.png){kind=small}
Sponsored Project
Summary
I facilitated meetings with Salesforce sponsors, and led research activities on a semester-long student project and designed 3 features using Slack design system to address data privacy and trust challenges in Slack AI integrations, particularly in the educational context.
Impact
Trust Score increased by 40%, measured by post testing surveys.
My Contribution
Facilitated Meetings
Designed Hi-Fi Screens
Led Research
Prepared Documentation
Team
8 UX Designers
2 Senior UX Sponsors from Salesforce
Timeline
5 Months
Tools Used
Figma
Notion
Zoom
Google Docs
Our 3 solutions built trust in Slack AI for the education industry by addressing student concerns around data privacy and transparency.
Screenshot from Salesforce website
We were given a broad and ambiguious problem statement "Build Trust With Data Privacy For Salesforce AI Tools".
We explored all of Salesforces tool and realized we had to scope down.
All tools by Salesforce
To scope down we did some stakeholder analysis we decided to focus on educational context as Salesforce had not explored much into it and we had access to students for primary research.
Conducted with graduate students to understand how they use slack and their concerns about AI and data privacy.
Analyzed how different AI tools provide data privacy and build trust with users about their data. We learned that most companies follow GDPR and CCPA guidelines and provide documentation on how their AI is trained.
Explored ethical AI practices and performed affinity mapping for themes that emerged from literature review, interviews and competitor analysis. We made our own definition of trust for our context.
Realized we did not have access to Slack AI and it was not publicly available. So we decided to explore the speculative methods.
.png)
We realized trust had many aspects to it and was more complex and interdependent on many factors. So we chose to define it for our context.
"Trust is the Calibrated Willingness of users to depend on AI capabilities"
Data Security (FERPA/GDPR/PII) : Confidence that student records and personal data are protected.
Contextual Accuracy : Confidence in interpreting human tone, sarcasm, and cultural nuances.
System Transparency : Confidence in explanations of AI mechanics, limitations, and zero dark patterns.
Human Autonomy & Control : Confidence in meaningful consent, opt-out control, and feedback loops.
Psychological Safety & Equity : Prevention of biased profiling or punitive monitoring of user behaviors.
Brainstormed with a group of designers about the worst case and best case scenarios for the use of Slack AI.
This helps teams spot dangers before a product is even built. It faces a simple truth: people can find ways to misuse any new technology, especially AI. By imagining the absolute worst-case scenario from day one, companies can build safety and privacy features directly into the core of the product. This way, they protect users right from the start, rather than scrambling to fix a disaster later.
From the black mirror brainstorming we focused on 3 scenarios which we thougth were most improtant. We explore 2 negative and 1 positive case and scoped to design for those.
All 8 of us made sketches for 3 scenarios to ideate and come up with concepts.
We narrowed down to 3 main features. The first one being in-line privacy alert which would alert the user while they were typing the messages to make them aware before they get flagged.
.png)
Explanations for AI flagging decisions where it gives users guidance as soon as a message is sent.
.png)
Having transparency by showing users how the AI is trained when their content gets flagged and shows where they detected the flagged content.
.png)
To test how users would behave we used our solutions with users by pretending to be the AI. We also used paper prototypes and the actual slack channels.
Users mentioned something interesting about how different channels have different tones.
We realized consistency in tones reduces uncertainty in AI responses, directly improving perceived predictability and user trust.
We built a strong collaboration with sponsors through weekly check-in calls, where they provided valuable critiques and insights on our designs and suggested a more professional look and replacing emojis with a more defined scale.
We refined our designs using the Slack Design system and addressed feedback on each of the concepts to address the privacy concerns.
Users can enable/disable alerts for sensitive topics and control the topics and contexts they would want the alerts for.
AI explains flagged content + allows user appeals giving them more control over their data and the AI.
Users set channel tone (e.g., professional, relaxed) for better AI context understanding.
We tested our users with graduate students with a situation once without our prototypes and, once with our prototypes. Final prototypes resulted in a 40% improvement in user trust metric scores, as measured through post-testing surveys.
We presented our slide deck to the sponsors and prepared a comprehensive documentation for all of our research inisights.
Planned for future steps to which Salesforce might take to enhance this product.
.jpg)
Building trust was one of the biggest lessons from this project. Success depended not only on creating a reliable experience for users but also on building strong relationships with sponsors and stakeholders. Regular communication and transparency helped establish credibility, leading to more honest feedback and a better final product.
I would start research scoping and usability testing earlier and continue it throughout the project instead of concentrating it in later stages. Earlier testing with users and sponsors would have surfaced adoption and workflow issues sooner, allowing faster iteration and clearer prioritisation.
If possible, I would have also involved engineers and product managers more directly in the design process through co-design sessions and more collaborative and inclusive research, to better align feasibility, product direction, and user experience from the start.
Big thanks to all my teammates, our sponsors from Salesforce and our professor for supporting us through this project.
.png)
.png)