Why Are 88% of Companies Wasting Their AI Budgets? | Marcin Rzepiel | Euvic Talks

Marcin: The people who are relieved in some way thanks to automation and AI will be able to do other things, things that are more creative and bring us closer to monetizing our work, for example. And this also applies to less creative departments, because there are still certain things that cannot be fully automated, or that we simply would not want to automate. 

[music]

Bartek: Hi, Bartek Śliwa here. Welcome to another episode of Euvic Talks, where we connect business and technology. And today, in addition to business and technology, we are adding strategy. More specifically, a strategy that has been on everyone’s lips recently: the strategy for implementing artificial intelligence. My guest today is Marcin Rzepiel, AI Transformation Lead at Euvic S.A. Hi, Marcin. 

Marcin: Hi, Bartek. Hello everyone. 

Bartek: Marcin, let’s start with a warm-up question. How does someone become an AI Transformation Lead? 

Marcin: That is a difficult question. But I think you need to show a certain kind of proactivity and come forward with initiatives. You also need a general feel for a space that combines three elements: processes, technology, and business. If you have spent enough time working across those three areas, then in today’s world, where AI is on everyone’s lips, this feels like a natural path. 

Bartek: All right: processes, technology, and business. How did you develop fluency in those three areas? 

Marcin: I was not afraid to step outside my comfort zone, or as people say now, to expand my comfort zone. Whenever interesting initiatives appeared, or whenever interesting customer projects came up, even if I did not feel completely confident or fully skilled in a topic, I was not afraid to take initiative, get involved, and simply learn on the job. That was true with technology as well. I joined Euvic as an intern at the very beginning, and I was strongly connected with technology and programming. 

Bartek: That was a while ago, right? 

Marcin: Ten years. More than ten years. When you say ten years, that sounds fine. But when I usually realize that ten years is one third of my life, it gives me a moment of reflection. But yes, it really was ten years ago. 

Bartek: Okay, moving on to the topic of our episode. Everyone is using AI now. Everyone, everywhere, all the time. Everyone is an expert, everyone says they built something in a chat. But we are reaching a moment where everyone says they use it, while at the same time they still say it is a toy or that they do not see results. What might that come from? 

Marcin: This is, of course, only my own thesis. There are statistics, for example from McKinsey, saying that around 88% of organizations declare that they use artificial intelligence. But a much smaller percentage says that they really see the effects of using it. I think this comes from the fact that no one examines how AI is actually being used in those 88% of companies. In many organizations, there is no concrete strategy. There is not enough empowerment from the top, from the board, or from some kind of center of excellence, so that people actually want to use AI and experiment with it. There is also a missing mindset shift. People are naturally afraid of artificial intelligence because of the information bubble that says there will be reductions, that AI will replace you, that AI will eliminate a lot of jobs, and that we are in the middle of a revolution similar to the Industrial Revolution from several hundred years ago. But we need to change people’s mindset a little. If we change that mindset, and if boards or managers give people the empowerment to use AI, and if an organization applies a systematic, strategic approach to managing AI, then both the effects and the way to measure those effects will come. This is my personal diagnosis and thesis, which I try to confirm empirically in many cases. I have a working sample of observations. 

Bartek: We like personal reflections, so I can definitely agree with that. As for layoffs, we have already gone through several waves which showed that after the outflow there is an inflow again, and the same companies hire people back because AI is not quite what they expected. 

Marcin: Anthropic is an interesting case. It is the company behind Claude and the Opus and Sonnet models, which are among the most popular in our software development world and, I would say, do the best job there. On one hand, Anthropic’sCEO keeps saying that programmers will not be needed soon, that there will be reductions, and that the market will change completely. On the other hand, if you look at the number of job openings Anthropic publishes for software engineers and programmers, it is around 100 positions. So they are still looking for the people they say will not be needed soon. I am curious whether this is really a transition period and whether those people are needed while they are building the rope for themselves, so to speak, or whether that CEO colleague is slightly wrong here. 

Bartek: Okay. Maybe he does not know yet himself. 

Marcin: Maybe. 

Bartek: You mentioned Anthropic and how AI helps us in software development work. Today, I think I would like to focus on how AI can support other departments, or companies that are not strictly software companies. From your experience, where do you see the low-hanging fruit, for example in back-office departments such as HR, finance, and procurement? 

