What can AI do in an enterprise helpdesk today, where are its limits, what is worth entrusting to it, and how can data protection be handled? IT customer support is one of the first areas where companies want to introduce AI. Vendor promises are big, while day-to-day experience is more nuanced. Our colleague, László Tóth, Director of Grepton’s Infrastructure Division, spoke about the topic on Jazzy Radio.
How does AI work on the helpdesk?
In enterprise customer support, AI is the first filter. It receives the incoming ticket, interprets what the user wants, and decides how to proceed based on the nature of the problem:
- it resolves routine, well-known cases on its own,
- it routes more complex ones to the right engineering team,
- so the specialist receives a ticket that is already interpreted and organized.
A password reset is a good example. AI never reveals the password, because only the user can know it. It walks the user through the steps, or starts the reset process. The requester simply notices that their problem was solved faster.
How much does AI resolve without human intervention?
Vendors often quote optimistic ratios, while in practice the picture is more modest. Still, in a significant share of cases no human is needed: “in forty-five cases out of a hundred, a person doesn’t need to do anything.” This is not the entire volume, but it is a noticeable relief, because AI does most of the preliminary processing, even for tickets where a human makes the final call.
Where are the limits of AI?
AI’s knowledge is uneven. It is strong in areas where it has learned from a lot of material: in software development, for example, it performs excellently because it has seen an enormous amount of source code and documentation. Assistants are trained to answer as many questions as possible, so in complex situations their answers should be treated with caution. Artificial general intelligence (AGI) is still far off; today, AI works reliably in specific areas.
What is worth entrusting to AI?
The most important rule is to entrust AI only with decisions whose workings you understand and that can be reversed if necessary. A few practical principles go with this:
- AI decisions can also be checked by another AI, and these checks can be automated.
- A human always stands at the end of the process, and responsibility and control stay with them.
- Excessive expectation is the biggest risk. AI is not a miracle cure, and anyone who expects miracles falls for the hype.
- Humans make mistakes, and so does AI. You can prepare for mistakes deliberately if the system is designed for it.
No results without investment
A well-configured AI system already delivers measurable benefits for large organizations today, but it takes time and expertise. After rollout, the system has to be trained and fine-tuned continuously. Grepton also helps with AI adoption in enterprise and government environments.
AI takes over routine, repetitive tasks. Specialists can then focus on problems that call for real detective work: mysterious faults, unusual correlations, and issues that span complex systems. Customer support does not become human-free because of this; human time simply goes where it matters most.
Data protection: where does the data go?
The data handled in IT customer support is often sensitive, so the path the data takes is a key question. Helpdesks typically use the leading language models, such as ChatGPT, Claude, Grok or Gemini. Several of them also offer European availability, so the data can stay within the EU.
The other option is local AI: the model runs on the organization’s own infrastructure, and no data leaves the company. This is especially worth serious consideration for particularly sensitive data or in regulated industries.
Summary
AI already provides real relief for IT customer support today: it resolves routine tickets and hands more complex ones to specialists with the groundwork done. Entrust it only with decisions whose workings you understand and that can be reversed, and always keep a human at the end of the process. Good results take time and continuous learning, and it is worth settling the data protection question at the start, that is, whether to choose a European provider or local AI.
Grepton on-prem AI
Would you like AI to work securely with your own data?
Grepton's on-prem AI solution runs on your own servers, from hardware to operations, with a single partner. In a free online consultation, we select the 2–3 most valuable AI tasks and estimate the capacity you need.