Travelogue #01 of the Humboldt Group: Read now ➜

Travel report #01

Join us on the Humboldt Group’s first adventurous journey and learn:

What we’re up to at Comma Soft out of curiosity.
How we got in touch with Circular Economy on the island of Oeconesia.
How Masinope, island kingdom of 1,000 rivers, gave us inspiration for Machine Learning Operations.
Who our crew members are and what makes them tick.
Why Alexander von Humboldt inspires us so much that we want to follow in his footsteps.

Enjoy reading and exploring!

If you’re interested in learning more about us and our innovation topics beyond our first trip report, you’ll find more exciting information on this page: from best practices to sources and notes to ideas for implementing them in your own company. Let us inspire you!

We also invite you to get in touch with us: For questions, suggestions, to exchange ideas – write to us and we will get back to you!

Contact the Humboldt Group

Circular Economy

Our visit to Oeconesia has gripped and inspired you? Would you like to read more about the topic?

Then our innovation lab technician Dr. Michael von Papen recommends the following texts.

Start with an initial ML use case

Are you planning to implement a specific use case with machine learning? The adjacent canvas helps you to clarify the most important questions for your project in advance. A template is available for download here.

Download Canvas

From Use Case to Transformation

If you want to use machine learning to make your company data-driven throughout, you should not stop at individual use cases. Use cases are tactical measures, comparable to the visible tip of the iceberg. Below that lies the strategy on the basis of which these cases are implemented and below that, in turn, the actual transformation that is triggered by MLOps principles.

Start small, but start!

It makes sense to start with the first ML use cases and gain experience. Non-critical use cases and short-term projects quickly create quick wins, while keeping the effort manageable. However, also realize that the focus is clearly on learning. You may not see a big business impact yet, but understand the value ML can generate.

In the next step, you will then dive deeper and develop a strategy aligned with your business goals, based on which you will implement further ML use cases. From then on, it also makes sense to automate initial processes needed for operationalization to save resources.

If you achieve your strategic goals with ML, you can eventually go as far as using Machine Learning to drive innovation. It’s here that the real transformation to a data-driven enterprise happens. At the same time, it is inevitable that you use MLOps as a concept to implement Machine Learning in an efficient and maintainable way now at the latest.

MLOps Bootcamp

Do you want to use Machine Learning Operations as a concept in your company?

In the MLOps Bootcamp, you will learn the conceptual basics and learn how to implement reliable and maintainable ML applications in your company using practical code examples.

Register now

More interesting resources on MLOps:

Overview study by the University of Leipzig on frequently occurring antipatterns with a focus on organizational challenges. Includes both the most frequently observed problems and their possible reasons and solution ideas. The contents addressed in the paper could be of particular interest to managers or team leaders.

Community Blog & Community Events: blog articles and papers on MLOps. For technically interested reader:s Contains many more resources to learn about the practical, especially the technical, side of MLOps.

Large interview study with 18 ML Engineers, which includes many quotes and presents the typical challenges of implementing and commissioning ML applications.

Collection of many different articles on MLOps, including content on motivation and processes, less tailored to specific tools. Often prepared with descriptive graphics.
https://mlops.community/blog/ and https://home.mlops.community/public/events
Blog articles and talks on MLOps for technically interested readers. Contains many more resources to deal with the practical, especially the technical, side of MLOps.

https://arxiv.org/abs/2209.09125
Large interview study with 18 ML Engineers, which includes many quotes and presents the typical challenges of implementing and commissioning ML applications.

Driving innovation

You’ve already read about the innovation formats that drive us at Comma Soft. If you are also considering such approaches for your company, then take a look here: We have collected guidelines, tips & possible actions of the three formats for you.

Task force

Task forces are a suitable means of evaluating innovative topics quickly and without much overhead for your company. The main aim is to use the personal motivation and interest of the employees for a topic. By dispensing with very rigid administrative and organizational structures, those involved are free in their thinking, which can also give rise to completely new ideas.

Task forces are the tool of choice for staying on the ball, especially when it comes to new topics or rapidly developing trends. They enable a rapid response to new developments.

If a topic has obvious relevance, but cannot yet be clearly located in a business unit (organizationally) or if it is likely to be relevant across business units. Here, a task force can give the topic an initial home so that it can get off to a fast start.

