Does your company have a Copilot or ChatGPT subscription that colleagues actually use, yet no results show?
Our colleague Nándor L. Szabó, Lead Developer and AI expert at Grepton, discussed this on the Business Class show of Jazzy Radio FM90.9. The starting point is uncomfortable but honest: a significant share of corporate AI pilots never get past the trial stage. The tools are already on people’s computers, yet measurable financial results are often missing.
AI is not the weak point
Many people treat large language models as a ready-made tool: switch it on and it works. In practice, it’s different. For the vast majority of corporate AI projects there is no demonstrable financial return – in Nándor’s example, 19 out of 20 projects. The tools are already in place; the reason for the stall is organizational: there is no owner, and the goal is vague.
Rolling out subscription tools doesn’t solve anything on its own. Enthusiasm lasts for months, then fades if the company doesn’t deliberately build its AI maturity.
Without training, shadow AI grows
According to a survey by the Hungarian Central Statistical Office (KSH), most Hungarian employees – eight out of ten – have not yet taken part in any AI training at work. At the same time, roughly one in three already uses some AI tool, often outside any company framework. This is shadow AI: data leaves the building, and control slips into a gray zone.
A successful pilot: a concrete goal, an owner, a deadline
An AI pilot counts as a success if its goal is written down at the start, in both forints (HUF) and working hours. The good feeling that work has become faster is not enough for a balance sheet.
That’s why it’s worth starting like this:
- Pick one small pain point; don’t try to transform the whole company.
- Appoint an owner.
- Set a deadline of two to three months.
- Work with a small team.
If the first round delivers measurable results, the mindset spreads on its own, and the next project can follow.
No data, no AI
The second big point of failure is data. A company may have the manual, tedious task, yet the necessary data never reaches the model. Just because a company is full of files doesn’t mean they are readable for AI and legally accessible. Compatibility and permissions can be as much of a bottleneck as the software itself.
The antidote to fear is information
Employees’ concerns often stem from a lack of information. Today, AI typically takes over tasks, rarely entire jobs, and the freed-up time goes to harder, more interesting work.
Leaders, meanwhile, carry a dual responsibility: they must set an example and know exactly what they have ordered. At a chatbot demo, an unprepared leader asking unrealistic questions can even discredit a solution that works well internally.
Time saved is not yet profit
Testing, verification, correction and coordination are invisible work, which is why companies often underprice the pilot’s budget. Models can hallucinate, so human approval is needed at the end of the process. For solutions built on company documents, an effectiveness of 95–98 percent can even be written into the contract, with links pointing to the sources.
There's much more to hear
In the show, Nándor shared further examples, stories and practical advice on how to turn a first AI project into real, everyday operations. If you’re curious, watch the video or listen back to the Business Class episode of Jazzy Radio.
Not sure where to start?
With Grepton’s business AI solutions and corporate consulting, we help you find where AI adds real value for your company. The maturity questionnaires on our website can help you get started. https://grepton.hu/en/artificial-intlelligence-business-ai-solutions-consulting/
