Stealth (AI)
AI dashboard design
The fashion industry is built on gambling with demand planners placing orders months ahead. I joined a pre-seed AI start-up for 4-months to transform complex data layers into intuitive insights for its 40+ year old user, working with engineers and users to lead sprints and build MVPs of AI dashboard to onboard customers worth 1M ARR.
design engineer, dashboard design
Role
Collaborators
Design Engineer
Christopher Hugentobler, founder; Jimmy Chu, Engineer
duration
tools
4 months
Figma, Google Sheets
Design
UI and explanations honored mission of lightweight demand-planning
The new dashboard metrics transformed insights from passive to verifiable action items. This saved clients weeks of manual excel work to identify and act on supply chain shocks
Size-level drop-down eliminates unnecessary information at dashboard level but makes granular size-level details accessible as needed
Intuitive explanations of industry jargon on hover helps the user onboard onto the product and avoid misunderstandings
Problem
Risk became our north star
the function of risk was abnormal movement
f(risk) = abnormal*movement
Low actionability
Users had to manually navigate cluttered ui to determine risky products, slowing execution
Low verifiability
Users couldn’t verify AI prediction sources, and so are less likely to adopt insights, reducing latency.
By creating a user journey, we isolated the key pain to be manual monitoring risks
We narrowed down contextual data to manage and display in a logic map
Process
Working questions for refining data-viz
How to best distill complex size, time and inventory-level data at first glance?
which information is most essential to display first?
How do we streamline onboarding to prevent low user adoption?
Self-imposed Constraints
Keeping data unbiased by eliminating convoluted, unclearly structured data, or dark patterns early
Standardized taxonomy - This builds trust and buy-in from demand planners by embedding familiar concepts
Scalable mvp that is buildable across rows and columns
"At-a-glance" insights to reduce mental load
Allow for direct feedback for the design and forecast accuracy
Breaking task down
Widgets Versus Tables
We had to accurately forecast, identify and assign actions for products tied to risk levels. Over-engineered widgets and charts caused unnecessary user context-switching. A fully integrated dashboard enabled scalable columnar data, row-by-row product comparison, and a seamless user flow. So we abandoned the widget concept, but it informed the final time-series chart design
Data across time scales
To visualize abnormality over time, I experimented with various time series charts. This informed the product feed. By asking the customers to explain the chart to us and how they reached their conclusion, we were able to narrow down visualizations directions most intuitive to regularly scan.
Development
Halfway through development, the team collectively decided to launch the final design as a responsive google sheets MVP. I adapted the Figma design accordingly using spark-lines. This enabled the client to onboard directly onto our data, and integrate it their google-sheet sheet-heavy work-flow while reducing front-end development.
I worked closely with the engineer to design this google sheet for data to programmatically update into.
Impact
After shipping the redesign, we were able to increase engagement amongst existing customers and engage clients worth over 1M ARR
50% of existing customers found this intuitively easy to use without an onboarding
50% of existing customers included this in their weekly SIOP meetings
Reduce time to act (from weeks to days) by 40% through actionable CTAs
Cut support and onboarding queries by 25% (simplified, granular workflow)
Qualified 2 trial customers to consider premium subscription worth 1M ARR for demand planning needs
Soft signs of traction
Users readily adopted alongside existing system
Users pulled up the sheet during board meetings
Learnings
This contractual project was a rare chance to collaborate closely with product, engineering and users to ship live changes via weekly sprints.
“Yes. This is well-designed. I don’t even know how you did this.”
Christopher Hugentobler (founder) on seeing Google sheet dashboard
No such thing as too many variations
I committed to ideating a new component daily. Through rigorous comparison, we identified why certain hierarchies or components outperform others. This iterative approach also helped me detach from early ideas, ensuring higher quality design.
Mindful Design
Tools that design for manual monitoring is the worst waste of human attention. When something designed once ends up used a thousand times over, we should be mindful of how a product is used and the deeper implications for users. When exploring forecast charts, my main motivation was Josh, a 40+ year old demand planning veteran. Our dashboard saved him hours of staring at excel sheets every week.
AI ethics
Predication accuracy metrics encourage prevent overlying on AI-generated insights to inform multi-million dollar purchase decisions. I enjoyed refining accuracy metrics with the users to reach a more verifiable dashboard.
