Problem
Schedulers needed AI support they could understand, question, and act on.
Mission schedulers were making high-stakes crew decisions across scattered data sources: training needs, deadlines, currency, flight hours, availability, and crew legality. The opportunity was not to automate the scheduler out of the workflow, but to help them see better options and understand the impact of each decision faster.
User context
Scheduling required constant cross-checking across disconnected tools
Problem analysis
The key problems clustered around trust, time, and decision confidence
Insight 01
Data lived in too many places
Schedulers had to reconcile multiple sources before knowing whether a crew member was actually a good fit.
Insight 02
Manual review slowed decisions
Availability, training deadlines, and flight-hour limits required constant cross-checking before assigning a role.
Insight 03
Errors could create downstream impact
A single assignment could affect another member’s currency, training opportunity, leave, or conflict status.
Insight 04
AI needed to explain itself
Users needed clear reasons behind recommendations so they could accept, override, or compare options with confidence.
Process
How the work moved from ambiguity to direction
Define the event rules and crew criteria
Show AI-recommended crew sets in context
Fit impact guidance into the scheduler workspace
Iteration 1
Using a stripped-down flow to validate the right information
The first iteration was intentionally closer to a wireframe: a stripped-down version of the flow used to validate what information schedulers needed before polishing the interaction details.
Step 1
Create event
The scheduler starts by creating the mission event.
Step 2
Input rules
They add role needs, timing, flight-hour constraints, and event requirements.
Step 3
Review AI sets
AI recommends optimal crew sets based on availability, flight hours, training deadlines, and who works best together.
Step 4
Understand impact
Impact details show what happens when certain people are not assigned, helping the scheduler decide faster.
Iteration 1 placed recommendations directly in the role assignment flow so schedulers could review fit and rationale in context.
Iteration 2
Making impacts more scannable without overwhelming the scheduler
Based on feedback from the first iteration, I refined the sidebar to complement the dropdown notifications. Each iteration focused on quick scanning and actions for the scheduler to take based on important information.
Iteration 2 paired the existing impact section with a clearer sidebar so users could scan affected members and issue types.
Gallery
Final design: AI support for faster scheduling decisions
Ultimately, I moved forward with the design users found most helpful for making quick scheduling decisions: clear recommendations, visible impact cues, and progressive disclosure for secondary details like person-specific impacts. Nice-to-have concepts, such as AI-suggested member replacement, were left out to keep the final experience focused and feasible.
The final design brought recommendations and impact details into one decision-support flow, while using progressive disclosure to keep secondary details available but less overwhelming.
Impact
Results and signals
Explainable
Recommendation flow
Recommendations moved from suggested outputs to visible decision support with rationale, constraints, and tradeoffs.
Human-led
AI interaction model
Schedulers stayed in control through comparison, override, and review patterns.
Scannable
Impact details
Sidebar and notification patterns made cautions easier to review without adding a separate workflow.
Aligned
Cross-functional language
The work gave design, product, data science, and scheduling SMEs a shared way to discuss AI trust and actionability.
Challenges
Constraints that shaped the design
Fitting AI into an already dense interface
The design had to preserve familiar scheduling patterns while adding recommendation rationale, impact details, and actions without making the page harder to scan.
Showing enough reasoning without overexplaining
Users needed to understand why the AI recommended someone, but too much detail could slow down a workflow that was already information-heavy.
Learnings & Reflection
AI features become useful when they fit the way people already work.
Balance familiarity with innovation.
I learned to introduce new AI patterns while preserving the visual language users already recognized.
Adapt to existing complexity.
The page was already dense, so the design had to improve clarity within real product and engineering constraints.
Use AI to reduce repetitive thinking.
This project showed how AI can process scattered data in real time and help users make faster, more informed decisions.