If people in the company are burning for a topic apart from their daily tasks. For our ChatGPT task force, there are colleagues who are very fascinated by the topic and are also privately involved. Since they have long been considered experts in Natural Language Processing (NLP) and Large Language Models (LLMs) in the company, they are what we call “natural authorities” for the topic.
If a topic evolves from existing processes in the company and/or can be clearly assigned to an organizational unit, the topic should also be dealt with there. This prevents ambiguities in responsibility and ensures clean (re)integration into existing processes.

Very complex topics that may also require a large amount of development or prototyping are not suitable for a task force. Here, the reasonable time, personnel and organizational framework is usually blown. In these cases, a classic project including project management should be set up. An initial evaluation by a task force can be useful, but should be carefully weighed against spikes (feasibility, implementation options, solution models) in the project.
For those involved, task forces are a task outside their usual range of activities. It is therefore particularly important to communicate and value the relevance of taskforce work. For us, this means, for example, that this work is budgeted and prioritized internally. This gives the taskforce the necessary freedom, detached from day-to-day business.

Even though task forces are very agile and explorative, there should be clearly defined goals and a time/budget framework. This prevents the taskforce from getting lost in unplanned side issues, and these goals also represent the strategic control of the taskforce.

In any case, participation in a task force should be voluntary and its foundation should also be bottom-up. The participants should bring along a lot of motivation to take on the topic, as this is the only way to create the necessary drive to deal with “taskforce-worthy”, complex and dynamic topics.

Communication into and out of the task force is very relevant. This requires suitable channels through which the rest of the company is kept up to date on developments, as this involves a wider circle of interested parties. At the same time, it must be possible to communicate questions or ideas to the task force. We generally use team channels here, and the task forces also ask very specific questions to the company (e.g., in the form of short questionnaires) in order to obtain information or possibly find fellow campaigners for specific sub-aspects. In addition, these channels are used to try out the findings of the task force in real applications at an early stage and to feed back the results.
The task force should consist of a small but heterogeneous core of personnel. In our case, teams of three to five people have proven successful, as they are large enough to handle the task and at the same time do not draw too many resources away from ongoing projects.

Additional experts can be brought in at any time for special aspects. In this way, the task force remains flexible and there is clear personnel responsibility. Ideally, the members of the taskforce come from different areas and complement each other ideally for the respective project. For our ChatGPT taskforce this was an NLP expert (concrete evaluation and expertise), an IT consultant (technical integration in IT landscapes) and a colleague from the business side (topic placement in customer projects).

Task forces are set up quickly and unbureaucratically with a (small) budget, independent of the other organizational structure, and start almost immediately. Especially with very dynamic topics, time is the most critical factor.

The task force decides independently when and how it meets, whom it consults and how it uses its budget. Too many bureaucratic requirements would unnecessarily restrict this process.

Ideally, an infrastructure already exists, e.g. Microsoft Teams, in which an exchange channel for the respective task force is created without any special effort. At the same time, the entire company is made aware of the taskforce and this channel. Regular updates are part of the taskforce’s duties.

Basic organizational issues around taskforces are clarified in the company. These include: Who releases the taskforce, how is budgeting and prioritization done, how should taskforces communicate. This prevents delays and ambiguities regarding responsibilities and ensures that the task force can start quickly. Other processes can then be defined for an expansion of the scope.

Stand-ups

Topic stand-ups are an excellent format for regularly exchanging ideas on relevant, evolving topics within an organization. They help to share knowledge, to get to know contact persons for specific questions and to develop a uniform understanding and internal best practices. In addition, such a format also strengthens the sense of togetherness, as cross-functional people come together to present topics. Those who have little contact with each other in their day-to-day work can share thoughts, questions and feedback here. For new colleagues, a stand-up also offers a good opportunity to get to know the company and to network. On the one hand, this broad exchange can give rise to innovative ideas; on the other hand, it also ensures a certain consistency and prevents the wheel from being reinvented over and over again.

There is a broad but clearly defined field of topics that is relevant to more than a handful and also has a certain dynamic. For example, we have established the Data Science Stand-up, in which we present tools, trends, methods and approaches, but also discuss topics related to Data Science. Since this is a very dynamic field, there is always something new to report, which is relevant for many of our colleagues.

It is a topic area to which everyone can contribute, as the format does not rely on an audience, but on the active participation of everyone present.

The colleagues are motivated and interested to share their personal experiences and also to critically deal with them collectively.

