Performance Marketing Knowledge Modules

Display advertising and programmatic buying across Google Display Network (GDN), DV360, and third-party DSPs. Covers targeting taxonomy, campaign setup, retargeting strategies, frequency management, viewability, brand safety, and optimization. Designed for agents managing upper-funnel awareness and mid-funnel retargeting through display channels. Programmatic campaigns follow a similar hierarchy to search/social but add supply-side concepts: Key distinction: In programmatic, you buy impressions on an exchange, not clicks on a platform. The buyer (DSP) bids in real-time auctions for each impression based on targeting rules. Frequency caps by campaign objective: Diminishing returns curve: Fatigue diagnostic: When CTR drops 15-20% from peak over 7 days while frequency rises, creative fatigue is confirmed. Rotate creative, don't just lower frequency. Cross-device: Caps per device = overcounting. A 5/week cap per device means 15/week for a 3-device user. Use people-based frequency via device graphs (deterministic or probabilistic). Cross-channel capping improves effectiveness up to 32% (IAS). Set frequency caps at the line item level. Campaign-level caps don't prevent clustering within a day. MRC/IAB minimum definition: GroupM premium standard: 100% of pixels in view. Video requires 50% completion + sound on. This is 2x the MRC baseline and what premium buyers demand. Viewability by ad size (benchmark): vCPM formula: vCPM = Total Spend / ((Impressions × Viewability%) / 1000) Transact on vCPM, not raw CPM. A $3 CPM at 40% viewability = $7.50 vCPM. A $5 CPM at 80% viewability = $6.25 vCPM — the "expensive" inventory is actually cheaper per viewed impression.

Performance marketing email sequences — automated emails triggered by ad-driven conversions, form submissions, and behavioral events. Covers welcome sequences, abandoned cart recovery, lead nurture, post-purchase, and re-engagement. Scoped to performance email only (sequences that directly support paid acquisition ROI), not general CRM/lifecycle email. This module is platform-agnostic. It works with any ESP (Klaviyo, HubSpot, Mailchimp, ActiveCampaign, Customer.io). Account-specific details come from media-context.md. Email sequences follow a trigger → sequence → branch architecture: Cadence by business model: PLG sequences optimize for activation speed (get user to value fast). B2B sequences optimize for qualification depth (educate over weeks). Applying B2C cadence to B2B causes unsubscribes. Subject lines (28-39 characters optimal for mobile): Personalized: +26% open rate (Experian). Formula: [Name], [curiosity hook]Numbers: +15-20% open rate. Formula: [Number] [adjective] [noun] [promise]Questions: +10-15% open rate (B2B). Formula: Are you [making this mistake]?Urgency: +22% open rate, but fatigues after 3 sends. Use sparingly. Preheader: 40-100 characters. Complement the subject line, don't repeat it. Adding a preheader lifts open rates ~7% (Litmus). Body (optimal length by type): Critical rule: Single CTA per email. Multiple CTAs = +371% fewer clicks (WordStream). Every email has one job. Segment by acquisition source and behavior, not just demographics: Technical requirements before sending any sequence: Warm-up schedule (new domain/IP, Infobip methodology): After 14 days, increase 40% daily based on previous day's delivered volume. First 45 days: exclude users inactive 90+ days. Monitor bounce rate (<2%), spam complaint rate (<0.1%), and inbox placement throughout. Test priority (highest impact first):

Conversion rate optimization for performance marketing landing pages. Covers message match, page structure, A/B testing, friction analysis, form optimization, and page speed. The bridge between ad spend and conversions — a 10% improvement here multiplies the ROI of every upstream dollar. This module focuses on landing pages for paid traffic — pages that receive visitors from ads and must convert them on a specific action. It complements the writing skill pack's copywriting module with performance-specific optimization. A performance landing page has a defined anatomy: Not every page needs every section. The principle: each section moves the visitor one step closer to the CTA. The most common conversion killer in paid traffic. Message match = the ad promise matches the landing page delivery. Message match impact (research-backed): Dynamic Text Replacement matching ad keyword in headline: +31% conversion lift (Unbounce/ConversionLab, 77-day test, 1,274 visitors, 100% significance)Ad-to-page visual congruence (same image): +48% conversion lift (NextAfter, 239K sample)Full messaging continuity (headline + body + CTA aligned): +63% conversion lift (MarketingExperiments)Personalized landing pages convert 25% more mobile users than static (Unbounce) Scoring message match: Strong — Visitor instantly recognizes they're in the right place. Ad headline ≈ page headline. Visual continuity.Moderate — Same topic but different framing. Visitor has to read to confirm relevance.Weak — Generic page used for multiple ads. Visitor questions if they clicked the right thing. Two principles (MarketingExperiments): Continuity (every step states the value prop consistently: ad → page → form → thank-you) and Congruence (every element on the page supports the same value prop). Common mismatch patterns:

