Problem
During the development 💻 of new features, we often need some reasonable data 📝 to test and verify the implementation. The data in the development or test environment is usually not as rich as in the production or not available at all. The standard solution for that is eighter create sample data manually which is time-consuming or generate some dummy data with a script. Usually, we take the latter option. The disadvantage 🔒 is that the quality of that data is not high. It looks something like “My test Ticket 01”, “My test Ticket 02” etc.
Solution
The idea is to let some AI 🤖 algorithm generate the Tickets with useful titles and descriptions. We decided to try GPT for that but we are evaluating other models as well.
We implemented a prototype of this solution in USU Service Management and tried different scenarios with various settings. The results were more than satisfying ✅. We are now able to generate large amounts of “good-looking” tickets with just one click.
How it works
The Demo Tickets Generator can be started via the SIT Generate Demo Tickets by AI smart tile. It opens a dialog for setting up the parameters.
You can configure these parameters:
Language: the Tickets can be generated in various languages 💬. We tested it with English, German, French, and Czech. The results were very good in each language
Maximal Number of Tickets: this is just a safety check so not too many Tickets 💯 and API requests will be generated
Number of Tickets per Category: for 💪 each selected Ticket Category the specified number of Tickets will be created
Ticket Class: Ticket class 📊 of generated Tickets
Minimal and Maximal Description Length: here you can specify how long should be the generated Ticket Description 📝
GI Action: Generci Interface Action
AI Model: model parameter for GPT API. The best result was achieved with the text-davinci-003 model.
Max Tokens Limit: parameter for the GTP API
Temperature: parameter for the GTP API
Ticket Categories: here you can select multiple Ticket Categories. For each category, the Workflow will create the desired amount of Tickets.
Once you confirm the dialog the workflow creates the tickets and contacts AI API to get the appropriate Ticket Title and Description.
Examples of generated Tickets
The results were quite convincing. The key setting is the context of the request. It can be configured in the Generic Interface Mapping. It can be fine-tuned accordingly for the specific project ✅.
Security
Security 🔒 is always the most important aspect in each scenario. In our use case, the only data 📊 which are sent to the GTP API are the master data (desired Ticket Class, Category, etc.) which do not contain any customer-specific information. From that perspective, it can be used without any concern.
Summary
This is just one example of how can AI improve our work in the ITSM area. We are investigating other interesting use cases which could be automated with AI algorithms.
If you are interested 🤝 in this solution or have any comments, please let us know.