Marketing Analytics — Performance Marketing Knowledge Module

Marketing Analytics

Measurement, attribution, and optimization frameworks for performance marketing. Covers conversion tracking setup, attribution models, incrementality testing, budget allocation, campaign auditing, and dashboard design. The measurement backbone that all other marketing modules depend on.

Content Structure

Marketing analytics operates across three layers:

Data Collection (tracking, tagging, pixel setup)
    ↓
Data Processing (attribution, deduplication, validation)
    ↓
Data Activation (reporting, optimization, budget allocation)

Each layer has distinct concerns. A mistake in collection propagates through everything downstream.

Key Concepts

Conversion Tracking Setup

Component	What	Platform
Pixel/tag	JavaScript snippet on your site that fires on events	Meta Pixel, Google Tag, LinkedIn Insight Tag, TikTok Pixel
Server-side tracking	API-based event sending (bypasses browser restrictions)	Meta CAPI, Google Server-side GTM, LinkedIn CAPI
Conversion events	Specific actions you want to measure	Purchase, lead form submit, signup, add to cart
Enhanced conversions	First-party data matching for better attribution	Google Enhanced Conversions, Meta Advanced Matching
Offline conversions	Import backend/CRM data back to ad platforms	All major platforms support offline import

Tracking hierarchy (install in this order):

1. Google Tag Manager (container for all tags)
2. GA4 (analytics baseline)
3. Ad platform pixels (Meta, Google Ads, LinkedIn, TikTok)
4. Server-side endpoints (CAPI for Meta, server-side GTM)
5. Enhanced conversions / Advanced Matching
6. Offline conversion import pipeline

Attribution Models

Model	How It Works	Best For	Weakness
Last click	100% credit to last clicked ad	Direct response, search	Ignores upper funnel
First click	100% credit to first touchpoint	Awareness campaigns	Ignores conversion assist
Linear	Equal credit to all touchpoints	Understanding full journey	Overvalues passive touches
Time decay	More credit to recent touchpoints	Longer sales cycles	Undervalues discovery
Position-based (U-shaped)	40/20/40 to first, middle, last	Balanced view	Arbitrary weighting
Data-driven (DDA)	ML-assigned credit based on path analysis	Sufficient data (300+ conv/month)	Black box, needs volume
Platform-reported	Each platform claims credit for its touches	N/A — this is what platforms report	Every platform over-counts

The attribution truth hierarchy:

1. Backend/CRM data — actual revenue, actual leads. Ground truth.
2. Incrementality tests — what would have happened without the ad? Causal truth.
3. Media Mix Models (MMM) — statistical allocation across channels. Directional.
4. Multi-touch attribution (MTA) — touchpoint-level credit. Useful but biased.
5. Platform-reported — each platform's self-reported numbers. Always over-counts.

Never use platform-reported numbers alone for budget decisions. Always reconcile with backend data.

Blind spots all models share:

• Viral/referral loops — PLG products where users invite teams (e.g., Notion, Slack, Figma) generate growth that no attribution model captures. If 40% of signups come from team invites, attribution over-credits paid channels by 40%.
• Brand halo — organic search for brand terms is often created by paid awareness campaigns. Last-click gives search 100% credit; the reality is shared.
• Dark social — word-of-mouth, Slack/Discord sharing, private messages. Untraceable but often the largest growth channel for PLG products.

Incrementality Testing

The gold standard for measuring true ad impact. Each method has specific minimum requirements.

