In podcast episode 29, Madeleine Mickeleit talks to Hendrik Nieweg and Martin Dimmler from Device Insight. The two IoT use cases this time revolve around the implementation of warning and escalation concepts for coffee vending machines from Costa Coffee and the AI-supported optimization of production processes at the company JUMO.
Podcast Episode Summary
As a full-service provider for demanding IoT and IIoT projects, Device Insight (DI) has already realized many exciting use cases together with partners. In this podcast episode, DI’s Hendrik Nieweg and Martin Dimmler showcase two of them.
The first use case revolves around a topic that is probably close to all of our hearts: Coffee! Together with the international coffee house chain Costa Coffee, DI has implemented a “pay-per-cup” business model. Costa Coffee’s idea: in return for a share of the profits per coffee sold, high-quality coffee vending machines were made available free of charge to gas stations, airports etc. Device Insight’s technology not only enables them to reliably track the number of coffees sold, but also to minimize downtime through early warning and escalation concepts that prevent profit loss.
In the second use case, Martin and Hendrik report on their collaboration with the AI specialist Sentian. At JUMO, a manufacturer of measurement and sensor technology, they succeeded in improving the production process with the help of AI. Thus, thanks to sensor technology and analysis of the raw material, adjustments can already be made in the production process. The result: More sensors can be produced at the highest quality level.
At the end of the episode, we show how relevant partner ecosystem collaboration is to solving customer problems.
Device Insight GmbH, founded in 2003 and based in Munich, is an IoT specialist that supports companies in digitization in the environment of Internet of Things, Industry 4.0 and Artificial Intelligence. Based on a flexible IoT framework, Device Insight combines ready-made IoT building blocks and microservices with individual applications for customized IoT services.
Hendrik, what is your role at Device Insight?
I manage the Solutions division at DI. On the one hand, this involves presales consulting – consulting is also what I mainly do in the company. On the other hand, the members of my team also provide Aftersales Support. My colleagues and I form a bit of a bracket around our projects, i.e. project initiation on the one hand, but then also, once it’s developed, operation and support of the solutions.
I have been with the company for 13 years and by background I am an industrial engineer specializing in mechanical engineering. So I’m not a classic software engineer, but I had already drifted strongly in the direction of industrial IT during my studies, so it was a good fit that I ended up working for an IT company that has a strong focus on Industry 4.0, mechanical engineering, etc.
Martin, what is your background and role at Device Insight?
I have now been with Device Insight for seven years. I have a degree in business informatics, and I joined DI directly after completing my master’s degree. That is to say, I really only know IoT, my entire career has been tied to the topic of IoT, if you will. I have had stints in sales and project management at Device Insight and am now responsible for developing our portfolio of services in the cloud and AI space. But I’m also still involved a bit in the project initiation and presales process. That means I do workshops with our customers and work out new business models, functional requirements, solution designs and so on.
What is your core business and what is your vision for IoT?
We have been on the market for over 17 years now and also see ourselves as an IT pioneer. So as one of the companies that was very early in this market. They just didn’t call it by that name in the old days. When we started, the whole thing was still called M&M, Machine to Machine Communication.
DI has also been explicitly launched as a company that takes care of networking machines and plants with central server systems. We started at the Munich site and still have our epicenter here. In the meantime, we have grown to around 100 employees, and I was employee number 7 at the time. We come from the connected products environment, so we started by connecting machines and systems to these central systems. We have done projects like smart home heating, where we also still have a large installed bassis and we have connected forklifts.
Due to our focus on mechanical and plant engineering, we then also built up a relatively strong footprint in the area of Industrial IoT. It was precisely with our current shareholder KUKA that we first started out as a technology supplier years ago and then intensified our partnership more and more. Since March 2019, we are now also the full subsidiary and the technology supplier for KUKA and also the development center for Industrie 4.0 applications.
That is our core business. We are building IoT services from the edge to the cloud and are also relatively heavily involved in standardization initiatives, working in the OPC Foundation, driving issues around the Open Industry Alliance and are also very active in other bodies.
How do you have to understand this holistically in the market? What does your offering look like?
The one key point is that it goes from the edge to the cloud. This means we can map across all levels and ultimately provide a turnkey IoT solution for our customers.
The other point is that we see ourselves as a 360-degree solution provider: This means that we also offer everything that belongs to the area in addition to the technology. We start with business case analysis and requirements engineering and then derive future-proof IoT solution architecture. Of course, we also go into implementation and take care of the operation of the solution.
The whole thing, of course, fast, agile and scalable. We are focusing on the hyperscalers such as Microsoft Azure, which we believe have found a perfect level of abstraction in terms of what work I can make easier for a solution provider, but at the same time how much flexibility I can still leave to them. Tailor-made solutions are important to us. True to: “No size fits all”. We always look at what the customer needs, what their individual business processes and the logic behind them look like, and then offer a tailor-made solution. This allows us to deliver the projects faster, but also more robust and stable.
What challenges do you see your customers currently facing in terms of digitization?
