Core Principles

Note: These are fundamental truths and rules. For step-by-step investigation processes, see Thinking Flows.

Monitoring & Observability

Principle: Self-Sufficient Monitoring

Don’t wait for notifications or alerts to tell you what changed. Observe and investigate proactively.

You should never be asking “what landing pages should I track?” - the system should show you ALL landing pages automatically
If something changes in user behavior or new pages get traffic, you should discover it yourself through observation
Being self-sufficient in data analysis is critical - don’t depend on others to tell you what to look at

Principle: Self-Service Event Discovery

Never ask developers “what events do we have?” - explore and discover them yourself.

The tool should let you see all events by exploring user activity
You don’t need documentation of every event - just look at real user sessions
Understanding the funnel comes from observing actual event sequences, not from spec documents
If you need to ask what data exists, your analytics tool is failing you

Principle: Comprehensive Data Collection

Track everything by default, not just what you think matters.

Don’t pre-filter what you think is important - capture all user entry points
This ensures you catch unexpected patterns and changes
Example: Tracking landing pages by path name should include not just designed landing pages, but any page users land on (/terms, /about-us, /how-it-works, etc.)

Data Analysis

Principle: Discover, Don’t Define

Let the data show you what’s happening rather than only looking at predefined metrics.

Systems that auto-discover patterns (like all landing pages) are superior to systems where you manually define what to track
You’ll catch marketing changes, new campaigns, or unexpected traffic patterns without manual setup
This reduces blind spots in your monitoring

Principle: Common Sense Over Complexity

Marketing analytics is not rocket science. Apply common sense.

Don’t overcomplicate analysis with complex frameworks or methodologies
If something seems logical to check, check it
Trust your intuition about what data would be useful
The best insights often come from asking obvious questions that others overlook

Principle: Trust Your Expectations, Then Verify

You know what data should exist based on your domain knowledge. Check if it’s there.

If you’re running Facebook ads, you expect to see ad IDs, campaign IDs, referrer data
If users interact with landing pages, you expect CTA click events
Use your expectations to validate data quality - if expected data is missing, something’s wrong
Finding what you expect confirms the system is working; finding what you don’t expect reveals gaps or opportunities

Learning & Exploration

Principle: Exploration-Driven Learning

Discover features by clicking around and exploring, not by studying documentation.

You don’t need YouTube tutorials or manuals to start analyzing data
Start with a need (“I want to see what users are doing”) and explore the tool to find how
Good tools enable natural discovery through intuitive UI
Learning by doing > learning by reading
The tool should support your curiosity, not require training

Principle: Follow the Natural Workflow

Start broad, then drill down into specifics based on what you find.

Begin with high-level views (all landing pages)
Identify patterns or interesting data points
Click through to drill down (individual users, sessions, recordings)
Each layer reveals more detail naturally
The workflow should feel obvious, not learned

Principle: Events → Funnel → Questions

After discovering events, immediately build funnels to understand the flow and spot problems.

Once you see what events exist, create funnels with them
Don’t wait for someone to tell you “we have a drop-off problem”
The funnel visualization will scream the problems at you
Start simple: landing page → first key event → next event, etc.
Anomalies are obvious when visualized (90% drop-off? That’s a massive red flag)

Principle: Let Anomalies Jump Out

Don’t hunt for problems. Build dashboards and let problems reveal themselves.

When you see 90% drop-off between landing page and first screen, that’s not subtle - it’s screaming
Trust your instincts: if a number looks wrong, it probably is
Don’t need statistical analysis for obvious issues
The data will tell you where to dig deeper
Your job is to set up visibility, then observe

Principle: Never Accept Numbers as Abstract - Drill Down to People

Numbers aren’t mysteries. Every aggregate metric represents real users you can investigate.

See 55,000 landing page views dropping to 5,000? Don’t just theorize - click through and see WHO those 50,000 people are
Every person in the drop-off has:
- Session recordings you can watch
- UTM parameters showing their traffic source
- Campaign IDs showing which ad they came from
- Full property data to understand context
Aggregate metrics are starting points, not conclusions
The richness is in the individual-level detail
Don’t accept “we have a bounce problem” - understand which campaigns, which sources, which user segments are bouncing

Principle: Context is Everything

Raw numbers mean nothing without context. Every data point should have rich metadata.

Not just “user bounced” - but which campaign, which landing page variant, which UTM term, which ad set
This context lets you pattern-match across segments
You can discover “all traffic from campaign X bounces” or “users with UTM term Y convert better”
Without this context, you’re just guessing
The tool should make context easily accessible, not buried

Principle: Traffic Quality Persists Through the Funnel

Good traffic converts well at every step. Bad traffic struggles everywhere.

