The right team and its organization is THE critical success factor of any software project - therefore, significant resources are invested into this area. Nevertheless, some IT managers would prefer to deal directly with the actual project content. A research team at the Hasso Plattner Institute (HPI) in Potsdam is already working intensively on a solution that is of interest all IT experts: The AI Assistant, which is fed with a large amount of relevant data, plans entire IT projects for managers and presents optimized team structures, schedules and expenditure forecasts. With a current research project, the HPI comes a significant step closer to its vision. We at Seerene are happy to help with this and are already able to give a first insight into this project.
The right team and its organization is THE critical success factor of any software project - therefore, significant resources are invested into this area. Nevertheless, some IT managers would prefer to deal directly with the actual project content. A research team at the Hasso Plattner Institute (HPI) in Potsdam is already working intensively on a solution that is of interest all IT experts: The AI Assistant, which is fed with a large amount of relevant data, plans entire IT projects for managers and presents optimized team structures, schedules and expenditure forecasts. With a current research project, the HPI comes a significant step closer to its vision. We at Seerene are happy to help with this and are already able to give a first insight into this project.
Every practitioner of modern software production is painfully aware that managing a larger team of developers involves not only technical challenges but also very demanding logistical difficulties. For example, the question of which developers best master certain tasks, how teams should be configured, or which efforts and processes determine the required project time. This elementary information is not easy to bundle even in a clearly structured and heterogeneous team. With teams of more than 30 or even hundreds of differently specialized developers, however, they already entail an unreasonably large amount of effort. The efficiency losses that can be caused by poorly calculated efforts, unclear technical challenges, mismatched specialized knowledge of individual developers or non-transparent time availability quickly add up to dangerous resource traps. Up to now, IT decision-makers have had few systematic tools at their disposal to assist them with these responsibilities.
Inefficiency and the always latent danger of "complete logistical failure" of complex software projects are well known and described. Many IT experts have therefore chalked them up as a given law of nature. "Not so!" says the Software Analytics team at HPI in Potsdam. One might easily think that the team is fighting a battle against windmills if the experienced software developers Daniel Limberger and Willy Scheibel, and mathematician Daniel Atzberger were not researching under ideal conditions under the direction of Prof. Jürgen Döllner, who holds the Professorship of Computer Graphics Systems. In addition, the team has full access to Seerene's powerful software analytics platform, which otherwise only corporations utilize in their quest to remain in control of their gigantic software systems. As such, the project is equipped with the best available means for the analysis and structured aggregation of complete software projects and indeed all associated data traces.
The research project has a very clear practical objective: to provide managers of large IT units with a software tool that simplifies the composition of teams and the planning of project plans. Proposals based on the best available data will shrink the planning effort from days to a few seconds, and the quality will be significantly improved by taking many dimensions into account.
The approach chosen in principle for the research project uses Natural Language Processing as an important pillar. What is astonishing for the layman is that here not only messages, work tickets, and any other digital human communication, but also the programming code is regarded as a linguistic utterance and both types of communication are included in the analysis in a certain respect with equal value. Thus, the program first analyzes all data and data traces of past projects up to the present without the bias of the human/machine distinction. On this basis, it creates clusters for the processed tasks on the one hand and profiles for all developers involved on the other. These continuously updated profiles can also be used to create ideal team/schedule configurations for upcoming projects.
So far, so good. But when you learn more about the approach, you quickly realize that the algorithms' approach to compiling performance clusters or developer profiles goes well beyond the ability and understanding of humans. For example, a developer is not simply given a profile whose parameters can be easily described. Thus, the values determined are not purely about common characteristics such as perhaps programming language skills, speed, endurance and complexity management that people would apply. They are parameters that are summarized by the AI from a wide variety of dimensions. Their relevance or predictive power aims to be significantly superior to conventional evaluation parameters in terms of speed and stability. How can one imagine this?
A perfect match between the developer and the task package to be solved is created when the analysis applies criteria, each of which contains a large number of different parameters in a situationally ideal weighting. Things like the depth of knowledge or the experience in a programming language still play an important role, but no sole or isolable role. The property mixes, in some proportion with many others, which together give the maximum predictive power. Only through this abstraction can matching make use of the maximum possibilities of data analysis.
While the first AI evaluations of real teams and projects with the help of the Seerene Digital Boardroom already inspire great confidence, the research group is currently working on putting the results of their analysis into a form that can be interpreted and visualized again for the target group of IT decision-makers. Decision-makers must ultimately be able to understand their "superhumanly analytical" digital assistant and its suggestions. Only then can AI lighten the load and optimize practical decisions. "AI uses a mathematical formalism that we cannot interpret directly and we must first understand it ourselves in order to be able to translate it into practical human decision logic afterwards. In doing so, we can initially interpret the machine-generated suggestions only with the utmost caution" says Atzberger. However, the research group's initial tests make them extremely confident that AI will generate an applicable solution here. "For example, we have been able to determine very specific characteristics of individual developers amazingly well through our system, which have been confirmed to us by the developers themselves and their colleagues. In addition, we have also already been able to uncover interesting correlations that the investigated developer teams themselves had never noticed before. This is the kind of added value we ultimately want to deliver methodically," says Daniel Atzberger. Indeed, it has already been possible to harmonize some human experiences with the AI results and strengthen confidence in AI. Indeed, the first procedures have already been tested and evaluated in an industrial environment. Now, the scalability of the procedures needs to be addressed. Next year, an important milestone will follow: then the research team would like to present the first prototype for the data-driven creation of project plans.
The HPI research project is extremely practice-oriented – after all, it aims to alleviate one of the IT industry's most significant pain points, which ties up incredible resources. Because according to Prof. Jürgen Döllner of the Hasso Plattner Institute, behind the pain point lies a fundamental weakness in computer science. "In the IT field, we apply "engineering-like" rationality mostly to our code but surprisingly rarely to our own working methods and production processes. That's why I'm delighted that Daniel's project works so closely with the mission of Digital Engineering, which is ultimately largely the initiative of our colleague and benefactor Prof. Hasso Plattner. With the term Digital Engineering we express our ambition to let practical software production mature in every respect into a real engineering science, i.e. one that reflects its own approach," says Döllner. The research group and HPI will share their results in the scientific community and look forward to discussion, expert input and cooperation requests.
Authors: Oliver Viel, Dr. Uta Morgenstern
www.seerene.com The Digital Boardroom provides IT decision-makers with the information needed to actively drive their organization towards software excellence. The analytics technology extracts and fuses the data traces left by all the tools in your software development infrastructure and derives insights of high strategic and operational relevance for both managers as well as development experts.
Translated in March 2021 by Brandon Lewis.
