Patrycja Welew,

Artificial Intelligence in Business Practice: How to initiate transformation on the road to becoming data-driven – a conversation with Wit Jakuczun, CTO, F33

There is much talk about the data-driven approach, and many organizations boast of employing it. But what exactly does it mean, and how does it work? How can one truly lay the foundations to become a data-driven organization? In pursuit of answers to these questions, we spoke with Dr. Wit Jakuczun, a mathematician and expert in the field of artificial intelligence, who not only delves into the theoretical aspects but has also been actively supporting companies in their transformation into data-driven entities for over 20 years as the CTO at F33.
Artificial Intelligence in Business Practice: How to Initiate Transformation on the Road to Becoming Data-Driven - a conversation with Wit Jakuczun, CTO, F33

There is much talk about the data-driven approach, and many organizations boast of employing it. But what exactly does it mean, and how does it work? How can one truly lay the foundations to become a data-driven organization? In pursuit of answers to these questions, we spoke with Dr. Wit Jakuczun, a mathematician and expert in the field of artificial intelligence, who not only delves into the theoretical aspects but has also been actively supporting companies in their transformation into data-driven entities for over 20 years as the CTO at F33.

Where should a company start to become data-driven?

Organizations must first be business-driven before they can become data-driven. The key to success is more than just having a data warehouse or focusing solely on the data-driven aspect. Being data-driven is not just about technology; it’s primarily a business approach.

Data warehouses are crucial tools, but they shouldn’t be the beginning and end of the transformation towards a data-driven approach. Though the concept of data warehousing has been known for several decades, it has sometimes led to failures. That’s because many projects started with the mindset that having a data warehouse would automatically make the company data-driven. However, infrastructure alone without a clear business strategy may prove insufficient.

Many companies have failed by beginning with the mindset that merely having a data warehouse is enough. Therefore, it’s crucial to determine why a data-driven tool is essential for a specific organization and what business goals it wants to achieve. Understanding this aspect is the foundation of effective transformation.

Organizations must first be business-driven before they can be data driven. - Wit Jakuczun, CTO, F33

Do you think that many companies still need help defining their organization’s business goals?

Unfortunately, there is a lot of truth in this. The logic for businesses should be to develop specific ambition, such as capturing a 30% market share in the sale of tissue paper. This is the ambition, but now the company must consider what that means. They would analyze their market, identify their advantages, such as a superb distribution network, and devise systems to support realizing this strategy, ambition, and this particular goal. Roger Martin described this method well in his book “Playing to Win: How Strategy Really Works.”

Of course, data can be utilized in this process, but the key is to avoid immediately focusing on costly initiatives, such as building databases or hiring a new team. These are significant and expensive undertakings that, if not preceded by a proper understanding of business goals, can lead to a lack of financial resources and emotional commitment for successful transformation. It’s worth noting that, regardless of whether you are developing a data warehouse or another project, this is insufficient. The mentioned “data warehousing” approach is already outdated. The same mistake is being made in artificial intelligence as in data warehousing 30 years ago.

What is the reason for this, in your opinion? 

I think the main reason for this mistake is the competitive pressure, which encourages taking action in artificial intelligence. Companies, especially engineers and technical people, must consider whether such technology is actually needed to avoid succumbing to this pressure. However, artificial intelligence is not a standalone management concept; it’s a technical tool. That’s precisely why the business should identify specific problems, such as competitiveness or delivery deadlines, and then decide whether and why it’s worth solving these problems. All of this should stem from a clearly defined business strategy.

How can companies avoid missing the business objective throughout the data-driven implementation process? 

Returning to the issue of business objectives, a good technique is to take a backward-looking approach to analysis. If you set a specific business goal, such as gaining an additional 5% market share, you should consider what that means regarding processes going backward. For example, the analysis may show that logistics departments are too rigid, affecting the speed of replenishment in stores, and these are critical to controlling the organization’s market share. As a result, the desire for more frequent promotions becomes challenging to execute as planned. To solve this problem, the logistics department may need new tools, such as planning or optimization tools, and more precise monitoring of specific indicators. However, the existing problem often stems from the fact that engineers start by looking for solutions based on artificial intelligence instead of precisely defining the problem they want to solve with these tools.

