Ecommerce Conversion Insights

Overview

The E-commerce Conversion Insights skill analyzes funnel data from uploaded CSV or Excel files (exported from any analytics platform, such as Google Analytics, Adobe Analytics, or Shopify) to identify where customers are dropping off in the purchase journey. It calculates statistical significance, estimates revenue impact for each friction point, and delivers a prioritized list of issues with actionable recommendations.

It replaces the manual weekly workflow where analytics teams pull data, segment by device/geo/category, and present findings in slide decks. Instead, it delivers on-demand insights from any funnel export.

Workflow

<Definition - Funnel Stages> Standard e-commerce conversion funnel stages (adapt based on available data):

1. Session Start (landing page view)
2. Product View (PDP engagement)
3. Add to Cart
4. Begin Checkout
5. Payment Info Entry
6. Order Confirmation (conversion)

Each stage transition has an expected range. A "friction point" is where the actual conversion rate between stages falls below the historical baseline by a statistically significant margin. </Definition - Funnel Stages>

<Definition - Revenue Impact Formula> Revenue Impact = (Visitors at Stage N) × (Expected Conversion % - Actual Conversion %) × (Average Order Value)

This gives the estimated revenue recoverable if the friction point were resolved to return to baseline performance. </Definition - Revenue Impact Formula>

<Definition - Statistical Significance> A conversion rate change is considered significant when:

• Sample size ≥ 1,000 sessions at the relevant stage
• Z-score ≥ 1.96 (95% confidence) comparing current period to baseline
• The absolute change is ≥ 0.5 percentage points </Definition - Statistical Significance>

<Definition - Uploaded File Format> CSV or Excel with columns: stage_name, visitors (required). Optional columns: device, geo/country, category, conversions, avg_order_value. Column names are matched case-insensitively with common aliases (e.g., "users" becomes visitors, "step" becomes stage_name). </Definition - Uploaded File Format>

<Workflow - Ingest Data description="Load the uploaded funnel file and standardize for analysis." tools=[run_python, file_read] triggers=["User asks about conversion", "User uploads a funnel file"]

1. [Ask user] Which file contains the funnel data?

   • Accept a file path to a CSV or Excel file
   • If already provided as {{funnel_file}}, skip to Step 2
   • If fails: no path can be resolved, re-ask and show the expected format from <Definition - Uploaded File Format>

2. [Agent] Run scripts/funnel_analyzer.py with --file-path {{funnel_file}} and --segments {{segments}}. The script reads the CSV or Excel file, normalizes column headers to stage/visitors/conversions/device/geo/category, and returns a standardized JSON structure.

   • Validate: script exits successfully and the output lists the required stage and visitors fields
   • If fails: the script reports the missing required columns and the columns it found. Show that error and the expected format, then ask the user to correct the file.

3. [Agent] Confirm the standardized data is ready for analysis:

   • Check that the parsed data has stage, visitors, and at least one segment dimension
   • Flag any data quality issues (e.g., conversions > visitors, missing stages)
   • Determine the AOV: if the data carries avg_order_value, compute a visitor-weighted AOV and record it; otherwise ask the user for an average order value, defaulting to $85 if none is given
   • If fails: report the specific data quality issue and ask the user how to proceed

</Workflow - Ingest Data>

<Workflow - Analyze Funnel description="Identify friction points, calculate revenue impact, and determine statistical significance." tools=[run_python] triggers=["Data ingestion complete"]

1. [Agent] Calculate baseline conversion rates:

   • If historical data is available (a prior period in the file), use that as baseline
   • If no history, use industry benchmarks from references/industry-benchmarks.md
   • Compute expected vs. actual conversion rate for each stage transition
   • If fails: no baseline can be established, note that findings are relative to industry defaults and continue

2. [Agent] Identify friction points:

   • Run scripts/revenue_impact_calculator.py with the standardized data and the AOV recorded in Ingest Data Step 3 (pass it via --aov)
   • For each stage transition, compute the absolute drop-off rate vs. baseline, the Z-score for statistical significance, and the revenue impact using <Definition - Revenue Impact Formula>
   • Filter to statistically significant drops only, per <Definition - Statistical Significance>
   • If fails: the script errors or returns no significant drops, report that no significant friction was found and show the raw stage rates

3. [Agent] Segment analysis:

   • Break down each friction point by available segment dimensions
   • Identify whether friction concentrates in specific segments (a segment rate more than 2x off the overall rate)

4. [Agent] Rank friction points:

   • Sort by estimated revenue impact (descending)
   • Keep the top 5, or fewer if less than 5 are statistically significant
   • For each, generate a root cause hypothesis based on the segment concentration pattern and stage-specific common issues from references/common-blockers.md

</Workflow - Analyze Funnel>

<Workflow - Generate Recommendations description="Produce actionable recommendations and deliverable report." tools=[run_python, file_write, open_in_session_tab] triggers=["Analysis complete"]

1. [Agent] Generate recommendations. For each of the top 5 friction points, create a specific recommendation:

   • What to change and why
   • Expected impact (revenue recovery estimate)
   • Implementation complexity (low/medium/high)

2. [Agent] Build the insight report:

   • Use assets/report-template.md as the output structure
   • Populate the executive summary (one paragraph, top finding plus total revenue opportunity), funnel visualization, top 5 friction points table (stage, drop %, revenue impact, segment concentration), detailed recommendations, and methodology notes
   • Ask the user where to save the report, or save it to the skill's assets/ directory if no path is given
   • If fails: the template is missing, assemble the report from the sections in instead

3. [Agent] Render visualization:

   • Create an HTML artifact with a funnel chart of stage-to-stage conversion rates (color-coded by severity), segment breakdown charts for the top friction points, and a revenue impact chart
   • Open it in a session tab for user review

4. [Ask user] Present findings:

   • Show the executive summary inline
   • Link to the full report and visualization
   • Ask: "Would you like me to drill deeper into a specific friction point, or export this as a slide deck?"

</Workflow - Generate Recommendations>

<Template - Executive Summary>

## Conversion Insight Report: {{date_range}}

**Top Finding:** {{top_friction_point_description}}

**Total Revenue Opportunity:** ${{total_revenue_impact}}/{{period}} across {{num_friction_points}} identified friction points.

**Biggest Segment Impact:** {{segment_with_highest_concentration}} accounts for {{segment_percentage}}% of the total drop-off at {{worst_stage}}.

**Immediate Action:** {{top_recommendation_one_liner}}

</Template - Executive Summary>

<Template - Friction Point Card>

### {{rank}}. {{stage_from}} → {{stage_to}} Drop-off

| Metric | Value |
|--------|-------|
| Current Conversion Rate | {{current_rate}}% |
| Baseline Rate | {{baseline_rate}}% |
| Absolute Drop | {{drop_pp}} pp |
| Visitors Affected | {{visitors_affected}} |
| Est. Revenue Impact | ${{revenue_impact}}/{{period}} |
| Statistical Confidence | {{confidence}}% |

**Segment Concentration:**
- Device: {{device_breakdown}}
- Geo: {{geo_breakdown}}
- Category: {{category_breakdown}}

**Root Cause Hypothesis:** {{hypothesis}}

**Recommendation:**
- Change: {{recommended_change}}
- Expected Impact: {{expected_impact}}
- Complexity: {{complexity}}

</Template - Friction Point Card>