← Back to portfolio
Amadeus Retro
A real-time retrospective tool for teams to reflect, vote, and drive action — powered by AI assistance.
Why another retro tool?
Agile teams run retrospectives every sprint, but existing tools create friction at every step — from setup to follow-through. We saw an opportunity to build something faster, more focused, and smarter.
Retrospectives are one of the most valuable ceremonies in agile, yet they're often undermined by tooling problems. Teams lose 10+ minutes to account creation and login walls. Facilitators burn time manually grouping 30+ cards into themes. And the most critical output — action items — gets lost between sprints because no tool carries them forward.
Amadeus Retro was designed from the ground up to solve these three problems with a zero-friction join flow, AI-powered facilitation through Amadeus Max, and persistent action items that carry forward automatically.
Understanding the problem space
We interviewed 8 practitioners across 3 roles to identify the core pain points in retrospective workflows.
Key Research Findings
Our research surfaced five critical findings that shaped every design decision:
- Setup Friction — Teams reported losing 10 minutes to logins and account creation. New team members face the highest barrier.
- Action Item Loss — Only 23% of action items from previous retros were ever referenced again. The primary loss point was the gap between sessions.
- Facilitation Burden — Facilitators spend roughly 40% of session time manually synthesizing and grouping cards into themes.
- Psychological Safety — 5 of 8 participants admitted they self-censor when card authorship is visible to the group.
- Viewer Needs — Stakeholders need passive observation without influencing team candor, but existing tools force them to join as visible participants.
Role Analysis
We identified three distinct user roles with different needs, goals, and pain points:
Facilitator (Scrum Master / RTE) runs the session end-to-end. They need structured phases, AI assistance for synthesis, and the ability to complete a productive retro in 45 minutes or less. Their biggest pain points are manual card grouping and keeping discussion focused.
Participant (Team Member) contributes cards and votes. They need zero-friction joining (no accounts), anonymous posting options, and transparent voting. Their frustrations center on signup walls and meaningless votes that never lead to action.
Viewer (Stakeholder) observes without participating. They need passive, invisible observation and access to retro outcomes for tracking team health. Existing tools have no read-only mode — viewers must create accounts and appear in the participant list.
Competitive Analysis
| Tool | Setup | Real-time | Voting | Action Items | AI |
|---|---|---|---|---|---|
| Miro | Complex (account + board setup) | Yes | Limited (dot voting plugin) | None native | None |
| FunRetro / EasyRetro | Medium (account required) | Yes | Good (built-in) | Basic export only | None |
| Retrium | Medium (account + subscription) | Yes | Good (multiple formats) | Good (carry-forward) | None |
| Amadeus Retro | Simple (name + room code) | Yes | Emoji voting with budget | Persistent + carry-forward | Full (Amadeus Max) |
Key differentiators: Amadeus Retro is the only tool offering zero-friction join (no accounts), persistent action items with carry-forward, and integrated AI facilitation assistance.
Who we're designing for
Each persona was derived directly from research findings, ensuring every design decision traces back to real user needs.
Sofia Martinez
Scrum Master & RTE
Sofia leads a distributed Amadeus engineering team of 9. She facilitates bi-weekly retros and PI-level retrospectives, and values structured facilitation that keeps discussions productive.
Goals
- Run productive retros in 45 minutes or less with clear outcomes
- Generate actionable follow-up items that get tracked and completed
Frustrations
- Manually grouping and synthesizing 30+ cards is tedious and time-consuming
- Action items from previous retros are forgotten because existing tools don't carry them forward
“I need the retro to produce results, not just a nice feeling. If nothing changes between sprints, we're wasting everyone's time.”
Key Behaviors
- Prepares retro structure and timing before the session
- Uses AI tools when available to help synthesize and summarize
Raj Patel
Senior Backend Developer
Raj is a 4-year Amadeus veteran who participates in iteration retros every two weeks. He values candor but is reluctant to post critical feedback when his name is visible.
Goals
- Voice concerns without fear of judgment
- See clear action items assigned with accountability
Frustrations
- Most retro tools require account creation and passwords just to post a sticky note
- Votes feel meaningless when action items never get followed up
“I'll be honest if I know it won't come back to bite me. And I'll only bother if something actually changes.”
Key Behaviors
- Posts multiple cards quickly when joining is frictionless
- Votes strategically on items most likely to drive change
Elena Kowalski
Engineering Manager & Stakeholder
Elena oversees three Amadeus teams and joins retros as an observer to understand team sentiment without influencing the dynamic. She reviews action item progress across teams.
Goals
- Monitor team health and recurring themes without disrupting candor
- Track commitment follow-through across multiple retros
Frustrations
- Existing tools don't have a read-only mode — she must create an account and appear as a participant
- No cross-retro view of action item progress
“I want to listen, not influence. My presence shouldn't change what people feel safe saying.”
Key Behaviors
- Joins via read-only link without interaction
- Reviews action items and completion trends after sessions
The facilitator's emotional arc
Mapping Sofia's journey across all 7 retro phases revealed specific moments where AI intervention transforms the experience.