Test Type	How	Min Duration	Min Budget/Scale	Best For
Geo lift	Run ads in some regions, hold out others	2-4 weeks + cooldown	10+ geos, 6mo+ history	Channel-level incrementality
Meta Conversion Lift	Platform holdout (5-20% of audience)	2-4 weeks	100+ conv/week, $5K+ annual	Meta campaign incrementality
Google Conversion Lift	Platform holdout (Bayesian)	14+ days	$5K+ budget	Google campaign incrementality
On/off test	Pause channel, measure total impact	2-4 weeks	Willingness to lose channel revenue	Is this channel incremental at all?
Ghost bidding	Bid in auction, don't show ad	2-4 weeks	Programmatic only	Display/programmatic lift

Geo lift test design (GeoLift methodology):

• Minimum geos: 10, ideally 20+ for robust synthetic control
• Pre-test data: 2x the experiment length (e.g., 12-week test needs 6 months history)
• Test markets: between 2 and half of total geos
• Significance level: alpha = 0.1 (90% confidence is standard for geo tests)
• Power target: 80% minimum
• MDE is simulation-derived, not formula-based — run power simulations across effect sizes

Platform lift study requirements:

Platform	Type	Min Budget	Min Conversions	Duration
Meta	Conversion Lift	$5K annual	100/week	2-4 weeks
Meta	Brand Lift	$120K (US)	N/A (survey)	2-4 weeks
Google	Conversion Lift	$5K	1,000 total (directional)	14+ days
Google	Brand Lift	$10K (US)	1.5M impressions	10+ days
TikTok	Brand Lift	$30K (US)	N/A (survey)	3-4 weeks

Power calculation for holdout tests (two-proportion z-test):

n_per_group = (z_α/2 × √(2p̄(1-p̄)) + z_β × √(p₁(1-p₁) + p₂(1-p₂)))² / (p₁ - p₂)²

Where p₁ = control CVR, p₂ = expected treatment CVR, z_α/2 = 1.96 (95%), z_β = 0.84 (80% power).

Practical minimum: 10,000 users in your smallest group for typical marketing conversion rates (1-4%).

Campaign Audit Framework

Structured audit across six dimensions:

Dimension	What to Check	Red Flags
Structure	Campaign/ad group organization, naming, segmentation	Mixed intents in one ad group, no naming convention
Spend efficiency	CPA/ROAS by segment, wasted spend, budget pacing	>20% spend on non-converting segments
Tracking	Conversion setup, event firing, data match rate	Platform conversions ≠ backend by >20%
Creative	Performance by creative, fatigue signals, testing cadence	Same creative running >30 days without test
Audience	Targeting quality, overlap, exclusions, funnel alignment	No exclusions between funnel stages
Bidding	Strategy appropriateness, learning phase status, target setting	tCPA target 50% below actual CPA

Budget Allocation Framework

Method 1: Historical performance allocation

Channel budget share = (Channel conversions × Channel CPA efficiency score) / Total weighted conversions

Where CPA efficiency score = 1 / (Channel CPA / Average CPA). Channels with below-average CPA get more budget.

Method 2: Marginal CPA allocation

• Plot CPA vs spend for each channel (diminishing returns curve)
• Allocate next dollar to the channel with the lowest marginal CPA
• Stop when marginal CPA exceeds target for all channels

Method 3: Portfolio allocation

• Set minimum viable spend per channel (below which data is insufficient)
• Allocate remaining budget proportional to ROAS/CPA efficiency
• Reserve 10-15% for testing new channels/campaigns
• Rebalance monthly based on trailing performance

Starting-point allocation by business model (adjust with data after 30 days):

Channel	PLG SaaS	Sales-led B2B	B2C Ecommerce
Paid Search	30-40%	25-35%	35-45%
Paid Social (Meta)	25-35%	5-10% (retargeting)	25-35%
LinkedIn	0-5%	20-30%	0%
Display/Programmatic	5-10%	5-10%	10-15%
Content/SEO	15-20%	15-20%	5-10%
Email/Nurture	5%	5-10%	5-10%
Testing reserve	10%	10%	10%

Key difference: B2B prioritizes LinkedIn (20-30%) while PLG/B2C prioritizes Meta. Search is always significant but highest for ecommerce (high-intent product searches).

Measurement Maturity Model

Self-assessment framework — identify current level, build a roadmap to the next.