Marcin: First of all, this is also only my opinion. But I am convinced that there are really a lot of low-hanging fruits, and they are very easy to pick. The main thing we should do is allow people working in back office to use tools available on the market in an organized way, not as shadow AI. For example, if we are strongly embedded in Microsoft, then we need to allow people working in back office, marketing, HR, finance, and accounting to use Microsoft Copilot. 

Bartek: Let me stop you there. How do we do that wisely? Everyone has Excel on their computer, but I assume maybe one in ten people can write a complex formula. So if we give Copilot to anyone, regardless of department, without proper training, I assume we should not expect good results. 

Marcin: We should not. But the advantage of Copilot, or any LLM, is that technically it understands natural language. So the entry barrier to writing a good prompt and providing the right context for the prompt, or for what you want to do, is much lower than in the case of complex Excel formulas. You do not need some long 80-hour training on how to specialize in Excel. Training is needed, but when organizations deploy Copilot, or already have Copilot but have not yet trained users, including back-office teams, they need to take care of that. They can use an internal specialist, for example a developer or an enthusiast who knows it inside out, and organize training sessions that should not be very long. Even a one-hour session on building context, building prompts, and using Copilot to search for information in a company SharePoint can already bring real effects. Back-office departments, for example HR teams searching through a maze of contracts, can retrieve specific information from those contracts without doing it manually, as long as they know how to build a good prompt. 

Bartek: Especially since technology can now read not only text written in Word, but even scans. So that is not a problem anymore. All right, let us get to the substance. Do you have a concrete AI use case in a back-office department that you could share, and whose effects you could show? 

Marcin: Yes. What I am about to describe was really a low-hanging fruit. In the project or organization I am talking about, we do not have Copilot and we do not have a Microsoft environment. We are on Google and use ChatGPT. Every employee in that organization has a ChatGPT license. We have an Enterprise account. 

Bartek: What organization are we talking about, so we have the context? 

Marcin: We are talking about Join The Crew. Join The Crew organizes yacht trips, mainly for DACH countries, meaning Germany, Austria, and Switzerland. They also have an international part, where they organize trips for other countries. In this organization there is a strong focus on marketing. As the CEO always says, it is a marketing-driven organization. The largest budget is in marketing, and the largest number of people work in the marketing department. 

Bartek: Now all the marketers from LinkedIn are heading to Join The Crew. 

Marcin: Yes, exactly. And that is where there were the most use cases in which AI could be used. The most recent case we managed to implement was done using simple ChatGPT, and more specifically using what ChatGPT provides out of the box. In the first version of the agent I am about to describe, there are no external integrations. It is only the interface and clicking things inside ChatGPT. What did we do? We started with the as-is state. I will keep referring to this: every introduction of AI into a specific process must be preceded by process analysis. 

Bartek: That is exactly what I wanted to ask you about later: how to conduct the process of implementing AI in a department or in a given process. But you can already hint at it. 

Marcin: Sure. There was an internal process that was not fully structured and was not done regularly. One or two people did it, and it involved researching the competition. Join The Crew has several competing organizations that operate in a similar way. One or two people, sometimes just one person, would go to competitors’ websites and check what types of offers had appeared since the last visit. They checked what new trips or new destinations a competitor had introduced. This is important. We need to know where the market is going. We need to know who to benchmark against. But we started wondering why anyone should do this manually in today’s technological reality, especially since we had a ChatGPT license. We were able to create an agent with a fairly complex prompt and fairly complex instructions, with our strategic goals for Join The Crew attached. It automatically scans all competitor websites once a week, on Monday, and builds a report with suggestions: what competitors did, what the suggestions and lessons learned are for us, and what we could do to improve our own offer. Moreover, the agent remembers previous results, so it does not suggest the same thing every time. It remembers what it caught the previous week and therefore knows the difference. Implementing such an agent took literally one day of work. Building the prompt and providing the right context so that it understood why we were doing it and for whom was the most complicated part, but it still took one day. And based on this, some ideas for specific trips have already emerged, which we will probably implement this year. 

Bartek: So the return is not only in saving employees’ time, but probably also in monetizing new ideas. 

