Many AI projects begin with the wrong question. 

A company selects a tool, purchases licenses, launches a pilot, and only then asks where it can actually be used. That puts the steps in the wrong order. 

The first question should be: 

What business problem are we trying to solve? 

Only then should the conversation turn to technology. 

Marcin Rzepiel points to a common problem with how AI adoption data is interpreted: 

The number of users or purchased licenses says little about the effectiveness of a transformation. A company can provide AI tools across the organization without changing a single critical process. 

Before launching another project, companies should answer five questions: 

1. What specific business problem are we solving? 
2. How will we know whether the project has succeeded? 
3. Who is accountable for the outcome? 
4. Which process or way of working will change? 
5. What value do we expect to create over the next 6–12 months? 

Only after completing this analysis should the organization select a technology solution. 

One of the most important points from the conversation concerns processes. Organizations often try to apply AI to a process that no one has thoroughly analyzed beforehand. 

“Any implementation of AI within a specific process must be preceded by a process analysis.” 

In practice, this means going back to basics. First, the organization needs to understand how the process works today, who performs each task, where delays occur, which steps are repetitive, and what work can be eliminated. 

“We map the current state, identify waste, and look for places where we can eliminate a step in the process.” 

Only then can the organization determine which technology makes sense. 

The sequence may look like this: 

1. map the process in its current state, 
2. identify unnecessary activities and bottlenecks, 
3. simplify the process, 
4. assess opportunities for conventional automation, 
5. evaluate whether and where AI adds value. 

This is an important distinction. Not every operational problem requires a language model. Sometimes the best solution is to remove a step, integrate two systems, or automate the transfer of data. 

AI makes sense when it provides a capability that a simpler solution cannot. 

The most compelling AI use cases do not always look like large-scale transformation projects with substantial budgets and months-long timelines. 

A good example comes from a situation Marcin Rzepiel described at Join The Crew. 

One of the company’s manual processes involved monitoring competitors. Employees regularly reviewed competitors’ websites, tracked new offerings and changes in their operations, and then preparedreports. 

The work was necessary, but it was also repetitive and time-consuming. 

After analyzing the process, the company implemented an agent powered by ChatGPT. Its role was to monitor competitors’ websites, detect changes, prepare reports, and recommend potential actions. The agent also drew on previous findings, allowing it to compare developments over time instead of starting each analysis from scratch. 

The solution itself was modest in scope. 

“Implementing an agent like this literally took one day of work.” 

This example shows exactly why it pays to start with the process rather than the technology. 

The starting point was not an attempt to create a groundbreaking AI product. The goal was to improve a specific task. 

The outcome turned out to be broader: 

“Our primary goal was to optimize this particular process (…). Along the way, we discovered that we could also monetize certain ideas more effectively.” 

This is where the example becomes especially relevant from a leadership perspective. Automation does not have to end with time savings. Freeing employees from repetitive work can create room for activitiesthat previously received too little attention—or none at all. 

The value, then, lies not only in the improved process itself, but also in what the organization can do with the time and insights it has gained.