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. Marketing analytics operates across three layers: Each layer has distinct concerns. A mistake in collection propagates through everything downstream. Tracking hierarchy (install in this order): Google Tag Manager (container for all tags)GA4 (analytics baseline)Ad platform pixels (Meta, Google Ads, LinkedIn, TikTok)Server-side endpoints (CAPI for Meta, server-side GTM)Enhanced conversions / Advanced MatchingOffline conversion import pipeline The attribution truth hierarchy: Backend/CRM data — actual revenue, actual leads. Ground truth.Incrementality tests — what would have happened without the ad? Causal truth.Media Mix Models (MMM) — statistical allocation across channels. Directional.Multi-touch attribution (MTA) — touchpoint-level credit. Useful but biased.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. The gold standard for measuring true ad impact. Each method has specific minimum requirements. Geo lift test design (GeoLift methodology):

Campaign management for Google Ads and Microsoft Ads search campaigns. Covers campaign structure, keyword strategy, match types, ad copy (RSA), Quality Score, bidding strategies, negative keywords, and ongoing optimization. Designed for agents that build, audit, and optimize search campaigns autonomously. This module is account-agnostic. Account-specific details (budgets, products, audiences) come from the media-context.md file provided at runtime. Search campaigns follow a strict hierarchy: Each level has one job: Campaign = budget container + targeting scope (geo, device, schedule)Ad Group = theme container. One intent per ad group.Keywords = the queries you want to matchAds = the message shown for those queries Match type strategy: Start with exact + phrase for proven terms. Use broad only with smart bidding and sufficient conversion data (30+ conversions/month at campaign level). Three components, weighted approximately: Quality Score impacts: CPC (higher QS = lower CPC), ad rank, eligibility for extensions, top-of-page thresholds. Learning phase: After any bidding change, allow 1-2 weeks (or 50 conversions) for the algorithm to stabilize. Do not judge performance during learning. PLG/freemium caveat: If the conversion event is a free signup (zero friction), platforms will optimize for signup volume — which may not correlate with paid conversion. For PLG products, import downstream conversion events (trial-to-paid, activation milestones) as offline conversions and optimize toward those instead. Without this, smart bidding will flood you with low-quality free signups.

Campaign management for Meta Ads (Facebook/Instagram), LinkedIn Ads, and TikTok Ads. Covers campaign architecture, audience strategy, creative testing, ad copywriting, bidding, and optimization. Designed for agents that plan, build, and optimize paid social campaigns autonomously. This module is platform-agnostic within social. Platform-specific specs and constraints are in references/platform-specs.md. Account-specific details come from media-context.md. All major social ad platforms share a three-level hierarchy: Percentage and performance (Meta, directional benchmarks): Seed quality hierarchy (most to least effective): Top 25% customers by LTV (value-based) — 20-40% lower CPA vs all-purchaser seedsAll purchasers (value-based lookalike) — 15-30% lower CPA vs binary event seedsAll purchasers (binary event) — baselineAdd-to-cart users — moderate qualityPage visitors — 2-4x worse CPA vs purchaser seeds Seed size: Meta minimum is 100; practical minimum is 1,000-5,000. Optimal range: 5,000-50,000. Above 50K, quality dilution in the seed itself can hurt. Refresh cadence: Every 30-90 days. Website visitor seeds refresh every 14-30 days. Performance degradation typically visible after 60-90 days (rising frequency, declining CTR). Build on auto-updating Custom Audiences to partially mitigate. Note: With Advantage+ and strong pixel data, broad targeting (no lookalike) increasingly matches or outperforms lookalikes. Test broad vs LAL before assuming LAL is always better.