Marcin: Yes. That is what we call unrealized value. We perhaps did not realize there was a value here that we were not realizing. Our main goal was to optimize the process so that someone did not need to manually go in and monitor those websites. But as a side effect, it turned out that we can better monetize certain trips, come up with new trips or new destinations, bring in more money, and reach a broader audience. 

Bartek: We like side effects like that. 

Marcin: Yes. 

Bartek: Based on that case, let us answer the question: what happens to people whose work changes as a result of implementing such an agent, or any other solution based on artificial intelligence? 

Marcin: I like this question, because in a way I am personally affected by it. I have always liked conceptual work. I have always liked work from a helicopter view, at a high level. I have always liked work where something needs to be invented or planned at a high level. When it comes to the boring part, in my opinion, sitting down and, for example, tracking competitors and reviewing pages, that work has never been enjoyable for me and still is not. It is similar here. The people who previously browsed the sites manually, caught new items, and identified new destinations introduced by competitors can now focus on creative work. Since it is a marketing department, they can shift their focus to thinking through, analyzing, and conceptualizing what from the report we could implement ourselves, rather than focusing on tedious, manual, repetitive work. 

Bartek: We are talking here about what you nicely called creative work. But we very often deal with work that is not necessarily creative, rather manual, and I assume that resistance and fear may be greatest there. How can organizations deal with that? 

Marcin: First of all: education, education, education, education. We cannot say and cannot run a narrative that our main goal with AI is to eliminate X FTEs in a given department. That is not the narrative we want to run, and it is not what we want to do. I believe that process optimization and reducing the time certain processes take, thereby freeing some resources, should be redirected to unrealized value. People who are relieved in some way thanks to automation and AI will be able to do other things, more creative things, things that bring us closer to monetizing our work. And this also applies to less creative departments, because there are still things that cannot be fully automated or that we would not want to automate. 

Bartek: So, paraphrasing, we implement an AI-powered process with openness to the fact that something new may emerge from it, and the people we relieve from work should keep in mind that they will be sailing into these new, undiscovered waters. 

Marcin: That is an important point, because the aspect and pillar of change and adoption in organizations, which we call people, is extremely important. We need to prepare people and build a mindset in which they are open to change. I know this is very, very difficult, because some people are naturally used to certain things. Maybe they have been doing something for twenty years and have become used to a certain lifestyle, behavior style, and work style. But today, and in fact always, change is the only certain thing in the universe. Now even more so. Building openness and showing people that stepping outside the waters they know perfectly well is not necessarily bad is something we should do. 

Bartek: Clear. Now the ways of working that I announced earlier: the process of approaching an implementation, how to start it, how to conduct it, how to identify it, and how to do it without tripping over your own shoelaces. 

Marcin: This is also a topic I really like. For many years, organizations have often worked with lean, often with agile, and used Scrum for different things. We know that and understand it. I strongly connect AI transformation with a lean approach. If we want to optimize a process and look for those mythical wastes, we approach it systematically. We map the as-is state, look for waste, look for places where we can eliminate a process step because it is waste, set a hypothesis of how the target process should look, and then step by step try to get there through the PDCA cycle, the Deming cycle. In general, I believe the situation is similar when implementing AI or AI-based optimization in specific processes and departments. Once we have mapped the process, we do not only ask which process step we can optimize in a human way. We also put AI on the table. We involve a person who is skilled in AI in the process of forming the hypothesis. Based on the mapped process, we also look for savings from the AI perspective. I really like the comparison that this is basically process mapping and process optimization from lean management. 

Bartek: Okay. So if we have manufacturing companies, they have definitely heard a lot about lean. The new element is bringing someone skilled in AI into the whole puzzle. 

Marcin: An AI person, yes. But of course there is hype around AI. AI is on everyone’s lips. I understand that some people are tired of constantly hearing about AI, especially if they fall into an information bubble where every post is about AI. That can be frustrating. But let us not forget that many organizations talk about AI while they do not even have simple automations in place, automations that have nothing to do with AI. When mapping processes, looking for optimization, and looking for that mythical waste, we should also take into account ordinary optimizations and ordinary automations, not necessarily related to AI. For example, if an email is sent to HR and, on the basis of that email, some kind of contract attachment with its own template needs to be generated, there is no AI in that. It is a simple automation that creates an effect and reduces lead time in that process, but it contains no AI element. We must not forget this. We can talk about AI and that is beautiful, but ordinary automation is still valid and is a foundation, I would say. 

