We encounter this every day – we arrive at a company where AI is humorously mentioned in every other sentence. Management enthusiastically talks about the latest technologies, has read articles, and listened to podcasts. They see AI as the key to greater efficiency and have an interesting budget prepared for its implementation. However, when we ask what specific business problem they want to solve with artificial intelligence, there is often silence.
Many companies are rushing after the trend of "everyone is doing it, let's do the same, or we'll miss the train." These companies often lack a clear IT strategy and do not know how AI should assist them. Although they allocate a budget for implementing artificial intelligence, they lack the resources and data to train their own model. For example, replacing a salesperson with twenty years of experience with artificial intelligence is simply not straightforward, and I would say that for many Czech companies, it is downright unrealistic. At that point, the company must look for other ways to simplify work and business with the help of AI.
When a company wants to implement artificial intelligence, there are essentially three paths it can take. The first option is to train its own model. As I mentioned, this is quite complicated and often unrealistic. Training requires a massive amount of quality data and a significantly higher investment. Therefore, for most companies, this scenario is unfortunately financially and technologically unattainable. The second option is to adapt an existing model to the specific requirements of the company, which is a kind of "fine-tuning." There are quality open-source models, but they still need to be fed with data. Thus, this approach is still relatively costly – just the hardware and infrastructure alone will cost at least a million crowns – and requires an experienced team to continue training the model.
The third and simplest way is augmentation, which means using a ready-made tool with built-in AI. This principle underlies the vast majority of "AI products," which are mostly well-known augmented models wrapped in a nice UI. Such a tool can already perform certain tasks, and companies simply "direct" it according to their needs. You can imagine this as a voice robot in customer support that can handle simpler requests. This principle is the foundation of most AI products, including our Balzaq system, which we have trained for our e-commerce clients who need to generate and translate all text for offered products into foreign languages. For an e-shop that offers four product variants in six different languages, this means translating hundreds of thousands of descriptions. AI can handle this automatically and saves the company about 80% of costs. Although it is necessary for an experienced editor to subsequently review the texts, AI can still save the overwhelming majority of the money that would otherwise be spent on an army of copywriters and translators.
Companies also struggle to envision what results to expect from AI. The reason is that they lack hypotheses and know-how for their business. They are limited by their inability to identify suitable use cases or the technology to apply to them. It is enough to take small steps. For example, during the development of Balzaq, we needed to determine the success rate of the model in translating texts into German. We are talking about percentage differences, which can mean saving dozens of hours of work. Unfortunately, the situation is further complicated by the fact that there is no comparison of models based on use cases for business. We only have technical data on what each model does, but there is no comparison for reading documents or long-term stability to ensure we continue to receive consistent outputs in all languages after a few months.
Ultimately, it only depends on how complex the task is and what quality of results the company expects. While we cannot replace an expert with 20 years of experience, we can provide them with an assistant. I like to compare AI to a trained monkey, which will never be a virtuoso, but if we give it a hurdy-gurdy (a tool that expands its capabilities), it can lead to a nice performance. Moreover, people who know how to lead others usually work well with artificial intelligence. To enable a company to work with it, we therefore use proof of concept (PoC). For example, a customer wants to use artificial intelligence to ease HR tasks for summarizing and sorting resumes. They need to navigate through all of this because they cannot orient themselves in the multitude of products and solutions. They may have three or four options to choose from and cannot select the one that will lead to the best outcome. PoC will tell us whether the client will be able to use the solution effectively and whether we can train it to deliver the desired results.
I personally believe that a company should set the right mindset from the very beginning. The key to AI delivering savings or profit does not lie in the technology itself. It is essential to connect with someone who understands both the possibilities of AI and the business. Does it make sense to invest in this technology? What problem will AI solve for me? And what savings or profit might it potentially bring? Implementing artificial intelligence should be considered in the same way as deploying new software. It is also necessary to work with the team – to clearly explain why the technology is being introduced and to demonstrate its practical benefits. Above all, it is important to alleviate their fears that it might replace them in their work.
We adopt exactly this approach. When a client approaches us, we utilize proof of concept. The client needs someone to guide them through a vast array of solutions and show them which one will best assist them in their situation. Comparing individual models, testing them, and resolving dilemmas that the company would otherwise face alone. What follows from this? There is no comprehensive comparison of individual models based on use cases for a given business. Therefore, it is necessary to start at the core of the company, derive from its internal processes, test possible variants, and only then attempt the first implementation. And when one of the variants serves its purpose, adapt it correctly across the company.
It is crucial to understand AI as a tool for building business, not as a cool technological toy. The future in this regard, in my opinion, belongs to those who can use artificial intelligence strategically and meaningfully. Not to those who blindly follow fashion trends.
This article is published with the consent of the author Petr Kameník (COO PUXdesign) and the editorial team of the AICrunch portal, where it was originally published on March 30, 2025. You can find the original version here.