So once a company sets its business goals, it should then define the problems it wants to solve before seeking AI-based solutions?

This is very important to emphasize. If the business goal requires data and algorithms, then you can discuss the necessary steps, but the key is first to determine whether such a path makes sense. To become a data-driven company, you need to have defined business goals, configured monitoring systems to assess the achievement of those goals, and management systems to support the business goal, and this is where the potential use of algorithms comes in. Simply put, specific actions need to be taken if the analysis shows that we cannot make significant changes. Take, for example, increasing the capacity of call centers. Limitations can occur due to consultant mismatches and scheduling deficiencies. The first step would be for the call center to monitor consultants’ skills and analyze the flow of calls. After that, you can then implement support systems, such as better matching consultants with customers or call topics. That is AI’s place in these support systems.

How should a company implement such new solutions while keeping sight of the goal? 

I had a client who planned production using an immensely complex spreadsheet. His work involved creating and maintaining just one production plan. During our conversation, he admitted that he could only answer a few crucial business questions due to the time constraints imposed by the technology he was using. Imagine a situation where such an employee works daily for 8-10 hours, deeply engaged in his work, and then hears from his boss, “Next year, instead of managing two factories, you’ll be managing six because we’re signing a contract to build another four factories. How do you feel about this idea?” Such a planner would likely feel overwhelmed because he’s already working at the limit of his capacity.

In the context of organizations, the company’s logic needs to focus on what needs to happen for the company to triple its growth – where does the secret of success lie? A planner burdened with a spreadsheet cannot meet such challenges. In such a situation, he should express the need for tools to enable him to work on a larger scale. This brings up both data-driven thinking and artificial intelligence. It’s necessary to identify where the barriers are in this employee’s planning process. For instance, the factory scheduling process might be too time-consuming, so it could be optimized using algorithms to significantly speed it up. This means the same employee could create, for example, five plans in the same amount of time, contributing to increased efficiency. In such a case, investing in AI-based solutions is justified, especially if the outcome is to triple the company’s growth without a simultaneous increase in team employment. It’s important to realize that, although the company is currently profitable, it doesn’t guarantee sustainability in the long run.

Does it happen that the companies you talk to need help to define what they want to achieve with the new tools? 

Sometimes, I meet clients who expect to create a model but cannot clearly explain why they need it. The model does not meet the business objective, which can lead to a high risk of failures that we want to avoid. Then, we don’t throw ourselves into making models right away – first, we sit down and give business advice on positioning AI well in the optimization process.

First and foremost, it is necessary to define a business goal, understand how to assess and monitor it and hen align data and business processes.

I will cite a specific case of a company that reported a problem with salespeople selling a mass service on the open market. The company owner felt that not all salespeople were equally effective in negotiating the service price. Some were selling cheaply, while others were selling expensively. He wanted to improve the company’s margins and avoid too low prices. Therefore, he asked me to create a model predicting what the prices of this service would be in the market for the next day, similar to the stock market. When I asked how this would affect margins, he said they would sell better by seeing the prices. I expressed doubt that only by knowing the average price would they be able to improve their operations. With knowledge of the average price, half of the services would still cost more, and half would cost less, making it challenging to determine margins accurately. By inference, the model was unrelated to the actual business decision. 

Aren’t other factors also affecting the organization’s margins? Just because a competitor may sell at a certain margin and not another does not mean such a margin will be profitable for this organization.

Yes, exactly. I asked him: What makes someone a better salesperson and someone else a worse one? And you know what came out in various conversations? The better ones in this organization are the more experienced ones because they get better deals. The new employees get the smaller, worse deals. I suggested that he create a different model to see if a salesperson, placed in a specific situational context, has reached their maximum potential. If we want to hire talents and promote them within the organization, what’s the point if someone gets better deals and someone else gets worse ones?

Imagine you’re a sales representative going to a poor rural area. You can’t expect someone who has meetings with 30 small shopkeepers in a small town to make the same turnover as someone who meets with 150 shopkeepers because they work in a more dense region, like a major city. We may need to think about a model that benchmarks whether the one who meets with 30 shopkeepers isn’t more efficient than the one who meets with 150 shopkeepers. Price is one thing, but the efficiency of these salespeople and their decision-making are another to improve the business goal.

