Skip to main content
AI Features

Smart Grouping

What Smart Grouping Does

Smart grouping uses AI to analyze the content of all cards in a retrospective and suggest thematic groupings. Instead of manually reading through dozens of cards and dragging them together one by one, the facilitator can ask the AI to propose groups based on shared topics, sentiments, or underlying themes.

Grouping is one of the most time-consuming phases of a retrospective, especially for larger teams. With 30 or more cards on the board, it can take 10 to 15 minutes of silent reading and shuffling. Smart grouping reduces this to a quick review step, giving your team more time for the discussion that actually drives improvement.

Smart grouping is available during the group phase of a retrospective. Only the facilitator can trigger AI grouping suggestions, but all participants can see the proposed groups and provide input before they are finalized.

How to Trigger Smart Grouping

To generate AI-suggested groupings for your retrospective:

  1. Advance the retrospective to the group phase.
  2. Click the "Suggest groupings" button in the facilitator toolbar at the top of the board. This button is marked with a sparkle icon to indicate it is AI-powered.
  3. Wait while the AI analyzes all cards across every column. This typically takes five to ten seconds depending on the number of cards.
  4. Review the suggested groupings that appear as highlighted clusters on the board. Each group is given a provisional theme label.

The AI reads every card in the retrospective, regardless of which column it belongs to. It looks for semantic similarity, shared topics, and complementary perspectives to form its grouping suggestions.

Reviewing Suggested Groups

After the AI generates its suggestions, the board enters a review state. Each proposed group is displayed with:

  • Theme label — A short, descriptive name for the group such as "Deployment pipeline friction" or "Cross-team communication gaps."
  • Grouped cards — The cards the AI has placed into the group, shown with a colored highlight matching the group.
  • Confidence indicator — A subtle visual cue showing how strongly the AI believes the cards belong together. High-confidence groupings have a solid border; lower-confidence suggestions use a dashed border.

For each suggested group, the facilitator can take one of these actions:

  • Accept — Confirms the group. The cards are merged into a card group with the suggested theme label.
  • Accept with edits — Confirms the group but allows renaming the theme label or removing individual cards before finalizing.
  • Reject — Dismisses the suggestion. The cards return to their ungrouped state for manual handling.

You do not have to accept or reject all groups at once. Work through them one at a time, starting with the high-confidence suggestions. This lets you handle the obvious groupings quickly and spend more time on the nuanced ones.

Manual Adjustments After AI Grouping

After accepting AI-suggested groups, you can make any manual adjustments you need. The AI groupings are a starting point, not a final answer. Common adjustments include:

  • Splitting a group — If the AI combined cards that cover two distinct subtopics, break the group apart and create two separate groups.
  • Merging groups — If two AI-suggested groups are really about the same theme, drag one group into the other to combine them.
  • Moving individual cards — Drag a card out of one group and into another, or back to the ungrouped area.
  • Renaming theme labels — Click the theme label to edit it. Use language that resonates with your team rather than the AI's phrasing.
  • Grouping remaining cards — Some cards may not fit any AI-suggested group. Group these manually or leave them as standalone discussion items.

How Smart Grouping Works

Understanding how the AI forms its groupings can help you evaluate suggestions more effectively. The AI considers several factors when analyzing cards:

  1. Semantic similarity — Cards that discuss the same topic using different words are identified as related. For example, "standups run too long" and "daily syncs eat into development time" would be grouped together.
  2. Complementary perspectives — A positive card about pair programming and a request for more pairing sessions may be grouped because they address the same practice.
  3. Root cause alignment — Cards describing different symptoms of the same underlying issue may be grouped, even if the surface-level topics differ.
  4. Column context — The AI considers which column a card was placed in (such as "Went Well" or "Needs Improvement") to better understand the intent behind the text.

Smart grouping works best with cards that contain enough detail for the AI to find meaningful connections. Very short cards (one or two words) may not group well. Encourage your team to write descriptive cards, or use Card Coaching first to improve card quality before grouping.

Tips for Getting the Best Results

  • Encourage detailed cards — The more context in each card, the better the AI can identify genuine connections between them.
  • Use card coaching first — Running card coaching during the reflect phase helps produce higher-quality input for smart grouping.
  • Review with your team — Share the AI-suggested groups on screen and ask the team if the groupings feel right before accepting them. This builds collective ownership.
  • Run it once — You can trigger smart grouping multiple times, but the results tend to be similar. If the first pass does not feel right, manual grouping may be a better fit for that particular retro.

Ungrouped Cards

Not every card will be assigned to a group by the AI. Cards that do not share clear thematic connections with others remain ungrouped. This is expected and healthy. Some feedback is unique and deserves its own discussion slot rather than being forced into a group.

Ungrouped cards are still fully visible during the voting and discussion phases. The facilitator can decide whether to discuss them individually, group them manually, or set them aside if time is limited.