If a campaign drives good traffic, it will outperform at step 1, step 2, step 3, etc.
If a campaign drives bad traffic, it will underperform at every stage
This is an audience quality issue, not a page/UX issue
When different traffic sources show different conversion rates on the SAME page, it’s about who you’re targeting, not what you built

Principle: Isolate Variables to Find Root Cause

When investigating problems, test one variable at a time.

Is it the landing page or the traffic? Compare different campaigns on SAME landing page, then same campaign on DIFFERENT landing pages
If bad traffic performs poorly everywhere (all landing pages), it’s a targeting/audience problem
If traffic performs well on page A but poorly on page B, it’s a page problem
If performance differs on the same page based on traffic source, it’s definitely an audience issue
Most problems are “usually both” - but you need to isolate to know which to fix first

Attribution & Tracking

Principle: End-to-End Attribution is Non-Negotiable

Every sale must trace back to the initial landing UTM parameters. Period. No excuses.

Sales are not mysterious - behind every sale is a user who landed with specific UTM parameters
You MUST persist UTM parameters from landing through the entire funnel to conversion
This is tool-agnostic: PostHog, Google Analytics, Mixpanel, whatever - they all can do this
This is trivial to implement: one day of developer work, maximum
There is no technical excuse for not having this
Without this, you’re flying blind on what’s working

Principle: The Attribution Chain is Sacred

Landing UTM → Funnel Journey → Lead Capture (Email/Phone) → Sale

Step 1: User lands with UTM parameters (utm_source, utm_medium, utm_campaign, utm_term, utm_content, campaign_id, ad_id, etc.)
Step 2: User goes through funnel (could be 1 technology or 10 technologies - doesn’t matter)
Step 3: User provides PII (email/phone) - becomes a lead
Step 4: Lead converts to sale/customer

At Step 3, you have:

A specific person (email/phone)
Their initial landing UTM parameters (from Step 1)
This creates the attribution: “This lead came from campaign X, ad Y, landing page Z”

This attribution must be maintained through to sale.

Principle: PII is the Attribution Anchor

Once you have email/phone, you can connect everything before and after.

Before PII: Anonymous user with session tracking and UTM parameters
After PII: Known lead/customer with identifiable behavior
PII (email/phone) is the bridge that connects anonymous → identified
Once connected, you know: “Customer john@example.com came from Facebook campaign 12023, landing page discount-01, converted in 3 days, LTV $500”

Principle: No System Limitations Excuse

Every modern analytics/CRM system can do end-to-end attribution.

If your system can’t do this, the system is inadequate
If your team says “we can’t track this,” they’re wrong or lazy
This is not a complex technical problem - it’s basic tracking
One day of setup, then it works forever
The data exists - you just need to connect the dots

Principle: First-Touch Attribution is Sacred

Always attribute sales to the original traffic source, not the last click.

User lands from Facebook ad → provide email → that’s first-touch
Days/weeks later they’re retargeted via SMS/email → these are nurture touches
Final conversion might be from email click, SMS click, or direct visit
Attribution goes to original Facebook campaign (first-touch)
Without the original ad, they wouldn’t be in your retargeting list
Retargeting is amplification, not acquisition

Principle: Know What You Want to Check

Having data isn’t magical. Knowing what to check is the skill.

Tools (PostHog, Moon, GA, etc.) store all the connections
The “magic” is knowing to check: Sales → Leads (via email) → Original UTM parameters
Not about advanced tools, but about understanding the data structure
When you know what you want to check, the answer is instant (if data is properly connected)
Most people don’t struggle with tools - they struggle with knowing what questions to ask

Principle: Understand Your Event Footprints (Schema)

Every marketing event has a schema with parameters. These are your attribution Bible.

If you don’t understand the footprints in your data, you’re a bad hunter
Every event has parameters: utm_source, campaign_id, affid, ad_id, etc.
These parameters are the ONLY way you see things
Without reading parameters, you’ll misattribute (thinking SMS sale = Google Ads sale)
Learn your schema once: Notice what IDs exist (PostHog distinct_id), what parameters you capture (affid, campaign_id), then use them
No meetings needed: See PostHog tracking exists → Use it. See affid exists → Segment by it. Simple.

Principle: Parameters Reveal the Truth

Surface-level source labels lie. Parameters tell the real story.

Source says “Google Ads” but parameters show SMS tracking codes → It’s SMS retargeting
Source says “Direct” but parameters show original Facebook campaign → It’s retargeting
Source says “Email” but parameters show campaign_id from paid ad → Attribute to the paid ad
Always read the parameters, not just the source label
Parameters are the definitive record of how user arrived

Principle: Never Trust Charts Without Checking Raw Events

Charts are abstractions. Raw events are reality.

