Compare User Journeys

Common useful group pairs:

• Converted vs churned (same acquisition cohort)
• Power users (top 20% by activity) vs casual users (bottom 20%)
• Activated vs non-activated (within first 7 days)
• Paid vs free (same signup cohort)
• High-NPS vs low-NPS users
• Users who discovered feature X vs those who didn't

Step 2: Pull Quantitative Metrics for Both Groups

Use mcp__Amplitude__query_amplitude_data to compute key behavioral metrics for each group. Run parallel queries for Group A and Group B:

• Session depth: average number of events per session
• Feature breadth: average number of distinct features used per week
• Return frequency: how often do they return per week/month?
• Funnel completion: what percentage complete each key step?
• Time to key milestone: how long (in minutes, hours, or days) does each group take to reach the activation moment?
• Event sequences: what are the most common sequences of events in the first session?
• Drop-off points: where do users in each group most commonly stop?

Compute for each metric: Group A value, Group B value, absolute difference, relative difference (%). This sets up the side-by-side comparison table.

Step 3: Find Session Replays for 5-10 Sessions per Group

Use mcp__Amplitude__get_session_replays to find recordings for users in each group. Filter by the segment criteria defined in Step 1.

Watch 5-10 sessions per group. For each session, note:

• Entry point (where did the session start?)
• Navigation path (what did they explore?)
• Time spent on key steps
• Confusion signals (rage clicks, repeated back/forward, hesitation before key actions)
• Feature discovery (how did they find key features — navigation, search, tooltip, onboarding?)
• Exit point (where did the session end?)

Do not summarize individual sessions — look for what recurs across sessions. A behavior that appears in 3+ Group A sessions but not in Group B sessions is a meaningful differentiator.

Step 4: Extract Behavioral Patterns from Replays

After watching replays, synthesize patterns by group:

Group A patterns (e.g., converters):

• What do they do in the first 5 minutes that Group B doesn't?
• Which features do they discover early?
• What navigation path do they follow?
• What signals of confusion do they show (or not show)?

Group B patterns (e.g., churners):

• Where do they get stuck?
• What do they attempt but fail at?
• What features do they never discover?
• Are there visible moments of frustration?

Step 5: Identify Key Differences

Synthesize the behavioral differences between the two groups into a ranked list. For each difference:

• Name the behavior
• Quantify it (what % of Group A does X vs % of Group B?)
• Note the source (quantitative data, session replay, or both)
• Assess the business impact: does this difference explain conversion/retention/churn?

The most powerful findings are those confirmed by both quantitative data and session replay evidence.

Step 6: Quantify the Gap

For the top 3-5 differentiators, compute the gap quantitatively:

• "Group A uses the collaboration feature in 67% of sessions; Group B uses it in 12% of sessions."
• "Group A reaches the 'aha moment' (first successful export) within 8 minutes on average; Group B takes 31 minutes or never reaches it."
• "Group A's first session has 24 events on average; Group B's has 9."

The gap quantification transforms observations into leverage: if you could move Group B toward Group A on metric X, what would the business impact be?

Step 7: Recommend Behavioral Interventions

Based on the differences identified, recommend specific product changes that could move Group B toward Group A's behavioral patterns:

• For each key difference: what specific product change could close the gap?
• Frame as a testable hypothesis: "If we [add feature discovery prompt for X], then [Group B activation rate] will increase by [Y%] because [Group B currently doesn't find X in the first session]."
• Prioritize recommendations by: (a) size of the gap, (b) feasibility of the intervention, (c) confidence based on evidence strength.

MCP Tools

• mcp__Amplitude__get_session_replays — find and watch session recordings for both user groups
• mcp__Amplitude__query_amplitude_data — pull quantitative behavioral metrics for both groups
• mcp__Amplitude__get_event_properties — discover user properties for group segmentation
• mcp__Amplitude__get_context — get projectId and organization context (always first)
• mcp__Amplitude__query_chart — build funnel or retention charts broken down by the two groups

Key Concepts

• Behavioral differentiator: A specific action or pattern that occurs significantly more often in one group than the other.
• Confirmation triangulation: A finding is high-confidence when it appears in both quantitative data AND session replays.
• Aha moment: The first moment a user experiences the core value of the product. Often the most important behavioral differentiator between activated and non-activated users.
• Rage click: Repeated rapid clicking on a UI element, indicating frustration or confusion. A strong signal of friction in session replays.
• Feature discovery: How and when users first encounter a feature. Users who discover key features early often have higher retention.
• Session depth: The number of meaningful events in a session. Power users typically have higher session depth.
• Gap quantification: Computing the specific numerical difference between groups on a behavioral metric. Transforms observation into leverage.

Output Format

The output combines a comparison table with narrative insights.

Structure:

1. Group definitions (2-3 sentences each): precise definition of Group A and Group B, with sizes.
2. Side-by-side metrics table (markdown):

3. Behavioral patterns (1 paragraph per group): what does each group's typical session look like?
4. Key differentiators (ranked list of 3-5): what Group A does that Group B doesn't, with evidence from both quantitative data and session replays.
5. Recommendations (numbered list): specific product changes to move Group B toward Group A behavior, each framed as a testable hypothesis.