The problem was that this client was very determined about what he wanted to do. He imagined it possible to forecast the perfect “tomorrow’s” price. That’s not possible. There are no models that hit the mark perfectly. All models make mistakes. This was a typical example of wishful thinking, where someone wanted to solve the problem starting from the solution rather than from the problem. Ultimately, a model monitoring the efficiency of employees was implemented, which gave them clear guidelines on how effective their actions were. Additionally, this model was linked to a new incentive system and a reorganized entire process.

What do you say to such customers you meet on your way?

To be data-driven, I always advise clients to formulate a business goal first. We must precisely define how we’ll assess progress and what parameters to monitor. The answers to these questions will help us identify the necessary data. Then, we need to understand the business process related to the issue, such as customer service operations, where inefficiencies occur, and how to improve throughput. Both algorithmic problems and the need to create IT systems may arise here.

For example, in a call center, a problem might be assigning the right consultants to specific customer service cases. Although we can identify the best consultants for specific situations, a lack of tools to manage their availability may hinder the effective implementation of algorithms. Ongoing operations can defeat any plans without systems to support consultants’ time management.

We often focus heavily on data and algorithms throughout the AI and data-driven process, but I believe understanding the business holistically is crucial. We must thoroughly understand the processes and expected outcomes and consider how the new solution will impact the current situation. Sometimes, it’s necessary to install new IT systems and adjust data. Data is crucial for observation and measurement, and algorithms help better analyze this data, but both elements should work together in collaboration with the business context.

In most projects I’ve worked on, it was necessary to adjust or expand IT systems, modify processes, and implement models or data warehouses. Monitoring is crucial for evaluating the situation after implementing the algorithm. Therefore, many companies that focus solely on algorithms or data preparation often neglect process updates, which leads to ineffective results. Furthermore, a lack of monitoring means we don’t know if the solution is already working, how it’s working, and if it meets expectations. Such a strategy dooms the project to a quick death.

I often notice that companies implement new systems that nobody wants to use afterward, and after a while, these solutions die. Then, the search begins for new tools, such as AI, to fix the mistakes of previous attempts. Do you notice this, too?

Indeed, unfortunately, this happens. The same applies to ERP systems and other similar solutions. Implementing an ERP system is only possible with a certain level of organizational maturity. Without defined processes and a certain degree of order, it’s out of the question. Implementing such systems is often challenging when companies assume purchasing them will provide them with standard processes and best practices. Companies try to adapt their business to pre-defined standards. It’s like thinking that buying the pen Hemingway used will make you as good a writer or that mimicking his morning ritual will transform you into a literary master. In reality, you’re forcing something that doesn’t fit you because you’re a different person with different needs, thinking, team, and culture. This is the most common mistake. We recommend thoroughly describing and examining processes, considering whether they are beneficial, and then introducing managerial and monitoring systems. Only then are we ready to implement an ERP system, not the other way around. Furthermore, ERP must be tailored to our methodology and thinking, not vice versa.

It’s often said that purchasing an ERP is supposed to generate a “competitive advantage.” However, how can you achieve a competitive advantage by buying processes and methodologies from other organizations that any company can acquire? Competitive advantage lies in having something unique that others don’t, creating a barrier to competition. Careless ERP implementation makes you one of many clones. This requires significant intellectual and organizational efforts which is why many companies struggle. ERP itself is also an old and well-known concept. An expert will emphasize that if there’s chaos in the company and a lack of order in processes and data monitoring, it’s not worth starting with ERP. These systems are designed to increase efficiency, automate processes, and improve quality by eliminating errors. They can bring many benefits, but as Greg Bigos [CEO, F33] always says – you must first define the process manually before moving to automation. Automation requires standards, so without them, there’s nothing to automate. In automation, a robot must make a clear decision – either left or right. If the company’s answer is “maybe,” then artificial intelligence can’t provide the solution.

Do customers often come up with ready-made solutions that they want to implement?