Level	Description	Tools	Advancement Criteria
1 — Ad Hoc	Siloed platform data, basic metrics, no cross-channel view	Platform dashboards only	Implement GA4 + UTM taxonomy
2 — Foundational	Centralized analytics, consistent UTMs, basic reporting	GA4, Looker Studio, UTM builder	Implement multi-touch attribution
3 — Attribution	MTA active, backend reconciliation, cross-channel view	GA4 DDA, CRM integration, BI tool	Run first incrementality test
4 — Incrementality	Regular lift tests, causal measurement, test-and-learn culture	Geo-lift tools, platform lift studies	Implement MMM
5 — Modeling	MMM + incrementality + platform data, triangulated decisions	PyMC-Marketing/Robyn/Meridian, data science team	Continuous optimization loop

Most accounts should target Level 3 and run periodic Level 4 tests. Level 5 requires dedicated data science and $1M+ annual spend.

Budget Pacing

Formula	Calculation	Use
Target Daily Spend	Monthly budget / days remaining	Daily pacing target
Pacing %	(Actual spend / Expected spend) × 100	Over/under-pacing detection
Projected Monthly	(Spend to date / Days elapsed) × Days in month	End-of-month projection

Pacing strategies:

• Even pacing — Spread evenly across the month. Simple but consistently underperforms (Guhl et al., 2025).
• Waterlevel pacing — Spend proportional to traffic patterns. Outperforms all heuristics in empirical studies. Formula: ideal_budget(t) = daily_budget × (traffic_in_period(t) / total_daily_traffic)
• Front-loaded — Spend more early in the month. Use for time-sensitive campaigns or when learning phase matters.

Day-of-week CPM patterns (Gupta Media, billions of impressions):

• Weekdays ~1.4% more expensive than weekends on Meta
• Friday is peak CPM on Meta; Thursday is peak on TikTok
• Weekends are 5-7% cheaper on TikTok
• Adjust daily targets by these patterns rather than spending evenly

Alert thresholds (three-tier):

Severity	Threshold	Response
Low	10-20% deviation from 30-day rolling baseline	Review within 24h
Medium	20-40% deviation OR consistent drift 3+ days	Investigate within 4h
High	40%+ deviation, zero conversions, tracking failure	Immediate action

Use z-scores on 30-day rolling baselines, separated by day-of-week. 2σ = warning, 3σ = critical. Target false positive rate under 30%.

Dashboard Design

A performance marketing dashboard answers three questions:

1. Are we on track? — KPIs vs targets (pace and actual)
2. What's working? — Top/bottom performers by channel, campaign, creative
3. What should we do? — Trends that require action

Essential dashboard sections:

Section	Metrics	Granularity
KPI scorecard	Spend, conversions, CPA, ROAS, vs target, vs prior period	Daily/weekly/monthly
Channel breakdown	Spend, conv, CPA, ROAS by channel	Weekly
Campaign performance	Top 10 / bottom 10 by primary KPI	Weekly
Creative performance	CTR, conv rate, CPA by creative	Weekly
Funnel metrics	Impressions → clicks → visits → conversions (with drop-off rates)	Weekly
Pacing	Budget spent vs plan, projected end-of-month	Daily
Trend lines	CPA and ROAS trailing 4 weeks	Daily

Inputs & Outputs

Inputs:

• KPI targets and budget (from media-context.md)
• Platform data exports (Google Ads, Meta, LinkedIn, TikTok)
• Analytics data (GA4, backend)
• CRM/backend conversion data
• Historical performance data

Outputs:

• Performance reports (executive summary + detailed)
• Attribution analysis
• Budget allocation recommendations
• Campaign audit findings with prioritized actions
• Tracking audit with fix list
• Dashboard specifications