A topic area should be anchored in the organization in such a way that a common understanding and a similar level of knowledge emerge. This helps us, for example, not to dwell forever on our own “favorite tool”, but to be inspired by the good experiences of others. This increases our knowledge and makes it easier for us to help out in other projects.
Even though a stand-up offers the opportunity to explicitly discuss a topic, this should be separate from a pure “workshop format”. In a workshop format, concrete, current challenges are presented and solutions are sought in the group.

If the focus is exclusively on “best practices” or “lessons learned” with a clear project reference in each case, classic project retros are more suitable. However, these can then naturally give rise to a topic for a stand-up.
We usually have around 20-30 people at a stand-up. A larger group is not suitable, because the size of the group means that not everyone can be included in the discussions. In smaller groups, on the other hand, the variety of different perspectives is smaller.

An average of about 1.5 hours per person per month should be planned for participation, and 0.5 hours per week for the moderators/organization team. In addition, the speakers need time to prepare their presentations. This free space should be granted by the managers and made available as working time.

The willingness to actively participate (present and discuss) must be present. In addition, it requires the willingness to organize and moderate the format.

It is a good idea to define some guiding questions that will guide the format. These can be, for example: What current trends are we observing? What experiences has a colleague had with a specific tool or method? What are the advantages of method A over method B? What are best practices in projects of type X or what turns out to be particularly difficult?
From our point of view, it has proven successful to have a moderation duo design the stand-up. Ideally from different departments in order to cover different aspects. The moderation duo proactively takes care of the speakers and discussion topics and leads through the meeting. Ideally, one moderator should be replaced from time to time in order to distribute the effort fairly and to give everyone the opportunity to try out the format.

A stand-up needs a fixed date, e.g. monthly, and should last about one to one and a half hours. This interval provides a regular exchange without becoming too busy. With two topics per appointment, the presentations do not become too long and there is enough room for discussion.

Each appointment needs a clear agenda in advance. Even for pure discussion topics, 2-3 people should be selected in advance who can report something on the topic to get the discussion going. It is also possible to discuss 2 different topics per appointment. Thus, we usually have a technical and a methodological part in the Data Science stand-up.

It is recommended to have and manage a topic backlog. Here, the organizing team collects and prioritizes interesting topics including potential speakers. This backlog should be available to everyone, so that all participants can also contribute a topic request. In particular, the Orga team always keeps its eyes and ears open in order to continuously find further topics.

At the end of the year there is a review of the topics of the past months. At this meeting, initial topics are also collected together with all participants to be presented next year. The Orga team should then proactively approach potential references at an early stage in order to set the agenda for the next dates.

The dates, at least the presentation portions, are recorded and documented in a wiki. After our stand-ups, we also provide presentation slides and, if relevant, code and data in addition to the recording. However, we usually refrain from recording pure discussion formats, as this can be an obstacle for some participants.

R&D projects

R&D projects serve to develop prototypes for solutions or products. In addition to the development of new products, which ideally refinance the research expenditures invested in advance, research projects can also bring other advantages. For example, active research and development can strengthen a company’s own position on the market or also serve as a door opener for new markets. Research cooperations are a good way to establish networks and expand one’s own knowledge. And last but not least, R&D also strengthens the company’s public image and is a relevant factor in attracting new talent.

If you deliberately want to open up a new field, be it a special topic, a market or a customer. The goal may be, for example, to develop a prototype that can later be put into operation.

If you want to position yourself as an innovation driver. The aim here is to help shape a developing topic and also to see your own contribution as a door opener into the respective community or to sustainably consolidate your own position there. For example, we used FASTGenomics as a gateway into the pharma & life science world.

Research and development can also be aimed at improving our own processes or increasing their efficiency. On the one hand, this can be done with direct reference to existing processes or as a project-independent development in order to be able to rethink processes or products.

You have identified a relevant topic (for example within a task force) and would like to contribute your expertise in a new area (cross-industry). For example, with FASTGenomics in the area of single cell analysis, many challenges were solved by already known approaches from data science. This technology transfer is then also an innovation in the respective area (innovation does not always have to be invention).
In research and development, ROI (return on investment) often only becomes apparent very late. If the risk is currently not acceptable, it may be worthwhile to look for ready-made solutions or partners on the market. In this case, a solution based on the task force can also be used for a kind of preliminary study.

