Analytics for Good Measure Giving

Analyze user behavior, track conversions, and understand engagement patterns using three data sources:

1. Cloudflare — primary traffic source (zone-level HTTP metrics + RUM beacon for real browser visits)
2. Firestore — user signups, feature adoption, giving activity
3. GA4 — event-level engagement tracking (blocked by ~80%+ of visitors)
4. PMF Analysis — engagement tier segmentation (Superhuman PMF framework)

---

When This Skill Activates

• Checking site traffic and engagement
• Analyzing user funnels (browse → view → convert)
• Reviewing search behavior and popular charities
• Tracking login/signup conversions
• Understanding time on site and engagement metrics
• Checking user signups, feature adoption, or giving activity
• Investigating Firestore user data or reported issues
• Comparing traffic across data sources

---

Data Sources

Cloudflare (Primary Traffic Metrics)

Script: scripts/cloudflare_analytics.py

source ~/.secrets/api_keys.sh && uv run python scripts/cloudflare_analytics.py 2>/tmp/cf_analytics.err

Outputs JSON with:

• daily[] — merged zone + RUM daily metrics
• totals.zone — HTTP-level totals (includes bots/crawlers)
• totals.rum — Real browser visits (beacon-based, most accurate)
• top_paths.rum[] — Top pages by visit count
• top_countries.rum[] — Geographic breakdown
• browsers[] — Browser/device split

Key metric relationships:

• zone.unique_ips = all unique IPs (bots + humans) — inflated
• rum.visits = real human browser visits — most accurate traffic count
• ga4.sessions = human visits without ad blockers — undercounts by ~80%

Environment variables (from ~/.secrets/api_keys.sh):

• CLOUDFLARE_API_TOKEN
• CLOUDFLARE_ZONE_ID (goodmeasuregiving.org)
• CLOUDFLARE_ACCOUNT_ID

Firestore

Script: scripts/firestore_analytics.py

uv run python scripts/firestore_analytics.py 2>/tmp/firestore_analytics.err

See Firestore schema section below.

PMF Analysis

Script: scripts/pmf_analysis.py

uv run python scripts/pmf_analysis.py 2>/tmp/pmf_analysis.err

Outputs formatted text (not JSON) with:

• Tier distribution — Champions (Tier 1), Interested (Tier 2), Passive (Tier 3)
• PMF proxy score — % of 30+ day users in Tier 1
• Tier 1 "Nicole" profile — feature adoption patterns of best users
• Gap analysis — what separates each tier (the activation gap)
• GA4 acquisition funnel — visitor → sign-in → registered conversion rates
• DAU/WAU/MAU stickiness — daily engagement trends

Uses both Firestore (gcloud auth) and GA4 (service account at ~/.secrets/Roshni-3eada2766db6.json).

Pass --detail for per-user breakdown.

GA4

GA4 Property ID: Set via GA4_PROPERTY_ID environment variable or .env file.

Important: GA4 undercounts traffic significantly (~80%+ of visitors have ad blockers). Use GA4 for event-level behavior insights (which charities get clicked, search terms, funnel events), NOT for traffic volume. Use Cloudflare RUM for traffic counts.

---

Project Context

Custom Events Tracked (GA4):

Event	Parameters	Purpose
page_view	page_path, page_title, flow_*	Navigation tracking
charity_view	charity_id, charity_name, view_type, flow_*	Detail page opens
charity_card_click	charity_id, charity_name, charity_tier, list_position, flow_*	Browse engagement
search	search_term, result_count, flow_*	Search behavior
donate_click	charity_id, charity_name, destination_url, flow_*	Donation intent
outbound_click	charity_id, charity_name, destination_url	External links
sign_in_start	method, flow_*	Login funnel start
sign_in_success	method, auth_type, flow_*	Login funnel complete (auth_type: 'signup' or 'login')
hero_cta_click	cta_name, destination_path	Landing engagement

Flow Tracking Parameters (on all major events):

Parameter	Example	Purpose
flow_id	1706234567890-abc123	Unique session identifier
flow_path	landing>browse>card_click>charity_view	User journey sequence
flow_step	4	Step number in journey

Key Goals:

1. Primary: Login conversions (sign_in_start events)
2. Secondary: Time on site / engagement
3. Tertiary: Feature adoption after signup

---

Cross-Source Analysis Patterns

Traffic Sanity Check

Compare these three numbers to understand measurement coverage:

1. CF zone unique IPs (inflated — includes bots)
2. CF RUM visits (real human traffic)
3. GA4 sessions (humans without ad blockers)

Conversion Funnel (full stack)

1. Cloudflare RUM visits → total real visitors
2. GA4 sessions → visitors without ad blockers
3. GA4 sign_in_start → login intent
4. GA4 sign_in_success → login completion
5. Firestore user count → registered users
6. Firestore feature adoption → active users

Content Performance Cross-Reference

• CF RUM top_paths → most visited pages
• GA4 charity_card_click → most engaged charities
• Firestore bookmarks → most saved charities
• Overlap analysis: are visited ≈ engaged ≈ saved?