Bartek: Now I have this thought: a CEO who is excited about AI and wants to implement AI may very often discover areas like the ones you just described, meaning areas that do not necessarily require AI, but still allow a certain kind of automation and therefore allow the organization to pick those low-hanging fruits. So again, we have one goal, and positive side effects give us additional value. 

Marcin: Exactly. And that is also a real-life case. In Join The Crew there is, of course, a lot of AI hype. We work intensely on it. 

Bartek: Let me guess who created it. 

Marcin: Yes, that was me. I admit it. But it brings results. Still, the AI hype is there. I fell into that trap myself. AI was at the front of my mind every time, and I did not immediately realize that many things can be automated the old way. You do not need to think in AI terms. Two examples. At Join The Crew we work on Google Workspace. It is not a Microsoft stack, it is a Google stack, and we have Google Workspace Studio available, where you can perform simple automations. Of course, now they can be connected with Gemini and you can send data there or ask Gemini to return or process data, but in general there are many things you can do simply by clicking in a drag-and-drop editor. You can create a simple automation that prepares an email draft. There is no AI in that, and I personally forgot about it. My first thought was naturally what agents or models I could create to solve the case. But sometimes you need to take a step back. Another interesting case, related to AI but not directly, was an automation we wanted to implement in accounting. Once a week, an Excel file was being updated manually. Data was downloaded from a CSV file on AWS, and someone manually imported that data into an Excel file available on Google Drive within the organization’s workspace. I was asked to automate it because there was no reason for someone to do it manually. My first thought was, of course, how to do this with AI. But then I thought: wait, why AI? Excel has macros. In Google’s version of Excel, meaning Google Sheets, we have Apps Script. It can simply be programmed so that once a week it automatically downloads the data. The connection with AI was that I did not write the script myself. I asked Claude to prepare a script that downloads the data and adds it to the spreadsheet. That was the AI element. But the automation itself does not have much to do with AI. So I cool the enthusiasm a little. We should think about AI, of course, but we should not be as excited as the information bubble around us. Sometimes we should take a step back and consider how to solve something the old way. 

Bartek: Exactly. How do you cool that enthusiasm? Because you enter an organization, here you build one agent, there an automation, and suddenly things happen that people had not even thought were possible before. Then, on this wave of excitement, 74 new ideas appear in everyone’s heads. How do you deal with that? How do you filter it? What can we do so that AI actually serves us, instead of ending up in a jungle of agents that, at the end of the day, no one knows what they are for and no one uses? How do we handle that overload? 

Marcin: First, whenever I personally enter an organization, I try to cool enthusiasm by example. When there is no AI somewhere, I try to say it openly, not build a narrative that everything is done with AI, but set expectations, for example with the CEO. CEOs are often visionaries, so sometimes you need to bring those visions back down to earth. As for managing all these ideas, because visionaries have the most of them, that is a challenge. The way we do it, for example at Join The Crew and in other organizations, is to introduce an innovation hub. It is a place where all ideas go. In this case it might be a board in Jira or another tool in which we manage the organization project-wise. We create a backlog of all ideas: those related to AI and those not necessarily related to AI. In Join The Crew, for example, they may be ideas for new trips or ideas for simple automation in a specific department, unrelated to AI. We collect all these ideas in one innovation funnel and try, with the right working group, to score them, for example every two weeks during a joint meeting. Nothing works more strongly on C-level than numbers, hard numbers. We do not always have those numbers. Sometimes our scoring is based on a simple matrix: impact versus effort. What will it give us and how much will it cost? Then we work only on the items at the top of our innovation funnel. We do not work on elements that we already know do not make sense. It is important that the working group evaluating the ideas includes a broad range of perspectives. Something that makes sense from the marketing department’s perspective may very often not make sense from the operations department’s perspective. Some automation or idea may simply add more operational work. In the end, from marketing’s perspective we optimize a process, but from operations’ perspective we add work. Is that good for the organization? Not necessarily. And that needs to be taken into account. 

Bartek: Okay. Following that logic, can an organization not be ready to use AI in its processes? 