If the necessary solution competence is not yet available or only very basic in the company for the respective issue. In this case, you should check whether it would not be better to link up with existing initiatives or to look for suitable partners and, if necessary, service providers.

If the market is already being courted very actively, you should evaluate your own entry very carefully. Under certain circumstances, it may make sense to enter into research collaborations and work with rather than against each other. In addition, sub-aspects can be developed in a targeted manner, for example, which may also have a certain value on their own.
The be-all and end-all is a very precise understanding of the requirements, for example of the customers or users. This also includes determining what the solution is worth to the users in the end. To do this, the entire requirements/market analysis is constantly updated throughout the development process.

If you are convinced that you can offer something for which a market has yet to be created, the sales department should find early adopters and ambassadors for the solution as early as possible. Here, however, the significantly increased risk of a later ROI must be clearly taken into account.

R&D projects need – even more than other solution development – strong support from management and the entire company. Setbacks are “normal” and even helpful with new topics. But they must also be treated as such (keyword “error culture”). Without strong personalities at the helm of such a project, those involved usually run out of steam at some point.
Dedicated R&D departments are one way of bundling research and development. In medium-sized companies in particular, however, it can pay off to assemble R&D project teams from all relevant areas and have them work on a focused topic. In any case, R&D projects should draw heavily on existing assets within the company and not reinvent the wheel unnecessarily.

Clear milestones and also possible jumping-off points are agreed. This reduces the risk of getting carried away by small successes without really getting closer to the actual goal. This also includes regular alignment with the market and its evaluation, or the evaluation of user requirements. These can change very quickly in some cases.

It makes sense to join forces with potential users for your solution at a very early stage and to involve them strongly in the development. This can also include experts from within the company, but should not be limited to this.

In order to increase the potential for success, regular evaluation of subcomponents can also be carried out in parallel with the actual project. Often, artifacts are created “quite incidentally” which, in themselves, represent a great deal of added value, but which may not actually be in the intended target area. This evaluation also helps to exploit synergies.

More interesting sources on innovations and innovation formats:

The Innovator’s Dilemma, book by Clayton Christensen
This book deals with the question of when one can, should and must invest in innovation, even though the current business model is still going well, in order not to be overtaken or even replaced by small, unimpressive companies. It clearly showed us that what is needed above all are small, courageous units in a company that can work on new issues with a strong culture of error, so that the company does not one day become too sluggish and slow.

“Reinventing Organizations visually: an illustrated guide to meaningful forms of collaboration”, Frederic Laloux.
A very nice alternative presentation of the original 360 page management book on the topic of meaning-making forms of organization. One of the ideas in this book also feeds into the way we drive innovation in our task forces and stand-ups: bottom-up rather than top-down.

“Business Model Generation: a handbook for visionaries, game changers and challengers,” Alexander Osterwalder and Yves Pigneur.
This visually appealing handbook helped us a lot to get into the appropriate thinking patterns and to think through and develop different business models for our FASTGenomics platform.
“Change and Corporate Culture”, lecture by Angelika Bachmann as part of the Petersberg Talks 2022.
This lecture by Angelika Bachmann as part of the Petersberg Talks 2022 not only highlights the topic of psychological safety, but also discusses what constitutes a healthy and innovative corporate culture and how to promote it. Absolutely worth seeing!

“Exclusivity is yesterday’s news”, opinion of Prof. Maren Urner – Professor of Media Psychology and Neuroscientist.
A successful dualism from Modern Work as well as Neuroscience on the topic of exclusivity of information – why sharing as a survival strategy is better than keeping everything to yourself. It reflects for me the central “why” of a stand-up or taskforce.
“Wissen durch Teilung”, article in DIE ZEIT on BarCamps:
Exciting, another format in which knowledge can be shared: A BarCamp, a so-called “unconference”, whose content and schedule are not determined in advance, but are developed by the participants themselves at the beginning of the meeting. There are no listeners, only participants.

“ChatGPT & Co under the magnifying glass”, KnowledgeHub on Generative AI
Absolutely worth reading: this is a collection of insights and categorizations that we have currently collected in the ChatGPT taskforce. In contrast to a white paper by a few authors, this potpourri brings together a wide variety of perspectives.

The journey continues!

We have already set sail again and continue to explore new worlds. Be curious about what awaits you in our next voyage report!

Would you like to receive another copy of the current report or would you like to register yourself and your colleagues for the next issue? Write to us!

Request travel report

On this page: