Transparency builds trust in AI-driven products, and AI Notice helps by informing users when content is AI-generated, encouraging them to verify its accuracy. As a sole designer, I was responsible to conceptualise the design pattern and successfully incorporate it in the FIORI design guidelines which can be used globally across within SAP. In the following case study I will take you through behind the scenes of drafting the design guidelines. Hope you enjoy it.
UX Designer — Research, Conversation Design, Interaction, Prototyping & Testing
8 Months
Figma
Why even do it?
In 2023 AI was one of the fast pace topics and every other industry was trying to induce or use AI capabilities in to their products. The story below covers how we got the opportunity to be a part of developing AI Design systems.
The Problem
Transparent AI labeling ensures trust, reduces misinformation risks, and improves user experience. Without it, legal, ethical, and operational risks can significantly impact businesses and users alike.
of users struggle to differentiate AI-generated from human-created content. (MIT Study on AI trust)
of professionals report difficulty trusting AI-generated reports without explicit labeling. (Gartner 2024)
in lawsuits filed against AI-generated deepfakes and misinformation in 2023. (AI Policy Institute)
of users prefer knowing if they are interacting with AI vs. a human. (ServiceNow Research 2023)
How did we achieve it?
Designing AI Transparency for a Trustworthy Future
I embarked on this project with a clear challenge—how do we ensure transparency and trust when AI-generated content is seamlessly blending with human-created content? I've spent years designing for enterprise software at SAP, but this was a unique and pressing problem. The rapid evolution of AI meant that users were interacting with AI-generated reports, insights, and decisions without even realizing it. The implications? Misinformation, loss of trust, and potential regulatory risks.
How to design AI transparency?
When I started, I had no predefined framework for AI content labeling. What should an AI label look like? Where should it appear? How do we make it unobtrusive yet noticeable? I conducted intensive research on industry practices, regulatory guidelines, and user expectations to define the best way to inform users without overwhelming them.
How to balance transparency with usability?
It was exciting to realize that my work could set the foundation for AI transparency standards, but this also meant balancing user awareness with seamless AI integration. If the labeling was too subtle, users wouldn't notice it. If it was too aggressive, it could disrupt workflows. This led me to collaborate closely with AI teams, legal experts, and product managers to define a strategy that aligned with both user needs and business goals.
Design Process
Discover
- Problem Statement
- Business Objective
- Desk Research
- Secondary Research
- Competitive Analysis
Define
- Existing Patterns / Use Cases
- Design Explorations
- Affinity Mapping
- Use Cases
- Design Principles
- Scoping & Versioning
Design
- Design Concepts
- UA Review
- User Flow
- UI Design
- Preliminary Guidance
- Specification
Deliver
- Usability Testing
- Publish Design Guidelines
How about some desk research?
What is the level of awareness of AI Design guidelines across SAP?
Participants across Ariba/Procurement · BTP · Cloud ERP · CX · HXM · IES · SAP Business AI
| Region | Ethics | Responsible AI | Safe AI | Sustainable AI | Trust Gen AI | Prompts | Components |
|---|---|---|---|---|---|---|---|
| EMEA (11) | 82% | 46% | 46% | 46% | 36% | 46% | 55% |
| APJ (7) | 14% | 14% | 14% | 0% | 43% | 14% | 43% |
| AMER (5) | 60% | 60% | 40% | 40% | 40% | 40% | 40% |
Quick Summary of the Findings
Awareness
Overall awareness of AI guidelines is moderate to low. Awareness of RAI is low. Multiple sources for information on guidelines makes users overwhelmed and difficult to follow.
Content
Based on the varied set of users, the key need is to build understanding of how AI and RAI Guidelines can be consumed. Some users felt that the guidelines have the format of a wiki and are wordy.
Misc
Industry is moving faster than some of the topics or references mentioned in the guidelines. Joule button in the toolbar is missed by 90% of users.
Key Need
The information regarding guidelines seems overwhelming with no single place. A unified, accessible, tiered approach to AI design guidelines is critical.
Competitive Study
Microsoft Copilot
Microsoft's Copilot uses notice as a footnote followed by the AI feedback buttons.
Microsoft PowerPoint
Even sometimes it comes along with the title of the body assuring that users know it is generated using AI.
Instagram now labels photos if user uses generative fill. Content detected to have AI signals will receive an AI label.
AI-Notice User Flow
This notice pattern helps in creating transparency and assists the user in identifying AI-generated content.
Problem Statement
To design an AI notice for SAP which would be able to address the content generated using AI that needs to be communicated to the user in a transparent manner.
The MVP
The MVP should make it clear when AI is involved without getting in the way. Simple, well-placed AI notices should inform without disrupting workflows — starting with the most common AI-generated content surfaces in SAP Fiori.
Who are the users?
Users interacting with AI-Generated content/outputs and workflows.
Business Domain Expert
Product Manager
SAP Customer
User Assistance
UX Designer
Accuracy & Reliability
Concern over the accuracy and reliability of AI-generated reports or data insights.
Sensitive Business Data
Worry about how AI tools handle sensitive business data, leading to hesitance in adoption.
Difficulty in Integration
Difficulty understanding how to best integrate Gen AI into business processes due to limited technical guidance.
Privacy & Fairness
Worries about the ethical implications of AI, including bias, privacy, and fairness.
"As a user I want some assistance that simplifies overall understanding of the AI generated content so that I can make informed decisions."
Design Iterations
Key issues discovered and iterated upon during the design exploration.
Chicken Pox Syndrome
The design itself was creating a lot of noise by repeating the same sparkle icon and recommendation elements, causing visual overload.
Component Clashing
It was a message strip which is an existing component within the FIORI design system and is mostly used for error/warning states — conflicting with the AI notice intent.
AI Footnote & Table Version
Whenever there is a text area or RTE used, AI notice can be represented by using footnotes — subtle, accessible, non-disruptive, and consistent with existing Fiori patterns.
Final Implementation
The final AI Notice pattern in context — clearly communicating AI involvement at the right level without disrupting the user's primary workflow.
With AI Notice in place
✅ AI notices are increasing awareness & trust. ✅ Users are more likely to verify AI content before acting on it.
Content Recognition
~60% of users correctly understood that content was generated or modified by AI when AI Notice was in place.
Trust & Perception
55% of users trusted AI responses with proper labeling.
Compliance Rate
Facilitated workshops and trained teams, achieving 95% compliance rate to new design standards.
Design Standards
Improved adherence to design standards across teams globally.
What all did I tackle...
How I prepared for the project
I worked with several stakeholders including Product Managers, Product Owners, AI Experts, User Assistance writers, and Legal teams to build a holistic understanding of what "AI transparency" means across different contexts within SAP.
How I learned the product landscape
Since the AI topic was already taking traction and it was getting adopted everywhere, this fellowship gave me the opportunity to deep dive into the product landscape and understand where AI content was surfacing — and where it wasn't being labeled at all.
How I learned about designing guidelines
I have always worked on designing applications using design guidelines, but here it was all about designing the guidelines themselves. Learning to write prescriptive guidance that is both flexible enough for varied contexts and specific enough to be actionable was a new muscle entirely.
What all things went into the collaboration
While it was a whole team involved in crafting the design guidelines, the collaboration spanned legal review cycles, accessibility audits, internal dogfooding sessions, and coordination with the Fiori design system core team to ensure the pattern was ratified and published globally.
How I negotiated design decisions
Every design decision had to be backed by research, legal requirements, or competitive precedent. Negotiating simplicity vs. completeness, and prominence vs. subtlety, required constant alignment with stakeholders who had very different priors on what "transparent" means to their users.
Some appreciation from the team
