AI Retrospectives: Templates, Prompts & 10 Formats to Try

Of all the scrum ceremonies I’ve facilitated, the retrospective has the largest gap between potential and reality. I’ve watched too many teams run the same Start/Stop/Continue retro, surface the same noisy themes, and let the same actions go unactioned. AI doesn’t fix that discipline gap, but in my experience it removes the friction that often gets used to justify skipping the work.

In this guide I cover ten AI-augmented retrospective formats I’ve used or recommend, the prompts behind each, the rituals that make AI retros produce real change, and how I measure whether a retro practice is actually moving the needle.

AI helps retros at three points:

• Async input collection: people contribute thoughts before the meeting in their own time.
• Theme clustering: AI groups raw input into themes faster than humans can.
• Action drafting: AI proposes specific actions per theme that humans can accept or refine.

Used well, AI saves 30-40% of retro time without losing depth. Used poorly, AI flattens unique inputs into generic themes and crowds out individual voice.

Running the same retro format every sprint produces fatigue. Rotate formats every 4-6 sprints to keep engagement high. The 10 formats below cover the most useful patterns.

Track which formats produce the best engagement and outcomes for your team. After 6-12 months, you’ll have data showing which 3-4 formats your team actually responds to.

Classic retro upgraded. Team members submit raw input async. AI clusters into themes. Team prioritises actions in 30 minutes.

Prompt:

“Below is the team’s Start/Stop/Continue input from last sprint. Cluster into 5-7 themes. For each theme, suggest a 1-line description, frequency, and a candidate action.”

Best for: standard retro cadence, established teams.

Stronger than Start/Stop/Continue at surfacing growth opportunities. AI helps cluster especially the Learned and Longed For categories.

Best for: teams in growth phase, when learning is a priority.

Visualise the sprint as a sailboat: anchors (what slowed you), wind (what helped), rocks (risks), island (goal). AI helps cluster anchors and rocks into actionable themes.

Best for: teams that respond to visual frameworks, hybrid sessions.

Emotionally focused. Useful when team morale needs surfacing. AI helps cluster emotional themes without flattening them.

Best for: post-difficult-sprint, after a major launch, when stress is high.

Categorise practices as: house of straw (fragile), house of sticks (some structure but weak), house of bricks (durable). AI helps identify which practices are fragile and which are sustainable.

Best for: process-focused retros, after process changes.

Team submits topics, votes, discusses each in timeboxed slots. AI helps:

• Cluster similar topics before voting.
• Generate discussion prompts for high-vote topics.
• Capture key points async during discussion.

Best for: teams that want highly democratic agenda-setting.

Categorise practices as: house of straw (fragile), house of sticks (some structure but weak), house of bricks (durable). AI helps identify which practices are fragile and which are sustainable.

Best for: process-focused retros, after process changes.

Team rates energy and engagement throughout the sprint. AI clusters causes and proposes interventions. Useful when burnout is a concern.

Best for: teams showing burnout signals, post-crunch periods.

For each focus area, the team assesses skill (do we have it), will (do we want to), and hill (what stands in the way). AI helps generate the hills - obstacles - by reading recent ticket comments and Slack patterns.

Best for: teams working on specific capability development.

Pick a single theme (e.g., quality, communication, planning accuracy) and run a focused retro. AI prepares the topic-specific data ahead of time.

Best for: persistent issues that warrant deep focus.

A general retro synthesis prompt:

“Below are 30 retro inputs from a 7-person team. Cluster into themes. For each: name, count of inputs, dominant emotion, suggested action. Flag any theme that appeared in the previous retro to detect patterns.”

Topic-focused prompt:

“Read these inputs about quality issues in the last sprint. Identify root causes. Suggest 3 actionable changes the team can make next sprint.”

Action drafting prompt:

“Take this theme: ‘Estimation is consistently low’. Suggest 5 specific actions. For each: owner suggestion, success metric, and target date.”

Surprise-surfacing prompt:

“Read these retro inputs. What is the most surprising or counter-intuitive signal? What might the team be missing because it’s uncomfortable?”

Pattern-detection prompt:

“Compare retro themes from the last 4 sprints. What recurring patterns suggest systemic issues vs sprint-specific issues?”

The most common retro failure is the gap between theme identification and action follow-through. AI helps but does not solve this. Discipline practices:

• Every action has an owner. AI drafts owner suggestions; the team confirms.
• Every action has a measurable success criterion.
• Last sprint’s actions are reviewed at the start of the next retro.
• Aged unactioned items get killed or escalated.
• One action per retro is enough; three is plenty; five is overload.

The team that closes 80% of retro actions in the next sprint outperforms the team that opens 5 per retro and closes none.

Track four metrics:

• Action close rate: percentage of actions completed within 2 sprints.
• Repeat themes: themes that recur across multiple retros (signal of unresolved issues).
• Engagement score: pulse survey on retro value.
• Outcome lift: do team metrics improve in the months following retro action implementation?

Strong retro practice shows action close rate above 70% and falling repeat-theme count over time.

AI is a tool, not a panacea. Some retro problems require human discipline:

• Skipping retros when busy. AI doesn’t restore the discipline.
• Blame culture. AI clustering can’t change interpersonal dynamics.
• Manager presence dampening honesty. Structural issue, not solvable by tooling.
• Poor manager response to feedback. If actions get blocked above scrum master level, retros become performative.
• Same actions opened and closed every retro. Signal of deeper organisational issues.