Modes

Mode	What You're Doing
Tracking setup	Configuring pixels, events, server-side, UTMs
Audit	Assessing campaign health across six dimensions
Report	Building performance reports against KPIs
Optimize	Recommending budget shifts, bid changes, pauses
Attribution	Analyzing cross-channel credit, designing incrementality tests

Common Tasks

1. Tracking audit — Verify conversion measurement:

   • Check all pixels fire correctly (use Tag Assistant, Meta Pixel Helper)
   • Compare platform-reported conversions to GA4 to backend
   • Verify UTM parameter consistency across all campaigns
   • Test conversion events on staging before production
   • Confirm server-side tracking is active and matching
   • Document any gaps and their impact on reported numbers

2. Campaign performance report — Weekly/monthly report:

   • Pull data from all active platforms
   • Normalize attribution windows for cross-channel comparison
   • Calculate blended and channel-level KPIs
   • Compare vs targets and vs prior period
   • Identify top 3 opportunities and top 3 risks
   • Deliver executive summary + detailed appendix

3. Budget reallocation — Optimize spend across channels:

   • Calculate CPA/ROAS efficiency by channel
   • Identify channels with headroom (below target CPA, impression share available)
   • Identify channels at diminishing returns (CPA rising with scale)
   • Propose reallocation with expected impact
   • Set review date to validate reallocation impact

4. Campaign audit — Full health check:

   • Score each dimension (structure, spend, tracking, creative, audience, bidding)
   • Prioritize findings by impact (estimated savings or conversion lift)
   • Deliver action items with owner and timeline
   • Schedule follow-up audit in 30 days

5. Design incrementality test — Measure true channel lift:

   • Choose test type (geo lift, conversion lift, on/off)
   • Define test and control groups
   • Calculate minimum duration and sample size
   • Set primary metric and minimum detectable effect
   • Plan analysis methodology
   • Document results and implications for budget allocation

Tips

• Reconcile platform data with backend weekly. Discrepancies grow silently and lead to bad budget decisions.
• Attribution is an opinion, not a fact. Use multiple lenses (platform, MTA, MMM, incrementality) and triangulate.
• Don't optimize a metric you don't trust. If tracking is broken, fix tracking before optimizing campaigns.
• Leading indicators (CTR, CPC, CPM) predict lagging indicators (CPA, ROAS). Monitor both, act on leading indicators early.
• Budget allocation is the highest-leverage optimization. Moving $10K from a 3:1 ROAS channel to a 5:1 ROAS channel creates more value than any single campaign tweak.
• Dashboard design matters. If the dashboard doesn't answer "what should I do differently?" it's just a data display.

Gotchas

• Platform conversion counting differences — Google counts conversions per keyword (can count multiple per click). Meta counts per user within attribution window. LinkedIn uses longer windows. Apples-to-apples comparison requires normalization.
• View-through attribution inflation — Display and video campaigns claim view-through conversions liberally. A user who saw a banner and converted via search 7 days later is unlikely a display-driven conversion. Use conservative windows (1-day view-through max) or discount view-throughs.
• Last-click bias — Defaulting to last-click attribution makes search and brand look great while making prospecting and display look terrible. This leads to over-investing in bottom-funnel and under-investing in demand generation.
• Sample size fallacy — "Campaign A has 50% better CPA than Campaign B" means nothing with 10 conversions each. Use significance calculators before declaring winners.
• Seasonality blindness — Comparing this week to last week without accounting for seasonality (holidays, paydays, events) leads to false conclusions. Compare same week last year for seasonal businesses.
• Vanity metrics — CTR, engagement rate, and video views feel good but don't pay bills. Always tie analysis back to business-outcome metrics (revenue, qualified leads, LTV).

References

• references/attribution-models.md — detailed model comparison, implementation guides, tool recommendations

Related Modules

• paid-search — search campaign data for attribution
• paid-social — social campaign data, platform attribution windows
• display-programmatic — view-through attribution, incrementality testing
• landing-pages — conversion rate data for full-funnel analysis