Planuvis – Projektmanagment-Tool with AI

Interview with Alexander Stefka

Hello Mr. Stefka, could you tell us something about your background in project management?

I have worked on and in projects throughout my professional life, albeit in very different areas. On the one hand, I like to have things under control and, on the other, I am a very visual type of person. From both points of view, the various project management methods are a very good fit - they create transparency, make communication clearer and thus ensure project success. I want to maximize this potential in our software by visualizing the essential things as well and clearly as possible and highlighting the most important aspects.

And there is a second important background - my proximity to next level. A seminar I attended at next level a few years ago led to contact and the opportunity to do something together in the area of software development.

The chemistry with Alexander Peschke and his team was an immediate fit and we then worked on some exciting topics and ultimately ended up where we are now - which makes me very happy: in a partnership between next level and uni software plus GmbH, for which I now work. As part of this partnership, we are now constantly working together on exciting topics - most recently increasingly on the topic of AI - and exchanging ideas.

In addition, we have set ourselves the joint goal of positioning Planuvis on the market - as a complement to the Excel solution next project. It's certainly a win-win: thanks to the valuable feedback from the PM professionals at next level, we can optimally align Planuvis accordingly, continuously improve it and shape it together. The wide-ranging state-of-theart knowledge of uni software plus covers all important areas from front-end to back-end development, cloud computing, DevOps/operations and also topics such as data science, AI and machine learning.

What was the specific reason or observation that led you to integrate artificial intelligence into Planuvis?

We have been working on the topic of 'machine learning' for some time. As part of a larger customer project, we developed a system that offers a service technician suggested solutions based on their observations and learns from feedback in dialog.

The results so far have been impressive, so I wanted something comparable for our project management tool. Building the knowledge base from scratch for a software like Planuvis is difficult to implement, so we chose a different approach and made initial tests with existing LLMs. It quickly became clear that this was the right way to go and that we should definitely pursue and develop this approach further.

Which AI functions are currently integrated in Planuvis and what specific advantages do they offer users?

It is currently possible to start with a completely empty project and let KIWI (the name of the AI assistant in Planuvis) 'build' a WBS including all predecessor-successor relations based on a project description. In addition, an existing project can also be subjected to a cross-check.

Specifically, KIWI can currently suggest stakeholders and risk elements in addition to phases or work packages and automatically update the corresponding views. It was important to me that these KIWI suggestions can be identified as such by color and that the user can then either accept or reject them in full or in part.

So what are the benefits for users? Put simply, you can quickly and easily get an initial structure or new perspectives or have the existing ones checked. Thinking in variants is now possible with very little effort.

So everything at the touch of a button - and that's it?

Definitely no. From my perspective, AI is a tool that needs to be used correctly.

While in most cases it is advantageous for the project manager to develop the WBS and all other aspects of the project together with the team at the beginning, instead of pressing the button in a quiet room and presenting everyone with a fait accompli, AI can also be very helpful in this phase. For example, AI can be used to quickly obtain new points of view or to carry out a completeness check on what has already been identified - but not usually right at the start of the preparation or planning process.

On the other hand, if you want or need to have initial suggestions/ideas or several variants quickly in the project identification phase, you can get this in a few seconds using AI, with surprisingly good content quality.

Nevertheless, it will be the human being who has the last word and must have a critical eye - AI is simply a tool and should be understood and used as such.

Where do you see the greatest potential for AI in project management in the future and what specific AI enhancements are planned for Planuvis?

An important focus is on improving the existing functions. Although the results are already very good, we want to become more accurate and more targeted.

We are also currently implementing new ideas - for example in the areas of project content, work package specifications, targets/non-targets or the assessment of risks.

I see a lot of potential in AI-supported review functions, for example with regard to the correct and complete formulation of objectives - we want to move further in the direction of “getting a second opinion”. AIWI could, for example, check the planned durations, costs or resources for an existing project in order to make suggestions for improvement, highlight any contradictions or point out potential problems and deviations at an early stage.

There are basically two broad areas in which AI systems can support us - on the one hand, correct and complete project planning in accordance with current PM standards and, on the other, the project content itself. In terms of content, AI could, for example, provide (early) warnings if key indicators or framework conditions change (construction price indices, market conditions, etc.) or are likely to change.

What were the biggest challenges or perhaps the most surprising AHA effects in the integration of AI?

To summarize, I can say that I was often surprised by the results in two directions. By 'two directions' I mean that, on the one hand, the AI often made surprisingly good suggestions for complex things without much optimization - but on the other hand, it was also difficult to tame for seemingly trivial things.

Teaching the AI to reliably number the phases from 0 upwards and then the work packages as a level below (without mixing the two, in the most creative variants) was almost an impossibility - while the correct linking of the content of a work package to one or more predecessors worked very quickly and plausibly.

The choice of a different language was and is also completely unproblematic - which is important, as we are also offering Planuvis in an English version for the time being. From time to time, however, the AI intersperses the term 'milestone' in a German-language project, so it says 'milestone: planning completed'.

So far, we have been able to eliminate all the strange things by improving the input and avoiding ambiguities or contradictions - which are not always easy for a human to recognize as such. In any case, we want to continuously improve and optimize KIWI, because the feature has already been very well received by customers at this early stage and still has a lot of potential.

Apparently, you get better results if you are unfriendly to the AI assistant or even end up using swear words. Be that as it may, we've managed to get by without swearing and insults so far and I hope it stays that way. We don't want the AI to become jaded or even be put off.

Thank you very much for the interview!