1. Peter Schooff
  2. Sherlock Holmes
  3. BPM Discussions
  4. Tuesday, 22 January 2019
  5.  Subscribe via email
Do you think Artificial Intelligence will have a significant impact on process improvement?
Accepted Answer Pending Moderation
Undoubtedly YES! BUT business need to recognise that this requires a back to basics understanding first business processes and then the supporting needs for the users internal and external in a more horizontal flow of work away from the hard coded inflexible silo systems. BPM can be the driver and it is easy and quick to learn and now at last supporting software can deliver directly from business user input. AI opportunities can be readily absorbed into business processes indeed users will be encouraged to see new AI ways to improve both experiences and productivity. RPA being embedded will automatically handle use of legacy data. The big challenge is how are the business executives going to be "enlightened" that there is now a new opportunity to enhance their business without being tied to costly coding with associated risks.
Accepted Answer Pending Moderation
Yes I see big impact. It will help drive consistency, speed of decision and insights to the business thus morphing process change faster. However, there is still a growth curve of acceptance not to mention a state early maturity. Organizations and technology still need to learn processes in this AI operational model and connect the pieces. Can't just plug in yet and go on to something else. The critical path is setting up models correctly, learning and being able to introduce change quickly in the business but in the technology infrastructure. That will require more investment and organizational discipline to direct the outcomes in the near term.
Accepted Answer Pending Moderation
I'm sure there will be, just not sure of when.

AI is still in its infancy in my opinion and for it to be truly useful in process execution and subsequently in process improvement (how can you improve a process you don't know how to execute?) will take some time, I guess. Many, many companies are still strugglig to even take the most basic steps towards Business Process Management and they do have to go through these motions first before you can start thinking about something as complex as deploying AI in CPI (Continuous Process Improvement).

Besdies that, I do believe (similar to RPA) that AI (at the moment) is nothing more than another tool to support BPM. It will not generate any disruptive movements to BPM in the coming 1-2 years if you ask me.

Then again, I like to be proven wrong, because I do believe that AI can make a difference once it grows up to maturity.
BPM is all about mindset first and toolset later....much later
Accepted Answer Pending Moderation
Depends how we define AI.

I don't consider data mining + "simple" predictive analytics

a) for improved run-time decision making,
b) for overlay of suggestions for changes to current process maps,

as AI.

Seems logical to proceed as follows:
1) What do we want to do that cannot be done using current tools?
2) Is AI capable of addressing such issues?
3) What is the ROI?
Accepted Answer Pending Moderation
Absolutely! AI and ML will play an increased role when it comes to the broader discipline of continuously optimizing business processes (as opposed to its mere one-time automation). It will be interesting to observe the underlying prerequisites that in the end will account for such an evolutionary step of the company's path to digital competitiveness.
I think there are already quite a few common denominators most subject matter experts can agree on:

1. Architecture
2. Structured historical data (even though an increased degree of "unstructured" data seems to be now admissible)
3. A mature BI strategy allowing for the KPI to BAM to BI to AI trail of obtaining deep process knowledge.

The classical challenge remains, though: There is an important investment period before any tangible ROI can be achieved. That, at least in my opinion, remains to be the main culprit explaining the lag many otherwise BPM enthusiastic enterprises reveal when it comes to data, BI and especially AI.
NSI Soluciones - ABPMP PTY
Accepted Answer Pending Moderation
As @Karl pointed out, predictive analysis derived from applying statistical methodologies to historical data is machine learning, not AI. And it's machine learning (ML) that we're actually talking about here, by and large.

The problem is that people don't understand the use cases for ML. And even if they do, they don't know which ML algorithms are appropriate for which use cases. And even if they do, they don't understand the predictions that those ML algorithms produce.

So, while ML in theory holds huge potential for digital process automation, in practice that potential is exceedingly hard to realize.
Accepted Answer Pending Moderation
On the face of it, it looks like an easy YES.


The question is about process improvement, i.e. optimization of some metrics as set in an objective function. The focus on ML severely limits the discussion only to those nonlinear optimization techniques that exhibit some form of "unsupervised learning". I get it, it's cool (who doesn't like to talk about antropomorphic machines?), however optimization techniques must be applied according to the optimization problem at hand. Sometimes, an optimization problem may be solved most effectively through a simple least squares linear regression, or by some more sophisticated non-linear techniques, but not ML. In my early career I have seen fascinating results from application of fuzzy logic, cellular automata and, most spectacularly, genetic algorithms, to optimization of objective functions. In my research back then, the neural networks (that are all the rage today) would have accomplished nothing without the mutation shock of genetic algorithms that were super-effective in avoiding local optima.

So, in order for AI to be really impactful on process improvement, the practitioner must have:
1/ a solid understanding and definition of the improvement problem (including multi-dimensionality of possible causes and effects as variables!);
2/ a solid mathematical ground of the suitable optimization approaches to that particular problem;
3/ a large (and high-quality) data set (of the right variables!) on which to train, test and extrapolate the outcomes.

That requires a far more solid understanding of math, econometrics and computer science than BPM consultants are willing to acquire.

As the joke goes: "ML is written in Python, AI is written in Powerpoint" :D
CEO, Co-founder, Profluo
  1. John Morris
  2. 4 weeks ago
  3. #5893
+1 cellular automata! Question - was used in practical situation? Also the ML/AI joke is the best.
yes, we dabbled with them when treating error-correction algorithms (purpose was to minimize predictive error for time series).
we finally settled for dummy-variable econometric models, pruned by genetic algorithms and then the result was fed into MLP's - the stack massively outperformed all known literature for our problem space, especially on directional tests (i.e. ability to predict the right direction of a stock 'tick', which is critical in high-frequency stock-market trading)
Accepted Answer Pending Moderation
As you can see, it is not so clear what we mean by AI. For some people rule based decision making might mean AI... If the situation gets very complex and we have tens or hundreds of rules effecting the situation i.g. healthcare, machine malfunction or complex service definition we human need help in action or decision making. This is kind of first generation AI and certainly growing area. An other area where I see growing interest is "simple" predictive analytics such as process mining helping to understand what really happens in our business processes and maybe helping clarifying the rules.

In the case of machine learning we certainly accept AI term. Here we need some kind heuristic analytics (= building connections between phenomena). Here I have seen some good cases in the area of marketing and how to effect people buying decisions. The challenge here is how teach correct answer to the computer... In the case of games like Chess this is fairly easy because of fixed rules and an other good example is face recognition because of a lot of material and simple right solution. In most cases in business this is not so easy, because the situations vary so much. I would say that this is very slow and it takes time to materialize.

br. Kai
  • Page :
  • 1

There are no replies made for this post yet.
However, you are not allowed to reply to this post.