In today’s ITSM systems, more and more processes are being automated for ticket processing, proactive error detection, incident management or in the form of virtual assistants. Helpers in the background more and more often: AI.
by Pierre Cordes, Solution Expert at handz.on GmbH
Artificial intelligence can be divided into two main areas of application in the ITSM context: Predictive Machine Learning on the one hand for the automation of ticket categorization, routing and major incident detection, Generative AI (GenAI) on the other hand for the generation of solution texts or summaries of logbook entries of a ticket. As advantageous as both approaches are, the requirements and challenges associated with them are very different.
Predictive machine learningAutomation and efficiency
Predictive machine learning is mainly used to control ticket processes. This includes the automatic categorization and routing of tickets. This artificial intelligence – often referred to as „classic“ – focuses on the automation of processes using AI-supported algorithms. The aim is to automate recurring tasks, such as automatically categorizing and prioritizing tickets, thereby increasing the efficiency of the IT department.
For example, if someone reports a problem with the printer, the ticket is automatically assigned to the correct category (e.g. hardware, printer) thanks to AI and forwarded to the relevant support group. This not only reduces the processing time, but also minimizes the risk of tickets ending up in the wrong department and being unnecessarily shuffled back and forth.
Another example is the detection of major incidents. If several employees report similar problems within a short period of time, the AI recognizes this and automatically combines the individual tickets into a major incident. This makes processing easier and speeds up the process of finding a solution. If a NAS system is unavailable, for example, the IT department would otherwise receive dozens, perhaps hundreds of individual tickets, each with exactly the same problem. Instead, the admins can concentrate on the major incident and resolve all the associated tickets at the same time. This saves an enormous amount of time and human resources.
The key advantage of predictive machine learning is that it can be used on-premises and is therefore GDPR-compliant, as no data leaves the company. The technology uses existing structured data and therefore a controllable learning space. It offers high-speed decision support, but is also heavily dependent on the quality and quantity of existing data. Once a model has been learned, it also requires continuous maintenance and retraining processes. And it is usually a „black box“ for people without in-depth data science knowledge.
Generative AI: text generation and data analysis
Generative AI (GenAI) goes one step further. It uses AI to generate solution texts, summarize logbook entries and extract knowledge. Instead of manually searching through long log files, GenAI can extract the most important information and reproduce it in just a few sentences. This opens up completely new possibilities in ITSM. Tickets are not only categorized, but their content is also processed. Based on the error description and the existing ticket data, the AI generates a proposed solution, which is made available directly to the IT department. In some cases, the AI can even resolve the ticket completely automatically.
Generative AI solutions can also be used to create knowledge articles. Based on the existing ticket data, they automatically generate articles for the knowledge database, which help IT specialists and users alike to solve problems quickly and independently. GenAI also results in the familiar effects in terms of time savings and faster problem solving. IT specialists can concentrate on the essential aspects of a problem instead of getting stuck in time-consuming routines and manual work.
What remains with generative AI is the well-known Achilles‘ heel of hallucination, which is why human validation of the results is always necessary. It is also important to prepare employees for the use of AI and to provide them with the necessary knowledge and rules for its use (legal and ethical). This is the only way they can fully exploit the potential of the new technology and minimize risks.
Data quality as a success factor
Regardless of which AI approach is used, success depends largely on the quality of the input data. An inadequate ticket description makes analysis difficult and leads to inaccurate results. It is therefore important to ensure that the ticket data is complete and precise. Only then can the AI correctly categorize the tickets, generate the right solution suggestions and fill the knowledge database with relevant information. So not „I have a problem“, but „I have been unable to use printer Z on my PC in office X since time Y“. The more precise and comprehensive the information, the more input is available for the AI to process.
Data protection and local solutions
Another important aspect is data protection. Predictive machine learning can be operated entirely locally, which means that the data does not leave the company. The key difference with GenAI is that the data is generally sent to external cloud services, which automatically raises data protection concerns. This means that all data that leaves the company for text generation must be agreed and approved in advance by the company’s information security officer (ISO).
Privacy-preserving AI concepts such as federated learning, in which an AI is trained decentrally on several end devices or servers – without the raw data leaving these devices – also provide a remedy. Instead, only the model updates are sent to a central server, which aggregates them into a shared model. However, operating such a so-called large language model (LLM) on your own servers requires additional resources (time, costs and computing capacity).
Step-by-step implementation
How can companies successfully shape the path to AI-supported ITSM? An important step is choosing the right ITSM system. Many modern systems already offer integrated AI functions or can be easily expanded with AI solutions. The analysis of existing data is just as critical to success. Companies should check what data is available, how good the data quality is and what data is required for the use of AI.
Once these steps have been taken, you can start implementing the AI solution. It is advisable to start with predictive machine learning, as this can be operated locally and in compliance with data protection regulations. In the first stage, this improves internal processes and gets users on board with the new way of working. Once this has been successfully completed and sufficient user acceptance has been achieved, the use of LLM services can be considered in the next step, provided that the company allows data to be shared.
Combined use and future
As sensible as a gradual introduction is, a combined, i.e. hybrid, approach should prove to be a real booster for ITSM. Predictive AI is used in the background for structuring, while generative AI is used in the front end for user interaction or documentation. Thinking further into the future, scenarios of domain-specific training are conceivable in which own service data is used for LLM training. Data protection hurdles can also ultimately be overcome through privacy-preserving AI concepts.