---

Analytics Queries (GA4)

Realtime Overview

Use mcp__analytics-mcp__run_realtime_report with:
- property_id: (use GA4_PROPERTY_ID from .env)
- dimensions: ["eventName"]
- metrics: ["eventCount"]

Top Charities by Card Clicks (Last 7 Days)

Use mcp__analytics-mcp__run_report with:
- property_id: (use GA4_PROPERTY_ID from .env)
- date_ranges: [{"start_date": "7daysAgo", "end_date": "yesterday"}]
- dimensions: ["customEvent:charity_name"]
- metrics: ["eventCount"]
- dimension_filter: {"filter": {"field_name": "eventName", "string_filter": {"match_type": 1, "value": "charity_card_click"}}}
- order_bys: [{"metric": {"metric_name": "eventCount"}, "desc": true}]
- limit: 20

Search Terms Analysis

Use mcp__analytics-mcp__run_report with:
- property_id: (use GA4_PROPERTY_ID from .env)
- date_ranges: [{"start_date": "7daysAgo", "end_date": "yesterday"}]
- dimensions: ["customEvent:search_term"]
- metrics: ["eventCount"]
- dimension_filter: {"filter": {"field_name": "eventName", "string_filter": {"match_type": 1, "value": "search"}}}
- order_bys: [{"metric": {"metric_name": "eventCount"}, "desc": true}]
- limit: 20

Engagement Metrics (Time on Site)

Use mcp__analytics-mcp__run_report with:
- property_id: (use GA4_PROPERTY_ID from .env)
- date_ranges: [{"start_date": "7daysAgo", "end_date": "yesterday"}]
- dimensions: ["date"]
- metrics: ["averageSessionDuration", "engagedSessions", "sessions", "activeUsers"]
- order_bys: [{"dimension": {"dimension_name": "date"}, "desc": false}]

Login Funnel Analysis

Use mcp__analytics-mcp__run_report with:
- property_id: (use GA4_PROPERTY_ID from .env)
- date_ranges: [{"start_date": "7daysAgo", "end_date": "yesterday"}]
- dimensions: ["eventName"]
- metrics: ["eventCount"]
- dimension_filter: {"filter": {"field_name": "eventName", "in_list_filter": {"values": ["sign_in_start", "page_view"], "case_sensitive": true}}}

---

Funnel Analysis Pattern

Browse → Card Click → Charity View → Donate Funnel

Step 1: Count users who visited /browse

dimensions: ["pagePath"]
metrics: ["activeUsers"]
dimension_filter: pagePath contains "/browse"

Step 2: Count charity_card_click events

dimensions: ["eventName"]
metrics: ["eventCount"]
dimension_filter: eventName = "charity_card_click"

Step 3: Count charity_view events

dimensions: ["eventName"]
metrics: ["eventCount"]
dimension_filter: eventName = "charity_view"

Step 4: Count donate_click events

dimensions: ["eventName"]
metrics: ["eventCount"]
dimension_filter: eventName = "donate_click"

Calculate drop-off rates between each step.

---

Standard Reports to Run

Daily Health Check

Run these queries and summarize:

1. Traffic (Cloudflare): RUM visits, pageloads, zone requests, unique IPs
2. Engagement (GA4): Avg session duration, engaged sessions (note: GA4 undercounts)
3. Top Pages (Cloudflare RUM): Most visited paths
4. Top Charities (GA4): Most clicked charity cards
5. Conversions (GA4): sign_in_start count
6. Users (Firestore): Total registered users, recent signups

Weekly Deep Dive

1. Traffic Trends (Cloudflare): Week-over-week RUM visits, geographic shifts
2. Search Analysis (GA4): What are users searching for? Gaps in charity coverage?
3. Funnel Metrics: CF RUM visits → GA4 sessions → card clicks → views → donates → signups
4. Device Split (Cloudflare): Browser breakdown, mobile vs desktop
5. Feature Adoption (Firestore): Which features are growing/stagnant?