The journey map surfaces three critical AI intervention moments where Amadeus Max transforms the facilitator experience:
- Card Collection phase — Duplicate detection reduces cognitive overload when 30+ cards flood in simultaneously
- Discussion phase — Theme grouping and summary generation save ~15 minutes of manual synthesis effort
- Action Items phase — AI-suggested action items with recommended owners give the facilitator a concrete starting point instead of a blank slate
Structuring for three roles
The IA had to serve three distinct user roles with different access levels on a single shared board — a harder problem than it first appears.
The architecture organizes around 7 top-level areas: Landing/Join, Retro Board, Amadeus Max Panel, Action Items, Facilitator Controls, Viewer Mode, and Admin. Each area has granular role-based access:
| Area | Facilitator | Participant | Viewer | |------|------------|------------|--------| | Board / Cards | Full CRUD | Create/edit own + vote | Read-only | | Amadeus Max (Board-level) | Full access | No access | No access | | AI Rewrite (Card-level) | N/A | Own cards only | No access | | Action Items | Full CRUD + assign | View + own status | Read-only | | Facilitator Controls | Full access | No access | No access |
Entry points differ by role: Facilitators create rooms, Participants join via room code (no account needed), and Viewers access a read-only link with no join gate at all.
From structure to layout
Low-fidelity wireframes validated our layout assumptions before moving to high-fidelity design. Key decisions included equal column widths, fixed headers, and a horizontal facilitator toolbar.
Key wireframe annotations:
- Equal column widths — No visual bias toward any feedback category
- Vertical card stacking — Supports natural scanning patterns
- Fixed header bar — Always-visible session context (sprint name, timer, phase, participants)
- Horizontal facilitator toolbar — Controls stay accessible without consuming vertical space
- Phase selector as primary action — Leftmost position reflects its importance in session flow
The product in detail
The final design translates wireframe structure into a polished interface using Amadeus's design token system.
Responsive Adaptation
The design adapts from 1440px desktop to a 1024px minimum supported width:
- At 1440px: 3 equal columns, full header nav, Amadeus Max as a 360px sidebar
- At <1280px: Columns stack to 2+1 layout
- At 1024px: Hamburger menu, AI panel as full-width overlay
Measuring success
We defined four success metrics before building, with clear targets tied to research findings.
Time to Join a Session
Time from landing page load to board render
Target
< 10 seconds
Result
8s average
Time to First Card Posted
Time from board load to first card submission
Target
< 30 seconds
Result
18s average
Action Item Completion Rate
Projected% of action items completed within their retro cycle
Target
≥ 60%
Result
Projected: 65%
Based on the persistent carry-forward mechanism removing the primary loss point (Finding 2: only 23% of items were previously referenced)
Facilitator Satisfaction with AI
ProjectedPost-session Likert survey (1-5 scale)
Target
≥ 4 / 5
Result
Projected: 4.2 / 5
Based on qualitative feedback from usability test facilitators who found Amadeus Max's summary and theme grouping features valuable
Design decisions & rationale
Every major design decision traces directly back to a research finding. Here are the four most impactful.
No-account join flow (name + room code only)
Removes signup friction that was the top-cited pain point. Participants enter a display name and 6-character room code — nothing else.
→ Finding 1: Setup Friction
Hidden authorship toggle for card visibility
Protects psychological safety so participants post candidly. Facilitators control when (and if) authorship becomes visible.
→ Finding 4: Psychological Safety
Persistent action items with carry-forward
Directly addresses the 23% action item reference rate. Incomplete items automatically surface at the start of the next retro.
→ Finding 2: Action Item Loss
Enlarged, labeled Amadeus Max button with pulse animation
Addressed the discoverability issue found in usability testing. The original icon-only button was missed by 3 of 5 facilitators.
→ Usability Testing: Major Finding 1
What I learned
Retroactively documenting a shipped product surfaced gaps between what we assumed users needed and what research later confirmed — several features (like hidden authorship) turned out to be more load-bearing for trust than initially expected.
Designing for three distinct roles (Facilitator, Participant, Viewer) with a single shared board required more careful permission modeling than a typical single-role app — IA work paid off significantly here.
Balancing AI assistance with the "AI assists, humans decide" principle meant constantly resisting the urge to make Amadeus Max more autonomous, even when it would have been technically simpler.
I would run usability testing earlier in the process — even informally — rather than treating it as a validation step after high-fidelity designs were built. The Amadeus Max discoverability issue could have been caught during wireframing.
Next steps & opportunities
Improve Amadeus Max discoverability with an onboarding tooltip
Add an onboarding tooltip that highlights the Amadeus Max button on a facilitator's first session.
→ Usability Testing: Major Finding 1
Explore cross-retro trend analytics for action items
Build on the persistent action item architecture to surface trends in follow-through over time — a natural evolution of the carry-forward system.
→ Finding 2: Action Item Loss
Investigate mobile-responsive layouts
Extend the 1024px minimum-width work into true mobile-responsive layouts as teams increasingly join from varied devices.
→ High-Fidelity Designs: responsive adaptation