AI Guidelines
Designing with AI starts by asking the right questions. Spark’s approach emphasizes trust, clarity, and purpose—ensuring AI is used where it adds value and can achieve success.
Purpose and Scope of Conversational and Generative AI
If you’re considering a conversational AI or a chat UI, ask yourself why. Chat AIs can be a good solution, but they’re not always the best solution for every problem. Consider the following learned from recent user testing and current AI projects.
Ensure Data Quality
Data quality plays a key role in the users’ experience. Product teams need to be clear on their goals and make sure the data supports those goals.
- Know your customers’ top issues to ensure the data includes solutions to key issues.
- Make sure all top, most-asked questions were topics in the bot.
- Review all data and evaluate:
- If it is needed for the bot
- What data can be combined and/or separated
- What prompts are needed for the data
- What content should be provided directly in the bot and what content should be linked
- What is the escalation or end of conversation look like based on different user groups
- Ensure data sources match your project goals rather than matching your goals to the data available.
- Fine tune data as needed to pick up on key words and acronyms that are specific to your users.
Match the AI Functions to Its Intended Usage
The kinds of questions users ask and the information they need will determine how AI can be most effectively used. What does your user need to do?
- Analyze or ask questions of the data? Rely on AI to connect dots that are not easily made visible. Examples: revenue analysis
- Retrieve info from multiple sources? Leverage AI to retrieve hard to find information in nicely formatted, customized results. Examples: reports, support chats, booking results
- Navigate complex, non-linear systems? Allow AI to eliminate the need for users to navigate the system to find relevant information. Examples: Profile retrieval based on context cues from other data, support chats, automation.
Measure and Optimize
Define key metrics for success such as user satisfaction, resolution time, engagement, abandonment and escalation rates and plan to update the AI accordingly.
UI Guidelines
Building Trust
While AI applications benefit from today’s customary UI best practices, it’s also important to remain transparent for users to successfully adopt AI and other new technologies. Accomplish this by providing users important information early in the experience:
- Clearly identify the system as AI (not human) up front to manage expectations and avoid confusion later in the conversation. Ex. “I’m Sabre’s AI chatbot. How can I help you?”
- Communicate if, how, and why information is collected; otherwise, users may lack trust and hesitate to engage with the system.
Humanizing AI
Not only do users engage more with AI when it has a human element, they expect it. In user testing, systems lacking human-like exchanges got associated with basic applications and search tools with limited or no AI capabilities, even when they had them.
Invest in Experience Design
Conversational AIs can be easily implemented, but that does not make them easy to use. Thoughtful design makes the experience smoother for users and should be factored into the project’s scope.
- Give time and resources to how search results appear. The goal of an AI chat is to provide data in a conversational, human-like way, not just to provide data or links. It’s essential to the product’s success that users can easily read and understand the results provided.
- Design human/AI interactions to help users refine the conversation for the smoothest, most “intelligent” experience. Design needs to be factored in early to ensure the AI API will support these interactions.
- Iterate and experiment collaboratively with all key players (product, design, development and the API architect).
AI Chat Assistants Design Only
AI-powered components and patterns are coming to Spark. In the meantime, use these recommended designs and resources to stay aligned with other Sabre products as you build your own.
View Spark's preliminary AI designs in Figma.