---

Interpreting Results

Engagement Benchmarks

Metric	Poor	Average	Good
Avg Session Duration	<1 min	1-3 min	>3 min
Pages per Session	<2	2-4	>4
Bounce Rate	>70%	40-70%	<40%
Card Click Rate	<5%	5-15%	>15%

Red Flags to Watch

• CF RUM visits high but GA4 near-zero → GA4 is broken (check initializeAnalytics call)
• CF zone unique_ips >> CF RUM visits → normal (bots), but ratio > 10:1 warrants investigation
• High bounce rate on /browse → cards may not be compelling
• Low charity_view after card_click → page load issues?
• Zero search events → search feature not discoverable
• sign_in_start with no sign_in_success → auth issues

---

Flow Path Analysis

Query common user journeys:

Use mcp__analytics-mcp__run_report with:
- property_id: (use GA4_PROPERTY_ID from .env)
- date_ranges: [{"start_date": "7daysAgo", "end_date": "yesterday"}]
- dimensions: ["customEvent:flow_path"]
- metrics: ["eventCount"]
- dimension_filter: {"filter": {"field_name": "eventName", "string_filter": {"match_type": 1, "value": "donate_click"}}}
- limit: 20

This shows what paths lead to donations.

---

Quick Commands

Realtime snapshot:

"What's happening on the site right now?"

Weekly report:

"Give me a weekly analytics summary"

Search insights:

"What are users searching for?"

Charity performance:

"Which charities get the most clicks?"

Engagement check:

"How's our time on site trending?"

Full report (all sources):

"Run /analytics"

---

Firestore Data Source

Script

scripts/firestore_analytics.py — queries Firestore via REST API using gcloud auth print-access-token. No additional dependencies (stdlib only). Outputs structured JSON to stdout, logs to stderr.

uv run python scripts/firestore_analytics.py 2>/tmp/firestore_analytics.err

Firestore Schema

users (top-level collection)

Field	Type	Description
createdAt	timestamp	Account creation date
updatedAt	timestamp	Last profile update
targetZakatAmount	number/null	Annual zakat target
zakatAnniversary	string/null	Zakat calculation anniversary date
givingBuckets	array	Custom giving categories (id, name, tags, percentage, color)
charityBucketAssignments	array	Charities assigned to buckets (charityEin, bucketId)
geographicPreferences	array	Preferred regions for giving
fiqhPreferences	map	Islamic jurisprudence preferences
givingPriorities	map	Cause area priorities

users/{uid}/bookmarks (subcollection)

Field	Type	Description
charityEin	string	Bookmarked charity EIN
notes	string/null	User notes
createdAt	timestamp	When bookmarked

users/{uid}/giving_history (subcollection)

Field	Type	Description
charityEin	string	Charity EIN
charityName	string	Charity display name
amount	number	Donation amount
date	string	Donation date (YYYY-MM-DD)
category	string	zakat, sadaqah, or other
zakatYear	number	Applicable zakat year
taxDeductible	boolean	Tax deductible flag
receiptReceived	boolean	Receipt received flag
matchEligible	boolean	Employer match eligible
matchStatus	string/null	Match status
matchAmount	number/null	Match amount
paymentSource	string/null	Payment method
notes	string/null	User notes
createdAt	timestamp	When recorded

users/{uid}/charity_targets (subcollection) — planned giving targets (currently empty)

reported_issues (top-level collection) — user-reported data issues (currently empty)

Feature Adoption Metrics

The script calculates adoption as: (users with feature) / (total users) * 100%

Features tracked:

• Bookmarks: user has any docs in bookmarks subcollection
• Giving History: user has any docs in giving_history subcollection
• Zakat Target: targetZakatAmount is non-null and non-zero
• Giving Buckets: givingBuckets array is non-empty
• Bucket Assignments: charityBucketAssignments array is non-empty
• Geographic Preferences: geographicPreferences array is non-empty
• Fiqh Preferences: fiqhPreferences map has any truthy values
• Zakat Anniversary: zakatAnniversary is non-null

MCP Tool Limitations

The Firebase MCP tools (firestore_query_collection, firestore_get_documents) cannot query subcollections. The Python script uses the Firestore REST API with allDescendants: true collection group queries to access bookmarks, giving_history, and charity_targets data across all users in a single API call per collection type.