Marcin: Yes. Very often, yes. Very often that is the case, and it is connected with many things that could probably be a separate podcast episode. It is the mythical AI readiness. Are we ready to implement a systematic approach to AI in our organization? Organizations are often not ready to implement AI right now, immediately. Very often some kind of preparation is needed to maximize the value coming from AI. What does that mean in practice? In practice, it means that our data may not be properly prepared. Our data may be scattered. We use Notion, Google Drive, Microsoft, and even physical binders to store company data. So we are not ready from a data perspective. For example, with data stored in five different places, we cannot create an agent that will act as an onboarding guide for new employees, because there is nothing solid to feed it with. If we feed it data that is not valuable, or is scattered, or contradictory, because there are fifteen versions of one document in fifteen different places, then we are not ready to implement a specific onboarding guide. At that point we need to take a step back and consider what we can clean up systematically in the organization so that the data is in one place, structured, and has ownership. Only then can we think about AI. 

Bartek: Okay. I wanted to ask you for one specific step that every organization can take here and now to become, in old Polish terms, more AI-ified. I assume the answer would probably be cleaning up data, but since you already said that, please give me a second answer. 

Marcin: Okay. Give people empowerment to start experimenting. I once saw data, I cannot quote the exact source now, saying there are several phases of AI implementation in an organization. One of the first phases is the experimentation phase. We cannot skip that phase and suddenly become AI-native without having any AI. That is impossible. We must go through the experimentation phase. That phase happens when we give people some kind of permission to experiment with AI. Not by using their own private licenses, but by giving them the right environment and the right space for experiments. That is key. Training them is key, but we need to give them space to experiment, because every employee is creative and may come up with an idea for using AI that no one else would think of. Everyone has a different way of thinking, and giving people room to experiment is a step that every organization can take. 

Bartek: Of course, it is worth keeping control over those experiments too. 

Marcin: Yes. 

Bartek: So not resisting it and covering your eyes, saying, “I am not here and AI does not affect us,” but actually taking a step forward and giving people that space in an appropriate and safe way. 

Bartek: Marcin, where can someone find a catalog of AI services that a company should or could consider? 

Marcin: [laughs] In our organization, we provide several services connected with AI. There are three, at different places and levels of detail. The first one, and actually I will start from the end, is that we were doing AI before the whole boom. 

Bartek: I really like when you repeat that. Always before the whole AI boom, before ChatGPT became so popular. 

Marcin: Exactly. Even back then, we were creating custom models connected with computer vision and predictive models in the areas we work in. So we can build custom models. 

Bartek: The classics of the classics. 

Marcin: The classics of the classics, exactly. And they are still relevant, because it is possible that ideas in the innovation funnel cannot be implemented using a simple agent in ChatGPT. The second element we often provide is consulting, but without the negative connotation of that word. It is concrete work together with the team. It means assigning an AI ambassador to a specific department, for example, to help that department become more AI-capable in the organization we are working with. The most comprehensive product is an entire AI transformation. It includes the people factor I mentioned, meaning a way to shift the mindset of employees. We also check whether the organization is ready for AI at all. We think about how to prepare the innovation funnel if it does not exist, or how to measure the ideas that are already there. We look at how to build a business case from them, meaning how to form a hypothesis about how much savings they may bring. And if needed, from that AI transformation we can also move into implementation, because as I said, we were creating AI models before it was fashionable. 

Bartek: The characteristic question at the end of this program. 

Marcin: I was waiting for it. 

Bartek: What has inspired you recently? 

Marcin: I was waiting for it, and I did not find a perfect answer in my head, but I will stitch something together. Recently I saw a video somewhere, on one of the platforms, where a man was showing five books worth reading. One of those books was Crime and Punishment by Dostoevsky. I went a little broader and thought: maybe it would be worth refreshing the school readings I read twelve years ago, or however many years ago. I am still in the middle of that inspiration, let us say. I have already finished Crime and Punishment. It is a great book. I do not know why, but I probably did not fully understand it when I read it in high school. And now I am reading Ferdydurke. 

Bartek: Let me know when you get to The Children of Noisy Village. Marcin, thank you for the conversation. Thank you all for listening and watching, and we will see you in the next episode of Euvic Talks. Thanks.