Digitization is already a very broad term. The expectations attached to this term are really very different. For a while, there was a strong trend to create positions such as the Chief Digitization Officer (CDO), who would deal with internal and external digitization. Many companies have discovered that there is a lot of work to be done. Perhaps a little disillusionment has set in at one point or another.
You have to look at what are the goals that I want to achieve. Simply driving digitization for the sake of digitization is not expedient. You have to look at what your focus is, what you’re trying to do and what you want to achieve. Do I want to make my machines and systems smarter, or do I want to leverage other potential in the area of Industrial IoT, such as increasing efficiency, reducing costs or generating new sales? You have to ask yourselfall these questions and think about what business models are behind this. We try to discuss these issues at eye level with the customer.
We don’t think that this technology-driven approach of “we’ll come up with the big toolbox of digitization products and then we’ll get started” is effective. You really have to sit down, do process mapping and figure out where the Pains and Gains are. For a long time, IoT in particular was a playground that everyone tried out and a lot of proofs of concept were made. And I think you can say today that it works. But disillusionment often comes when it comes to determining the added value. Often the cost-benefit ratio is not all that ideal. So our advice is always to look in and highlight and prove the proof of value. This is the challenge of today, especially in more difficult economic times.
Do you have a use case with you from actual practice where people understand how you develop the product with the customer and what the added value of Industrial IoT is for the customer?
I’d like to talk about our work with Costa Coffee, one of the largest coffee house chains in the world. They now belong to the Coca Cola Group. There is the challenge that such a business model is a bit harder to scale. That means I first have relatively high investment costs if I want to open a new coffee shop somewhere: I have to find a property, I have to build it, set it up, find staff, etc.
That’s why they took a closer look at the coffee vending machine business back then. Coffee can also be sold through a vending machine, which is much faster to set up, though it often doesn’t taste that great either. Costa Coffee wanted to address this challenge. They said they are building this coffee bar as a vending machine, but replicating the overall great experience at a coffee bar. When I walk by, flavors are sprayed on and it smells delicious, they have an elaborate design by Pininfarina, an Italian design studio that also did a lot of body designs for Ferrari models, there are fresh beans in there, fresh milk that makes a really good cappuccino foam and I can take syrup shots to go with it. There’s also a large touchscreen with great menu navigation, so that really creates a special coffee experience. While I wait for my coffee, music is also played with some background noise from a bar.
You can find these café bars at gas stations, airports, or on college campuses, for example. Worldwide, there are a five-digit number of machines that we have connected. We even have one machine in our office.
But building such a machine is of course relatively expensive at first. These are too high costs for a service station operator, for example. To overcome this hurdle, they have provided the machines and also the inputs such as Costa Coffee beans free of charge and also provide the technical maintenance. The business model is that they get a small cut for every coffee sold. In reality, however, this is not quite easy, so IoT has been sought as an enabler here. I need to know how many cappucinos I sold, if maybe a vanilla shot was put in with it, and I need to record the whole thing in a very fail-safe way. No incorrect data must be transferred and no gaps must occur, otherwise I will lose sales or upset my customer if I bill incorrectly.
How do you establish such connectivity from the field to the cloud?
A vending machine runs an application that is also operated when I select my coffee on the touchscreen. And we simply docked to them. We have a technical component, a small agent or edge solution you can call it. It extracts and pre-processes the data and can also buffer it if there is no connectivity and then sends this data to the cloud.
What was the process like with you guys and what was the collaboration like with Costa Coffee?
We were involved relatively early in the process. That means we were still involved in the fine-tuning of the business model definition, and at that time we were also regularly in England and first worked out the most important cornerstones of such a solution in workshops with the entire company – the vending machine business runs under the name Costa Express.
This makes you realize once again how many special cases there actually are in a business process like this. How do they actually have to manage such a pay-per-cup model in terms of processes? You don’t get very far with a ready-made solution in this case. It was clear relatively quickly that we would have to do a lot of customizing here, and we worked with them in a prominent, agile software methodology. We flew over every three weeks, presented the results of the last sprint and thus successively upgraded functionality.
That’s how we ended up with one of the most prominent functionalities, which is the whole alerting and escalation concept, which is very important in this solution. After all, the new business model is a big challenge at first. If I stop selling these machines and instead participate in my customer’s success, suddenly I don’t even care how they use my machines. I have a very strong interest in the customer having high availability with the machines and making a lot of sales. It’s all the more exciting that I have a third party on board in the form of service partners, which are also external companies internationally. I have to get them all around the table. Then we have all the operating messages, status messages, error messages and also the operating resources that we record centrally via a root engine that sort of kneads through this data and spits out relevant alarms and then, depending on the alarm type, informs the right people and also escalates these alarms. This means that if someone doesn’t refill the milk quickly enough, after an hour or two the text message goes to the supervisor. And if it’s a technical alarm, it goes directly to the service partner via the ticket system.
Then you sit down with the account manager on a monthly basis and calculate the lost trading hours. You can then determine very conveniently on the basis of the opening hours how many sales hours I have lost due to unnecessary downtime and how much revenue I have lost as a result. This is how I create an incentive that benefits as many people as possible.
