AI Design Quality Checklist
Foundations / AI and Joule Design / Overview / AI Design Quality Checklist
Intro
It is important to establish standards for the quality of SAP AI experiences. To get started, we are providing a first set of guiding questions and mandatory checkpoints to enable design teams to assess AI Design quality confidently and identify important follow-ups with technical experts. This AI Design Quality Checklist is also tied to the Design Intent Reviews and Design Excellence Reviews hosted by Chief Design Officer Arin Bhowmick.
Role of the Designer
Provides a human-centered understanding of the product by:
- Developing a user-needs strategy that connects stakeholder goals, end-user needs, and the product roadmap
- Conducting user research, mapping user flows, and running usability evaluations to generate evidence that refines the strategy.
- Designing and validating AI interactions (voice and conversational) based on research insights.
Methods
Our development process requires user research and design thinking. Write clear, well-defined user stories that capture goals, context, and acceptance criteria.
AI Design Quality Checklist Questions
1. Value
2. SAP AI standards
3. AI excellence
4. Generative AI and Joule embed
SAP AI Ethics UX Considerations
* Additions to item 2.4
SAP has committed to our customers and users to deliver only Business AI that is responsible. The AI Ethics UX considerations listed below serve as general guidance for designers and entire product teams on delivering responsible AI experiences. Overall, all teams at SAP working on AI use cases must adhere to the SAP AI Ethics Policy and follow the mandatory process, which includes, for example, the AI impact assessment for all AI use cases and the AI Ethics Steering Committee approval process for high-risk use cases.
Human-centered AI
- AI software is user-centric, addressing the widest possible range of applicable end-users, and following relevant accessibility standards, regardless of the users’ age, gender, abilities, or characteristics.
- Make AI systems identifiable as such to appropriate end users, where applicable, when interacting directly with humans (including via Conversational AI or ‘chatbots’).
- Develop AI systems in a way that it doesn’t encourage humans to develop users’ attachment and/or empathy towards the AI system. Clearly signal to end users that the social interaction of AI systems is simulated.
- Where feasible, apply a fairness function to test AI systems for unbiased output.
Human in control
- AI systems are subject to appropriate human oversight. The rights and freedoms of a human exceed that of AI systems.
- In situations where humans may be directly impacted by a decision made by SAP’s AI system, introduce human oversight to safeguard that the AI system doesn’t undermine human autonomy or introduce unintended consequences.
Transparency and explainability
- Provide a clear and simple explanation as to how decisions were made by an AI system used in automated decision processes, as far as is practical.
- AI systems that engage in profiling or automated decision-making must be able to provide explanations to data subjects upon request. They must be able to describe the data segment the subject was placed into and the reasons they were placed there. This is in alignment and compliance with applicable data protection and privacy law.
- In addition, the reasons as to why the decision was made must be provided if requested by the data subject. Present the reasoning in a way that the data subject can challenge it.