Why the next major AI opportunity lies in factories, not e-commerce

Why focus on industry

Jiří Vicherek: Why industry? Why manufacturing? You've done e-commerce, you've done start-ups, you grew up at AVG. Rockaway's most interesting bets were mostly on digital companies – new markets and steep growth. So how come in 2025 you're saying, "You know what? The next growing market will be factories." How did you come up with that?

Petr Mahdal: We are convinced that we can leverage all the knowledge and experience we have gained in the world of e-commerce. That's the first thing. 

The second aspect is the scope for innovation. For example, the adoption of AI technologies, AI approaches and AI products has a steeper curve in the context of e-commerce. They are simply taught that way – it's a digital world, it's a transactional business, they have to adopt these things and innovate quickly. Whereas in manufacturing, when I look at the European region, for example, progress is very slow. On the other hand, this represents enormous potential for implementation. 

And then it's a great opportunity to push the potential in the European Union further through the digitisation of industry. Because when we look at it geopolitically or geoeconomically, China is the clear leader today in the context of automation, robotisation and the adoption of AI technologies. 

And of course, it's not black and white – there are many pros and cons. In the context of our region, America is also very progressive. But I would say that Europe is lagging significantly behind. 

On the other hand, this is a huge opportunity to help. To move forward and start doing things smartly in the context of the industry. And that is one of the basic theses of RockawayQ. And we at FLO – as part of RockawayQ – recognise this and are trying to build a competence centre that can guide companies through these complex transformations.

The challenges of digital transformation in factories

Jiří Vicherek: What is the difference between transforming an e-commerce business and automating a factory in Central Bohemia? How is it that so few digital companies with any competence in technological transformation focus on one of the most significant parts of our economy?

Petr Mahdal: One aspect is a kind of evangelism and awareness that the level of digitisation must be significantly better in order to move the factory forward. Of course, if you want to do it, there must always be a will, a goal, and a demand. You need relevant people who want to make that transformation happen. And often, when you talk to people in factories, for them, digitalisation or progress means replacing a production process that currently has five steps, with a person doing manual work at each station, with some kind of automated line of robots.

Jiří Vicherek: They will buy robots.

Petr Mahdal: Yes, and they can't imagine anything more. But in the context of other advanced technologies, such as methods and approaches, such as predictive maintenance, traceability, the use of video, and so on, evangelism is not yet at a level where they explicitly want this. So one of our tasks is to help them identify such things, come up with ideas, and offer them options. 

And then there is the level of digitisation itself. Today, many manufacturing processes are purely manual. And not only at the level of the manufacturing process itself, but also at the level of the data footprint, you need to be able to measure production. 

In my opinion, these are the key aspects today. But at the same time, this is precisely why we believe there is enormous potential to move the industry forward, make it more efficient and speed it up. 

Jiří Vicherek: I recently interviewed someone from a company that manufactures packaging, and I was quite surprised when he said that while they have the production part down to a fine art, they are still in the early stages of all other processes. They have invested in robots and production lines, but when it comes to the rest of the company, it is more difficult for them. Is that your experience as well?

Petr Mahdal: Yes. Many factory owners see progress in the fact that if they manufacture windows today, for example, automating production is a symbol of progress from point A to point B for them. This means that people will no longer be manufacturing, machining, assembling, welding and who knows what else is needed to make a window, but that they will order an automated line from a supplier that will produce the window profile at the end. And that is progress for them. 

In my opinion, a holistic approach is relevant, looking at the whole organism, at the entire business life cycle. That means not only manufacturing a window, but also how to manufacture it as quickly, cheaply and with the highest quality as possible. 

Jiří Vicherek: So what do these operations need most often? Are there recurring requirements, products, or problems?

Petr Mahdal: To be fair, it must be said that every customer is largely unique. But to a large extent, parts of what we can do and offer can be replicated and are consistent. I would call it a data layer, data analytics, working with data. The majority of our projects start with the need to make sure we have data. And that we have complete, high-quality, structured data that we understand.

If we want to change anything, we naturally want and need to measure it. In order to be able to change, evaluate and move forward, we need to collect data from every step of the production process, from every location, and ideally from every machine. And preferably starting with the input material – that means, for example, connecting via ERP systems to procurement so that we know that we purchased a given raw material on a given date from a given supplier, because this can have an influence and impact on the production process. We may not see it today, but it will become apparent when there is a deviation or anomaly. 

Let me give you an example. We encountered a case where an automated line was already in place. One would therefore assume that a complete data footprint would be available. But in this particular case, the customer did not have access to their data. The data was collected in a completely granular, deterministic and perfect manner, but it was dumped somewhere in the cloud to the provider of the technology.

Jiří Vicherek: And the factory owner had to pay to access it.

Petr Mahdal: Well, and then we had to figure out how to get to the data itself. And then, of course, we had to go into the details of the contracts – what data it was, where it was being transferred, why it was being transferred, and to ensure that we either had unlimited access to it or could transfer it somewhere else. 

And then there are customers who collect everything but don't work with it. Or customers who don't collect anything. But there are also cases where you combine, for example, a production process that involves advanced machines with a data layer. But at the same time, there are purely manual parts of the process – and so we are dependent on digitisation, on people writing down values from specific sensors at a certain stage on tables or on paper. And we need this data too, so that we can analyse the entire end-to-end process. Only then can we say that we have a complete data footprint and can measure and evaluate something.

Jiří Vicherek: Can you digitise a factory without collecting slips of paper? It seems terribly expensive to me.

Petr Mahdal: In my opinion, it cannot be done effectively or responsibly. That's the crux of the matter.