Never look at charts without knowing which raw events formed those charts
Always check a few events manually to see what’s actually happening
Go through the funnel yourself, trigger events, see how they appear in the system
Understand how each event comes through: which parameters, from where, in what shape
Only after understanding raw events can you trust the aggregated charts

Principle: Parameter Schemas Evolve - Track the Changes

Your event schema changes over time. This evolution tells a story.

Change the funnel → Parameters change
Change landing pages → New tracking codes appear
Change payment system → Content IDs format changes
Parameters shift shape: Arrays become numbers, fields disappear, new ones appear
Evolution of schema = history of system changes
Track these patterns over time to understand what changed when
The better you understand schema evolution, the better hunter you are

Principle: Raw Data is Not “Developer Shit” - It’s Your Job

Understanding ugly formatted data is core marketing analytics work, not technical work.

Those “ugly spreadsheets” with IDs, affids, UTMs are THE CORE
Everything else (charts, dashboards, reports) is just a consequence
If you understand the raw parameters, you don’t need anything else
This is not rocket science, not developer territory - this is marketing analytics
Being comfortable with JSON, parameters, raw event data is fundamental
Don’t delegate this to developers - own it yourself

Principle: Use ChatGPT/Google for Unknown Parameters (1 Minute Research)

Don’t know what a parameter means? Look it up. Takes 1 minute.

See gtm-msr.appspot.com/render? Google it → Google Tag Manager checking traffic
See _gl parameter? ChatGPT it → Google linker parameter for cross-domain tracking
See fbclid? Look it up → Facebook Click ID
Every day you’ll encounter unknown parameters - just research them
Build your parameter vocabulary over time
No shame in not knowing - shame is in not looking it up

Data Validation & Cross-Comparison

Principle: Never Trust a Single System - Always Cross-Validate

Never make decisions based on one chart, one system, one source. Always cross-compare.

This is not marketing-specific - it’s how data works, period
Single system view = blind spots, biases, tracking limitations
Always compare: System A vs System B vs System C
Discrepancies tell you about system limitations, not necessarily about reality
Cross-validation is mandatory, not optional

Principle: Your Attribution System is Source of Truth, Not Ad Platforms

Ad platforms (Facebook, Google) undercount due to consent/tracking limitations. Your system sees more.

Facebook Ads dashboard shows X conversions
Your system (PostHog, Moon, billing) shows Y conversions (Y > X)
Your system is more accurate because:
- No consent tracking limitations (first-party data)
- No ad blocker issues
- Server-side tracking persists longer
- You have the actual transaction data
Use your system as source of truth for decision-making
Use ad platform data to understand the tracking gap

Principle: Billing is the Ultimate Source of Truth

When systems disagree, billing system wins. Money doesn’t lie.

Marketing system says 100 sales
Ad platform says 80 sales
Billing system says 95 sales
Billing is correct - these are actual transactions with money exchanged
If marketing system shows more than billing → Overcounting (duplicate events, test transactions)
If marketing system shows less than billing → Undercounting (tracking broken)
Always reconcile marketing data with billing data

Principle: Understand and Adjust for Tracking Limitations

Ad platforms have consent limitations. Factor this into your decisions.

Facebook requires user consent for tracking (GDPR, iOS privacy)
Your system doesn’t require consent (first-party tracking)
Facebook might show 20% fewer conversions than reality
Don’t panic and stop campaigns when Facebook shows low conversions
Check your attribution system first
Adjust/scale campaigns based on real data, with awareness of platform undercounting

Principle: Unattributed Sales Are Normal (5-20%)

Not all sales will have marketing attribution. This is expected.

Marketing shows 100 sales, Billing shows 120 sales → 20 unattributed (16.7%)
Unattributed sales happen: Direct calls, cookie clearing, device switching, tracking failures
5-20% unattributed is normal and acceptable
20-30% is high but manageable
30% means tracking is significantly broken
Create “no-affid” / “unattributed” category for these sales
Adjust marketing ROAS calculations for unattributed percentage
Track this percentage over time - if it spikes, investigate

Principle: Balance Micro and Macro Analysis

You need both detailed event analysis and high-level system comparisons.

Micro: Individual event footprints, parameters, user journeys
Macro: System totals (100 vs 110 vs 120), gap percentages, trend lines
Micro without macro = Missing big picture (50% of sales untracked and you don’t know)
Macro without micro = Can’t fix problems (know there’s a gap, don’t know why)
Daily: Check both levels (macro totals + spot check 5-10 events)
Weekly: Full reconciliation + deep dive if gaps changed
Always be cautious about both perspectives

Meta-Organization

As we document insights:

New Principle: Create a new section
Addition to Existing: Add subsections or bullets under existing principle
Example: Add to the examples section with reference back to the principle