I see it as a challenge to work intelligently with the mass of data, i.e. to look at data analytics to see which alarm is triggered by which circumstances and how I have to inform whom. Is this algorithm developed by you?
That’s right, it’s being developed by us. There is also further potential for data analysis: As a company, you also want to maintain the brand and always offer consistent coffee, and you don’t want to dump beans from another manufacturer into the vending machines. That’s also when they thought about how to monitor reorder coffee and coffee sold.
You are also on a journey with the AI startup Sentian. Are you working with them on a case like this?
IoT and AI should actually find a natural symbiosis in the context of digitization projects. In other words, AI requires data above all else, and depending on the use case, quite a lot of it. And of course, IoT can provide these very well in an automated way and in large quantities. Conversely, the machines and systems in IoT projects are often embedded in complex environments, and I can map them much better with artificial intelligence.
The thing is that these are nevertheless two different technologies. This means that entering into a partnership and bringing know-how into the project is actually indispensable and definitely makes sense. Sentian is a great partner for us and one of the pioneers in AI. They’re from Sweden and we’ve been working with them to design an AIoT offering (Artificial Industry of Things) and we’ve been exploring its possibilities. Predictive maintenance is one thing, of course, but in the context of AI it is not always the most ideal use case. So we looked at intelligent control of production processes. How this works exactly is best explained with an example.
That sums it up very well. You shouldn’t always just look under the topics of AI and ask yourself, can I somehow do predictive maintenance or something. But there are other interesting use cases that I can look at. Something like supporting production control is just such an issue.
We have a case here with the company JUMO, a manufacturer of sensor technology. There was a request to support and improve the production process with the help of AI. There is an astonishing amount of tacit knowledge in the processes, e.g. from the production managers and machine operators. They simply know how to operate their machines and must also be able to react from their experience and gut feeling if, for example, I have fluctuations in the quality or grade of materials.
At JUMO, it was the case that relatively complex high-tech laser processes are used in the manufacture of the sensors, and this process always has to be optimized on a batch-by-batch basis. Until now, this was the responsibility of the machine operator who, based on his experience, tried to set the process perfectly. With Sentian, a so-called imputation model has been created, which uses sensors and analysis of the raw material and what is produced at the end, and then processes this data and better adapts the process within the batch. Thus, the process can be adjusted better than a very, very crafty machine operator is capable of.
At the end of the day, the quality assurance department found that more sensors of the highest quality level can now be produced. For the first time ever, the manufacturer is now able to offer sensors of the highest quality level in a meaningful quantity. That means the production process as a whole has also been raised a higher level and more output is now predictably available at the highest quality level, and the whole thing is AI-supported.
What data are we talking about exactly that is being collected there?
Above all, it’s about how much the laser has to ablate to find the perfect resistance – that’s what I’m controlling here. It is not a question of which data I now need from a physical point of view, but simply to determine piece by piece via the process of laser training, I am ablating more and more and I am trying to optimize this and can forecast the quality via this imputation model. In turn, I need that to take this reinforcement learning approach.
The approach is to optimize already in the production loop. Otherwise, it is often the case that I have a production process and at some point there is quality assurance. And it can happen that I’m already relatively far through production and then I come to my quality gate and realize that something is wrong. But it may be that my problem was already very far ahead in the production process. This means that many steps have been gone through unnecessarily. This reduces output and increases costs. With this model, one is able to intervene very early and make optimizations.
We can see from many customers that this is where the journey is heading. Really looking at the production process and then bringing smart technology to the process.
What other cases are there in this direction and what does your partner ecosystem look like in this respect?
I think a partner ecosystem is extremely important. What Martin already said at the beginning, that no “one size fits all” solution works on the IoT platform side. It is even more difficult regarding the digitization of entire production processes. That’s why we believe that collaboration at eye level in ecosystems is extremely important. With customers, you have critical challenges and you don’t usually solve them alone. There are specialists who can take care of this case, but you have to consider relatively many associated things. That can be sensor manufacturers, that can be the manufacturer of the PLC logic, safety technology, etc. – that’s why these ecosystems are extremely important, so that everyone can bring their stakes in.
KUKA is one of the founding members of the so-called Open Industry Alliance, which is dedicated to exactly this topic. The Open Industry Alliance puts the customer requirement in the foreground and gathers together manufacturers of machines and systems, but also software providers, who commit themselves to wanting to solve the customer’s problems together.
Just as important for us on the cloud side is the collaboration with Microsoft, also due to the geographical proximity in Munich. Microsoft supports us very strongly as a Gold Partner. There we also have very good access to the experts and can discuss at eye level. And that also goes in both directions, because we are also close to the market and to the use cases, and we also go into implementation.
Then we have very specialized partnerships like the one with Sentian, which is also really dedicated to analysis and AI model development for production processes and brings in an extremely large amount of know-how, and we like to tap into that as well. We are very reflective and do not see it as our competence to be the experts in all areas. Only together can you be strong enough.