In principle, AI can be used for any task, but this does not offer any real added value and increases both costs and system load. Cost-benefit considerations must always be a top priority. Every AI query consumes enormous amounts of computing power in the background, and external AI services incur costs.  External AI providers (ChatGPT, etc.) naturally charge for the use of AI. The current rule of thumb is: 1 token = 0.75 words. 

The most obvious use case is analyzing all job reports for anomalies and errors. This makes sense, since text analysis is one of AI’s strengths. However, this falls short in a cost-benefit analysis, as 99% of all analyses reveal no anomalies. A better strategy would be to search for keywords using the familiar Automic script tools as usual and then consult the AI when anomalies are found. If a centralized error management system is in place, AI can also provide added value here.

Error analysis can be implemented very easily using ASK_AI, since ASK_AI can access the job reports directly.

:SET &ANSWER# = ASK_AI(“Analyze the following report in German and check whether it contains an error message. Here is the text: {data}.{query}”, “JOB T.JS.UNIX.ERROR with Runid 1242003, analyze report type REP”, “If there is an error message, describe a possible solution.”)

Another use case is the integration of line-of-business applications via the “Model Context Protocol” (MCP). MCP is an industry standard that has been implemented by all modern solutions. Similar to web services, AI functions are made available, which the AI uses to respond to prompts. Each MCP interface provides a set of “tools” that can read information as well as modify it. For example, if the AI is supposed to create an ITSM ticket, the MCP must provide a corresponding “tool.”

I’ll cover the important topic of permissions and security in the following article.

Through the MCP interface, AI can utilize virtually all solutions used within the company. However, this also raises the question of whether AI capabilities are actually needed. ITSM solutions expect and deliver structured data. This data can be processed much more effectively and reliably using traditional integrations. If the flexibility of AI is not necessary for the use case, the traditional method should be chosen, as it is more cost-effective. Even though the AI (token) costs for a single job execution are not high, they add up significantly if that job is run dozens of times a day.

When deciding whether to implement a task using an AI job, it is important to consider how much variability is required in the call parameters. In most use cases, the parameters can be determined at runtime. It is clear which parameters are needed to create an ITSM ticket, and even for fields such as title and ticket description, text templates are the preferred solution.

The AI use case often cited as an example—“List all ITSM tickets of error type ‘XYZ’ from the last month”—can be handled more efficiently with a REST job. In this case, AI can be used to define the corresponding REST call. However, if an analysis of individual ticket categories—including classification and changes compared to the previous month—is required, then the AI job is the right choice.

AI jobs are also ideal for analyzing data from big data systems.

Using AI jobs and ASK_AI is undoubtedly useful, but these are very powerful—and also very expensive—tools.

Outlook