In this blog entry I will share my story about a very challenging business rules project, how mathematics helped me to understand complex rule sets and how mathematical logic helped me to get deeper insight. In fact this was a joint work together with Jann Müller from SAP UK with whom I wrote a joint paper that was not accepted in the KES-IDT-15 conference about Intelligent Decision Technologies. So this is a story of research and collaboration in a very important field since most processes in ERP can be improved by automated decision management as well as decision support. At the end I discuss how to bridge the gap between academia and the SAP ecosystem.
A rules management project
End of 2013 I had a very challenging project in the area of health insurance: we used the ABAP based BRFplus framework to classify scanned accident reports for further process automation. In fact we did the following:
- Questionnaires for accidents reports have been standardized so that they could be scanned easier.
- We created an ABAP application where those archived questionnaires together with scanned data are processed. There is also workflow integration for clarification cases but I will come back to this later.
- We used BRFplus as rules technology to for process automation: the scanned data are checked by a rule system. If the facts are “obvious” the insurance claim solution takes over and finishes the process by using its own rule systems.
- Here we found a sophisticated solution since and this is how it goes: we submitted the information how our rule system thinks the insurance claim should be processed. We calculated tasks (f.e. creation of correspondence or workflows) for the claims system. This leads to further process automation.
We have been quite successful and even the first version of the rule system was in 2/3 of all accident reports as good as an end user. Of course we spent further effort to make the rule system even better by accessing data of the SAP Claims backend system and got more information about the accident from claims bundles.
Ambiguity strikes back
Unfortunately the data are not always “obvious” since many accident reports are ambiguous and so the rule systems created clarification cases which have to be controlled manually. One reason for this was that accident reports are far more complex than purchase orders for example since accident reports tell a story about the real life. These stories can be quite complex like the following: “I was working in the house of my mother in law and fell over her stupid dog down the stairs. I got a small injury, had to go to a physician and came too late to work.” There are many of those stories and some parts of it can be verified since the SAP backend already has information about the insurance claim.
The classification of the accident type is very important for the insurance company because in some cases there is a chance for legal recourse. In this project I learned that our IT systems are very good to process standardized data but today more and data comes from un- or semi structured domains. In the case of classification of accident reports 61% of all accidents reports have been classified as “ambiguous”.
It would be interesting to know whether this effect occurs in other domains that deal with real world data too. In fact this leads also to very interesting philosophical questions. When we try to evaluate an accident report we ask for its meaning. Meaning can be seen as a conceptualization in a cognitive model. There are many reasons why this conceptualization is not possible: we don’t have all information and there is an incomplete mapping of the real world transferred to the conceptual model. But the cognitive model could suffer from the fact that it doesn’t allow a unique conceptualization. In our case both is the case and the result is ambiguity.
Our Approach dealing with ambiguity
In above mentioned project some solutions have been found using prioritization of accident types. I wanted to understand this in more detail. From a mathematical perspective prioritization means to create partial or linear orders. So my question was: what kind of partial order was already coded in the rule set specification? Can this be visualized and perhaps improved? In the case it can’t be improved by the existing rule system it would be useful understand how ambiguity can occur. This gives many opportunities for improvement of existing questionnaires or additional checks if this questionnaire comes from electronic channel like web applications.
I started to model the existing framework using description logics using standards like OWL. This allows classifying a certain incident characterized using facts from the questionnaire as concept in terms of an ontology. Of course there can be more than one than one possible classification for an accident. Then Jann introduced me into argumentation theory which be used in the following way: Some classifications can attack others. Using this method it allows to modes aspects like agreement, disagreement, skeptical argumentation and consensus between different points of views. And this is what we have been working on.
What I have learned during this small research project
Although I am member in two scientific organizations and I reading scientific books and scientific literature from time to time I learned that I missed much of the most recent work in artificial intelligence especially non-monotonic logic. Without help of a researcher our small contribution wouldn’t have been possible. So I had to read and to learn a lot – of course in my free time.
So this is the reason why the whole process took some time compared to my activities at work, where projects have to be completed very quickly. But in general research is a slow process since the results must be correct and convincing. The usual work in software industry is different: we have requirements analysis, apply best practices and from time to time we do edge development.
I’m really looking forward to the conference about intelligent decision technologies this summer and I’m sure that I will get much inspiration and I will blog about it.
Why Research is necessary
IMHO professionals need at least a vague idea what people who do research in applied computer science and related areas are doing. Without research we won’t be able to keep track to latest advances in science and don’t understand the concepts young employees learned at university.
This is also problematic since ERP has to become smarter since the future directions in IT will be advanced analytics, IoT, rule based systems and so on. IMHO IT professionals will need other skills in the future. Reading scientific publications and doing small research projects is a perfect opportunity to fresh up his knowledge.
Bridging the Gap between Academia and SAP Ecosystem
In the past I supervised some students who worked in my company and wrote their diploma thesis about their projects. For me this was a perfect opportunity to get deeper insight in a certain topic. I also recommend to supervise students since this can establish a close connection to the company and hopefully they will become employees in the future.
Supervising students and doing own research (together with a highly skilled Ph.D. student) is something different. In this case the challenge is that it is hard to find the right person since the research topic is usually highly specialized.
I believe that Ph.D. students can benefit from working together with people in the IT industry since they get contact to the industry and can also test whether the research topic they are working on has applications in the real world. Moreover they can get inspiration for their own research.
The problem is that it is hard to bridge the gap between academia and SAP Ecosystem. One reason is that SAP seemed to restructure their research organization into SAP Labs with a slightly different focus. It seems to me that there is still more classical research in HPI, SAP Innovation Center and of course in the HANA team of Franz Färber.
I think that perhaps the University Alliance Program may help. I would appreciate if there would be the possibility to connect senior and Ph.D. students with the SAP ecosystem. Maybe SAP University Alliance already has already some concepts and experts like Ann Rosenberg can tell more about it. I would be very interested to discuss this.