regulatory, high process compliance designates high quality of operations. It is rather standard that a particular business process has one design and that the design has one defined end-to-end flow like the process flow below:
Current process thinking has taught us that bi-directional is bad, forking is largely to be avoided, and loop-backs are waste. Lean Six Sigma thinking has provided us with plenty of philosophical and methodological reasons why to reduce the variation. Therefore, the process designs do allow some headroom for the unexpected, but all too often they are simplified as series of concatenated process steps or events.
This process design thinking has a fatal flaw in contrast with matured digital systems and current existing capabilities of AI. The process design in its efforts to define ideal process flow falls victim to its oversimplified and subjective assumptions of processes that are not in-line with the new, emerging, learning-capable digital systems and their complex interoperability. Furthermore, typical process designs are not derived from data reflecting the real process, but conceivably detached from it which allow the designs to be developed even in an unrealistic fashion.
As a function of digital systems and their learning capabilities, be it simple prediction capability, machine learning or a full-scale AI, it must allow an increasing number of process variations to be legitimate and thus compliant with your design. We can safely assume that in business process execution, AI out-smarts us humans if only it is allowed and enabled to do so.
Process Design Matrix & Multiple Sets of Rules
What we can say for sure is that once digital platforms become capable of learning, they become central to your systems interoperability. This creates completely new conditions to your process design. It becomes vital that you create conditions where systems learning is inevitable. Sticking to a traditional and simplified process design won’t allow systems to learn and consequently AI to have a positive impact on your daily operations.
It is highly unlikely that when AI plays a significant role in systems interoperability, the number of legitimate process flows will decrease. AI will be able to find new, unexpected but legitimate ways to execute your business processes. In the new, learning capable digital environments, the number of legitimate process variations is instead set to increase. Therefore, there is no other option but to adapt to the change in your process design.
So, what do you need to do then? How should you adapt the very basic building blocks of your process design? I am going to point out a few important topics that you should consider as you prepare for the arrival of the AI:
1. Start with knowing your as-is processes
We must learn to walk before we can run. First, you should learn the reality of your processes in order to be able to transform the process design. If you think you know your as-is processes because you have been working hard in workshops and interviews to map out your processes with stakeholders, think again.
Your true as-is processes are actually stored in the data in your IT systems. We have at QPR witness time and again the amazement on the faces of our customers when they have seen, for the first time, how they business processes actually go as opposed to their understanding of how the processes should go. With the help of QPR ProcessAnalyzer, we have been able to dig out the evidence of how the simplified process design thinking has not at all prepared organizations to see the reality in their numerous digital systems.
2. Think process design as a matrix of compliant variations
Among the 400+ process mining projects that we have completed, we realized that not all variations are bad as there are process deviations that make sense. These legitimate process variations should be incorporated into the process design. As digital systems become more capable in independent learning, the more we will witness process deviations that make sense and should be allowed to happen.
There can be a substantial number of unique end-to-end process flow variations that are legitimate and fulfill your business rules. The larger the processes are and the greater the number of involved digital systems, the greater number of valid variations can occur in a learning capable digital environment. An example of a list of valid process variations:
3. Set multiple sets of business rules to extend process compliance
The fact that process design consists of matrixes of various compliant process variations does not mean that you need to give up business rules. Your business rules are very much needed, they still exist, and they must be enforced in your digital systems. But, setting the rules must follow the flexibility enabled in the process design matrix.
To introduce more flexibility to your rules, you can set a hierarchy of multiple rule-sets, such as:
- Higher-level rules must be executed precisely each and every time.
- It might be sending an invoice, lower-level rules allow systems learning to have access to the needed flexibility.
- How the invoice is delivered to the customer.
Naturally, regulatory compliance needs to be carried out at all times, so set them high in your rules hierarchy. Not all rules are created equal.
4. Your process KPIs remain
In the new context, you might find yourself looking at an unexpected process flow wondering how to give it an appropriate KPI. Setting a KPI to each variation that AI executes might prove an impossible task. But, is it even necessary?
If you look closer to KPIs and their current implementations, you will realize that not much actually needs to be changed. It is a standard that every core business process has a set of KPIs presenting targets or goals. They might measure the number of orders handled overall, the mean duration of order handling process, the number of orders that went first-time-right and those that did not, and so on. These remain valid in the era of AI too, keep your key KPIs as they are because not everything needs to be changed.
5. Set new KPIs for AI learning
What is required now is that we set KPIs that measure the learning capability of our systems. As AI invents new ways of executing business processes, some unique process flows may first become less common and then eventually go extinct. When such a happy occasion presents itself, let the related KPIs go extinct too! As AI is authorized to have a wider reign over business process execution, the number of failures and errors should start to decline too. Here we have a perfect KPI for monitoring your AI implementation success:
Since AI is capable, to a substantial extent, of predicting and preventing failures and errors, the number of areas of deviations should diminish. You are soon able to do okay with less KPIs than you have done before. With AI, you can have a greater number of unique compliant process variations yet less KPIs are needed for measuring them. Once AI reaches a good maturity, the number of KPIs should have gone down too.
It is so very tempting to declare AI as the ultimate and final evolution of process design. However, that would be too much of a speculation since real impacts of AI are yet to be largely materialized. Of course, the future might prove me wrong as AI is executing even the most complex processes infallibly. This small tinkering with the concept of process design revealed that a major shift could be the reality soon. Not only can we gain greater simplicity, reliability and efficiency on the surface and on the level of human responsibilities, the processes and digital systems interoperability related complexities can also be managed to an increasing extent by AI.
There is nothing artificial in that !