One client came to me with a model that a provider had developed for them, which his company did not want to use. I asked for details about this model, and it turned out that they had high-quality code and a well-prepared model for a specific business case. However, the business wasn’t involved in the provider’s project and didn’t provide data. The provider created the data on their own! The project was completed as planned, but when the business saw the results, they found them unsatisfactory. The project ended, and the contract with the provider expired, but the model remained. Evaluating the quality of the model creation project itself, one could say it was executed at the highest level. However, from a business perspective, this project lacked value. It’s difficult to speak about objective value; it was certainly an important step for the project team, but it didn’t bring any benefits from a business standpoint. The problem was their  approach of “we want AI in the company” without fully understanding the business process. There is a lack of prior analysis of the project’s purpose.

Therefore, when working with clients, I always emphasize that they must have a business justification for their goals. They must justify why a particular model should be created and why such data should be used, not others. It’s a challenging task for our clients that is often a significant barrier. I believe communication and the ability to articulate thoughts clearly are crucial in this industry. Sometimes, clients need help understanding what I’m asking for. The more questions I ask, the more they reveal that they are not sure what they want to achieve. This can lead to frustration and a sense that it might be better to give up. The model should have one specific task defined. If you want it to provide forecasts for a week, it will deliver for a week; if for a day, then for a day. You need to clearly know what you want to achieve so that the model can effectively deliver it. In discussions, it often turns out that it won’t be just a single model but a solution with many interconnected models.

It’s a tough task when you sometimes have to make people realize that they don’t know everything, after all, like they assumed they knew about their business.

I don’t look at it that way. I also don’t know much about many businesses I help. I can’t teach someone how to run a business a certain way. They must know what business goals they want to achieve. I, of course, try to help them formulate them, and I can advise up to a certain level. I approach such businesses humbly, but I also expect to have decision-making power in creating models because I’ve been doing this for 20 years and am good at it. Sometimes, clients don’t know about models and they try to lecture me that one model will be better than another, yet I know it’s not. If I went to some manufacturing company making half a billion dollars in revenue and lectured them on how to organize production, they would quickly know I’m not an expert on this subject. If someone makes half a billion in revenue with a good margin, it means they know how to run their business. You must understand everyone’s role in the project and work together towards the goal. That’s the hardest part. A company participating in a project where models are used often focuses heavily on those models because they don’t understand them, so they perceive them as a risk in the project. People need to be reminded that there are areas they don’t know about in every project, and they should trust the experts to fill their gaps. People often pay a lot of attention to things they don’t know about, ignoring risks in areas they are familiar with. 

When I introduce model-based solutions in companies, people often ask me if it will really work. I tell them that if we meet certain conditions, we will definitely get a solution that will make the business operate more efficiently, but I also emphasize that there are no guarantees of success. My role is to make clients aware that if their expectations are unrealistic, they cannot be fulfilled. It’s a matter of honesty.

The hardest moments are when a company gets really excited that they will have AI-based solutions, and then I come and make a very simple model for them. They were disappointed later that I made such a simple model when they wanted advanced AI-based systems, like a deep neural network. I tell them this is a sufficient “good enough” solution to meet their business goals. It’s simple, efficient, low-maintenance, and everyone understands it. Why make some complicated neural network that is difficult to maintain, expensive, and doesn’t provide better results in the end? It doesn’t make sense.

Someone tuned in expecting a “wow” solution, got something simple, and is disappointed?

For me, the “wow” is that the solution works and the business objective is met. That kind of customer mindset is hype-driven, not data-driven. You have to ask yourself honestly—do I want to do something because I need it, or do I want to do something because others have it? If it’s the latter, then you need to consider whether trying on such a project makes sense.

Finally, let’s return to the question: How do we become a data-driven organization? 

The answer is that you need to build an organization with a culture of continuous improvement. Only then can you talk about being a data-driven company.

Behind F33
Greg Bigos, CEO

Greg Bigos is the CEO of F33, bringing over decades of experience in delivering ERP solutions for manufacturing, logistics, and retail industries.



Contact Greg
Behind F33
Wit Jakuczun, CTO

Wit Jakuczun is the CTO/Chief Data Science Officer at F33 with a PhD in Applied Mathematics and over 18 years of experience in mathematical modeling, data analysis, and simulations.

Contact Wit

Consult us about your needs

Get A tailored 30-minute 1-on-1 consultation with an AI expert.