Jiří Vicherek: Basically, if the factory operated on slips of paper, then you have to understand slips of paper.

Petr Mahdal: Sure, if you want to move from point A to point B, to some kind of progress or innovation. And I'm deliberately not saying what the resulting business case is, because it can mean something different for every factory owner. Some just want to serve the same customer portfolio, but want to have a higher margin on it. In other words, they are looking for ways to make production cheaper, faster, and more efficient. Some are looking for a business case in that they want to be able to produce twice as much. That means they are looking to penetrate the market and expand their customer portfolio. 

We are not here to tell a factory owner what to do, as they are likely not running their business for three months, but perhaps three generations. We can suggest and advise, but they have to say where they want to go and what they want to achieve. And we are there to help them through that process. 

In one factory, for example, we found that the temperature of the oil in a kind of cooling bath was recorded every hour. And when you walk through the factory, you have to go through it several times. You have to talk to people, look at the environment, see how people work, what their body language is like, what their dynamics are like... These are all aspects you need.

Jiří Vicherek: Can't you do it remotely?

Petr Mahdal: It's not possible remotely at all. For us, it's crucial to be part of the production process as it works today for a certain period of time. We go there, talk to people, observe, evaluate. Well, what happened here was that they looked at the temperature chart and we wrote down: OK, this station is marked in this process, the technological procedure specifies that the current oil temperature is recorded every hour – fine, let's move on. But after a while, we realised that the table was somehow strange. We realised that we had gone there around lunchtime and the values for that day were written there until 4 p.m. – the end of the shift. The oil temperatures were written there four hours in advance.

Jiří Vicherek: You want to come up with predictive maintenance... and they've already done it.

Petr Mahdal: We went to that person and asked him if the date was correct, if it wasn't a table from yesterday. And the answer was: "It's experience. It doesn't change significantly, and that's how it works." The person simplified it by writing the table for the entire shift in advance. And that's an example of the level of improvisation that, in my opinion, you don't want or need in the manufacturing process, and certainly not in data analysis. Such data is useless to you. Even if it may be true, it simply does not reflect reality. 

AI in manufacturing: automatic quality control and prediction

Jiří Vicherek: You say that AI is part of the FLO and RockawayQ proposition. On the other hand, we're talking about digitisation and recording. Getting information from pen and paper into ones and zeros sounds to me like something that could have happened 20-30 years ago in exactly the same way. How does artificial intelligence come into play? If you don't have that data layer yet, how is AI an opportunity? It sounds like the opportunity is more in digitisation with software that has been around for a long time, not in the latest OpenAI model.

Petr Mahdal: It's a combination. Today, everyone is talking about AI, but machine learning has been with us for years. Data science as such, too. But it's a non-functional thesis if you don't have that digitisation, if you don't have a data footprint. Because only when you have that, and you're sure it's good quality, and you understand it, can you interpret, measure and manage it adequately, and only then does space arise for various AI components. 

And we could talk about this in much more detail. There are certain situations where AI is gaining more and more space and establishing itself. And that is exactly our direction. At RockawayQ, we want to create something like a holding company that can provide not only hardware but also software for these transformations. And FLO is supposed to be the competent force, the partner that will carry out these transformation projects through a process of change.

Jiří Vicherek: Where does AI make sense in manufacturing, where is it already beneficial? 

Petr Mahdal: It could be quality control at the end, for example. This means that when you have a physical product at the end, it usually goes through some kind of inspection. In manual operations, this role is currently performed by a person – there is an operator who receives some products and usually performs a visual inspection, perhaps a weight check, perhaps measures an angle, and so on. They simply do it manually. 

But today, we can apply AI technology so that such checks are not done manually 100% of the time, but we can filter out, say, 80% because we have video transmission, for example, and some models running in the background that can check 80% of the products at the visual level with high quality. This means that you only need to manually visually inspect twenty percent of the products. As a result, you don't need ten people at this station, for example, but only two, because you can handle eighty percent of the workload through the digital data layer and AI.
Another aspect may be that today we often encounter situations where a product or the output of a process is of poor quality and either has to be completely scrapped or sent back for modification. And that is quite expensive. On the one hand, it is time-consuming, it costs money for materials and labour, and on top of that, everything else costs money. It goes through the whole process, and you only find out at the end. 

But if you have complete analytics, you can also use AI models that you train to stop a certain batch, for example, in the first third of the production process, because the technology identifies a combination of factors. This saves a lot of energy, a lot of work, and a lot of costs. And AI technologies can be applied here as well.

The AI transformation team: data engineers, consultants and modellers

Jiří Vicherek: Let's take a look at FLO's Data & AI team. Your guys and girls visit factories and manufacturing companies, but also more traditional clients in e-commerce and retail, for example. What kind of people are they? How many of them are pure consultants with some technological know-how? How many of them are specialists, and what are they specialists in? How many of them are hardcore data scientists?

Petr Mahdal: We try to have everything you mentioned in our team. We already have it today, and we are trying to develop and build on it further. Starting with the simplest, we have experts in traditional data tasks – such as transformation, migration, BI reporting, and so on – on various technologies. 

Then we have people who specialise in industry, for example. These are consultants. Typically, these are people who partner with the customer, go through the process, go through the physical aspects, analyse it, discuss problems and potential with them, and so on. Ultimately, they are also part of the delivery, the whole journey. 

And then we also have data science experts who are able to combine AI tech with machine learning. They are able to train and create LLM models for specific use cases, for specific customers, on diverse technologies from diverse providers. 

Every project, every transformation, requires a slightly different team. We are aware of this and therefore approach each project in such a way as to put together the team that is needed.