Funnel Drop-off Investigator

Diagnose why users drop out of key flows by chaining three data sources — funnel metrics, session replays, and qualitative feedback — into a single actionable report. This is the fastest way to move from "conversion is low" to "here's why, and here's evidence."

Workflow

Step 1: Gather Context

Ask the user:

• Which funnel or user journey to investigate (e.g., "Trial signup", "Booking confirmation", "Onboarding")
• Time period (default: last 30 days)
• Product area or page names if known

Get app context:

Use list_all_applications to confirm subId and appId

Step 2: Run the Funnel Analysis

Find the funnel by name or search for relevant pages:

Use searchEntities with:
- itemType: "Page" (and optionally "Feature")
- query: [user-described flow name]
- limit: 20

If the user has a named Pendo funnel, use queryFunnel directly. If they're describing a journey rather than an existing funnel, construct the step sequence from the pages/features found above and confirm with the user before running.

Run the funnel:

Use queryFunnel with:
- subId, appId
- steps: [ordered array of pageId/featureId]
- startDate / endDate
- Returns: visitors at each step, conversion rate per step, time between steps

Identify the critical drop-off:

• Find the step with the largest absolute visitor drop
• Calculate the step-level conversion rate for each transition
• Flag any step where conversion < 50% or drops >20% compared to the previous step

Present a step summary to the user before continuing:

Step 1 → Step 2: X visitors → Y visitors (Z% conversion)
Step 2 → Step 3: Y visitors → W visitors (V% conversion) ← ⚠️ Biggest drop
Step 3 → Step 4: ...

Step 3: Pull Session Replays for the Critical Step

Target the step immediately before the biggest drop-off point — these are the sessions where users reached the problematic moment but didn't convert.

Use sessionReplayList with:
- subId, appId
- startDate / endDate
- pageId: [page ID of the critical drop-off step]
- minDuration: 10 (seconds — filter out accidental bounces)
- activityPercentile: 20 (focus on lower-engagement sessions — the confused or stuck ones)
- limit: 5

Present the replays as a numbered list with:

• Session ID / visitor ID
• Duration on page
• Activity level
• Direct link to replay (if URL available from result)

Note to Claude: Don't summarise what's in the replays — you can't watch them. Simply surface them with enough context for the PM to click through and watch the 3–5 most informative sessions. Explain what to look for: rage clicks, hesitation, repeated scrolling, form abandonment.

Step 4: Pull Qualitative Feedback for the Drop-off Area

Use get_feedback_items with:
- subId (Listen subscription)
- dateRange: last 30 days
- limit: 50

Then filter/score feedback relevant to the funnel area:

• Match on page name, feature name, or flow keywords
• Look for sentiment patterns: confusion, friction, missing info, error states
• Prioritise verbatims that mention the specific step or page name

Surface the top 5–8 most relevant feedback items as direct quotes with source/date.

Also run topic clustering if volume warrants it:

Use generate_feedback_topics with:
- same filters as above
- Returns: AI-clustered themes with representative quotes

Step 5: Generate the Investigation Report

Produce a Markdown artifact saved to /mnt/user-data/outputs/funnel-investigation-[flow-name]-[date].md

Report Structure

# Funnel Drop-off Investigation: [Flow Name]
**Date:** [today]  **Period:** [date range]  **App:** [app name]

## TL;DR
[2–3 sentence summary: where drop-off is worst, what the replays suggest, what feedback confirms]

## Funnel Overview

| Step | Visitors | Conversion | Drop |
|------|----------|------------|------|
| Step 1: [Name] | X | — | — |
| Step 2: [Name] | Y | Z% | -N |
| ⚠️ Step 3: [Name] | W | V% | -M ← Critical |
| Step 4: [Name] | ... | | |

**Critical drop-off:** Step 2 → Step 3 loses the most users (N visitors, V% conversion)

## Session Replays — What to Watch

These 5 sessions show users who reached [critical step] but didn't complete it.
Look for: rage clicks, repeated scrolling, hesitation before CTA, form errors.

1. [Visitor/Session ID] — [X]s on page, [activity level] — [link or ID]
2. ...

**What to look for in these sessions:**
- [Specific UI element or CTA name] — did users find it?
- [Form field or input] — were there errors or abandonment?
- [Navigation pattern] — did users go back or exit?

## Qualitative Feedback — What Users Are Saying

**Top themes from [N] feedback items in this area:**

### Theme 1: [Theme name] ([N] mentions)
> "[Representative verbatim quote]" — [Source, Date]
> "[Second quote if strong]"

### Theme 2: [Theme name] ([N] mentions)
> "[Representative verbatim quote]"

[Repeat for 3–5 themes]

## Hypotheses

Based on the funnel data, replays, and feedback, the most likely causes of drop-off are:

1. **[Hypothesis 1]** — Supported by: [data point / feedback theme]
2. **[Hypothesis 2]** — Supported by: [data point / feedback theme]
3. **[Hypothesis 3]** — Supported by: [data point / feedback theme]

## Recommended Next Steps

| Action | Priority | Owner |
|--------|----------|-------|
| [e.g. Watch 5 replays at step X] | High | PM |
| [e.g. Add tooltip/guide at step X] | High | Design |
| [e.g. A/B test CTA copy] | Medium | PM/Eng |
| [e.g. Add NPS/feedback prompt at drop-off step] | Medium | PM |

Step 6: Present the Report

Use present_files to share the Markdown artifact.

Summarise in chat:

• The single biggest drop-off step and its conversion rate
• The #1 replay signal to investigate
• The dominant feedback theme
• Your top hypothesis for what's causing the drop

Pendo MCP Tool Reference

queryFunnel

• subId, appId: required
• steps: ordered array of { type: "page"|"feature", id: "..." }
• startDate / endDate: YYYY-MM-DD
• Returns: visitors per step, conversion rates, median time between steps

sessionReplayList

• subId, appId: required
• pageId: filter to a specific page (the drop-off step)
• startDate / endDate
• minDuration: seconds (filter accidental bounces — use 10)
• activityPercentile: 0–100 (lower = less active = potentially confused users)
• limit: number of replays to return (5 is usually enough)
• Returns: session IDs, visitor IDs, duration, activity %, replay URL if available

get_feedback_items

• subId: Listen subscription ID
• dateRange or startDate/endDate
• limit: number of items (50 is a good starting point)
• Returns: raw feedback verbatims with source, date, sentiment

generate_feedback_topics

• Same filters as get_feedback_items
• Returns: AI-clustered themes with representative quotes and counts

searchEntities

• itemType: "Page", "Feature"
• query: semantic search term
• itemIds: resolve IDs to names

list_all_applications

• Returns available subscriptions and apps with IDs

Error Handling

• Funnel has no data: Check date range — widen if needed. Confirm page IDs are correct.
• No replays returned: Widen date range, lower minDuration, or raise activityPercentile.
• No relevant feedback: Run without page filter to get broader feedback pool, then manually filter for relevance.
• Can't resolve page names: Show IDs and ask user to confirm which pages map to which funnel steps.
• Funnel steps are unclear: Present candidate pages from searchEntities and ask user to confirm the sequence before running.