Best practices for designing and optimizing services for BMC HelixGPT
To ensure services are both discoverable and usable through BMC HelixGPT, we recommend that you design services with conversational AI in mind. Services designed for visual interfaces might not translate well to AI-based conversational experiences. Traditionally, services that are meant to be requested via the application UI greatly rely on structured forms and questions with predefined logic. To guide users through a decision path, they ask a broad category question like Software or Hardware, then narrow down to options like Laptop or Mobile.
In contrast, BMC HelixGPT aims to understand the user's intent with fewer questions. Therefore, creating simple and focused services can enhance usability and user experience in BMC HelixGPT.
Best practices for different user roles
Different user roles contribute to optimizing services for BMC HelixGPT. The following table outlines best practices for each role:
| Role | Goal | Best practices | References |
|---|---|---|---|
| BMC Digital Workplace Catalog administrator | Define and structure services to ensure that they suit well for BMC HelixGPT. These best practices apply to services made available in chat by using either the Publish wizard or the Catalog Request Agent. | Create smaller, purpose-driven services that are easier to prompt and maintain. Avoid bundling multiple requests, such as a laptop, Wi-Fi, and accessories, into a single catalog item. Bundled services make it difficult for BMC HelixGPT to determine the user's exact intent. | Designing service catalog - best practices |
Create dedicated services for each specific use case. Avoid creating services that cover multiple scenarios so that risk of misinterpretation is reduced. | |||
Break down complex conditional paths into multiple, simpler services that are easier to design, test, and maintain. Avoid using complex conditional logic in your services. Conditional questions that change based on user input can be difficult to represent in conversational flows. | |||
Use simple and intuitive input methods that support a natural dialogue flow. Avoid using forms that rely heavily on drop-down fields because they are designed for visual interfaces and might not translate naturally into chat conversation. | |||
Use clear and precise titles for catalog services. Ensure each service title is descriptive and unambiguous. Descriptive titles help BMC HelixGPT to quickly identify the requested service. | |||
Create unique titles for your services to improve accuracy in recognizing the user's intent. Avoid similar service titles. If multiple services have similar names, BMC HelixGPT might confuse them. | |||
Aim for clear, user-friendly language that is easily understood by a broad audience and is easier for BMC HelixGPT to interpret. Avoid using overly technical terms or internal jargon in service descriptions. | |||
| Limit the number of questions because too many questions can affect chat performance and user experience. Keep questions concise and relevant to the request. | Adding service questionnaires - best practices | ||
When you create a questionnaire, consider the following question types that BMC HelixGPT doesn't support in chat:
Additionally, consider chat limitations related to the following fields:
Important: This limitation applies specifically to services imported into BMC HelixGPT by using the Publish wizard. (Controlled availability) For details on unsupported question types when using the Catalog Request Agent, see Using BMC HelixGPT to make services requestable via chat. | Importing chat-enabled services from BMC Helix Digital Workplace Advanced into BMC HelixGPT | ||
| Application business analysts in BMC Helix Virtual Agent and BMC HelixGPT administrators | Enhance the discoverability and usability of services in BMC HelixGPT. These best practices apply only to importing services into BMC HelixGPT by using the Publish wizard. | When publishing a catalog service to BMC HelixGPT, include multiple ways users might phrase their request. This approach improves BMC HelixGPT’s ability to recognize and match the user's intent.
Important: Even though variations enhance the intent recognition, if you define too many similar variations for different services, BMC HelixGPT might confuse them. |
|
Define a clarification phrase to be displayed when BMC HelixGPT cannot determine the user’s intent.
| |||
Enter a simplified form of the service question to suit your business needs. Question labels that are understandable in a visual interface might not be as clear in a conversational chat. Rephrasing questions for chat ensures that they are easier for users to understand during a conversation.
|
| ||
| BMC HelixGPT administrator | Test and customize prompts to improve how BMC HelixGPT understands and responds to user queries. These best practices apply only to services imported into BMC HelixGPT by using the Publish wizard. | Review how BMC HelixGPT generates prompts from catalog services to understand how BMC HelixGPT interacts with users. | |
| Review and refine prompts that are generated from catalog services to ensure they reflect natural language and are user-friendly while meeting your organization's needs. However, you must not modify the variables, tags, and the JSON input. | |||
| Use the skill testing capability to simulate conversations, debug issues, and refine prompts to achieve the desired results. | The To test a skill procedure in Creating and managing skills. | ||
| BMC HelixGPT administrator | (Controlled availability) Modify the settings and prompt of the Catalog Request Agent. | Modify the default prompt of the Catalog Request Agent. In addition, modify the number of services, questions, and probable answers that the Catalog Request Agent provides to the end users. These changes help tailor the user experience, improve response accuracy, and enhance performance. | Using BMC HelixGPT to make services requestable via chat |
| End user | Request services through BMC HelixGPT. | Select values from the lookup or drop-down fields rather than entering them manually. When users respond to questions in a chat, the Lookup and Drop-down question types support the following input methods:
| Not applicable |
