# Attribution > >- - **Type:** Skill - **Install:** `agentstack add skill-leadmagic-gtm-skills-attribution` - **Verified:** Yes — security-reviewed for prompt injection and unsafe behavior - **Seller:** [LeadMagic](https://agentstack.voostack.com/s/leadmagic) - **Installs:** 0 - **Category:** [Communication](https://agentstack.voostack.com/c/communication) - **Latest version:** 0.1.0 - **License:** MIT - **Upstream author:** [LeadMagic](https://github.com/LeadMagic) - **Source:** https://github.com/LeadMagic/gtm-skills/tree/main/skills/analytics/attribution - **Website:** https://github.com/LeadMagic/gtm-skills ## Install ```sh agentstack add skill-leadmagic-gtm-skills-attribution ``` Requires the [AgentStack CLI](https://agentstack.voostack.com/docs/cli). Works with Claude Code, Cursor, and any MCP-compatible agent. ## About # Attribution ## Overview Attribution answers the existential marketing question: which of our activities actually produce revenue? The core principle is that single-touch attribution lies. First-touch attribution over-credits awareness channels. Last-touch attribution over-credits bottom-of-funnel conversion channels. Both lead to systematically bad investment decisions — over-funding channels that appear in the credited position and starving channels that create the conditions for conversion. The non-obvious rule: the "best" attribution model depends on your GTM motion. Product-led growth companies should weight product-qualified signals higher. Sales-led companies should weight sales engagement higher. Channel-led companies should weight partner influence. There is no universal attribution model — only the model that matches how your customers actually buy. This skill produces: a Multi-Touch Attribution Model Report comparing 4-6 attribution models on revenue credit allocation, channel ROI calculations with cost and revenue attribution, a UTM governance framework with taxonomy and enforcement rules, a source-of-truth reporting structure, and optimization recommendations for budget reallocation. ## When to Use - User says "attribution" or "attribution model" → activate this skill - User asks "which channel generates the most revenue" → use this skill - User says "marketing ROI" or "campaign ROI" → attribution is required - User mentions "UTM tracking" or "UTM hygiene" or "UTM parameters" → use UTM governance module - User asks "how do I measure multi-touch attribution" → implement a model - User says "source of truth reporting" or "marketing and sales disagree on numbers" → use this skill - Trigger phrases: attribution modeling, multi-touch, first-touch, last-touch, channel attribution, campaign ROI, marketing ROI, UTM parameters, tracking governance Do NOT use for: - Campaign-level analytics (open rates, reply rates, meeting rates) → use campaign-analytics - Pipeline management and forecasting → use pipeline-management - A/B testing methodology → use a-b-testing - CRM data architecture → use crm-integration - Real-time channel monitoring and alerting → use proactive-alerts ## Authoritative Foundations This skill draws from the following established methodologies: - **Bizible (Marketo/Marketo Measure) — Multi-Touch Attribution** — The most widely deployed B2B multi-touch attribution system. Bizible's model maps every marketing and sales touch across the customer journey, weighting touches based on position (FT, LC, even, U-shaped, W-shaped, custom). Acquired by Adobe/Marketo, it's the enterprise standard. - **Full Circle Insights (Salesforce-native)** — Attribution built inside the CRM rather than appended to it. The implicit criticism of marketing-platform-first attribution is that it misses sales touches. Full Circle captures both marketing and sales activity within the CRM. - **Google Analytics — Attribution Models** — The most accessible attribution implementation, now with data-driven attribution as the default in GA4. Google's shift from rules-based to machine-learning-based attribution reflects the industry trend toward algorithmic models. - **DreamData — B2B Attribution** — Account-based attribution for companies where multiple stakeholders influence a deal. Critical distinction: person-level attribution credits the individual who converted; account-level attribution distributes credit across the buying group. ## Prerequisites - CRM data with opportunity records including: create date, close date, deal amount, primary campaign source (HubSpot, Salesforce, or Attio) - Marketing platform data with touch history: email opens, clicks, form fills, content downloads, webinar attendance, ad clicks (Marketo, HubSpot, Pardot, or GA4) - Sales activity data: calls, emails, meetings logged on opportunities (CRM activity history) - UTM-tagged URLs across all marketing channels (website, ads, email, social, partner) - Cost data by channel: total spend per channel per period for ROI calculation - Recommended: 12+ months of touchpoint data for stable attribution modeling - Recommended upstream skills: crm-integration (for CRM data architecture), campaign-analytics (for campaign-level metrics) ## Step-by-Step Process ### Phase 1: Intake Gather required information from the user. Ask all questions at once. Do not proceed until all answers are received. Required intake questions: 1. **GTM motion**: What is your primary GTM motion? (Product-led growth, sales-led, channel/partner-led, marketing-led, hybrid?) 2. **Average sales cycle**: How long from first touch to closed-won? (30 days, 90 days, 180 days, 12+ months?) This determines the lookback window. 3. **Buying group size**: How many people typically influence a purchase? (Single decision-maker, 3-5 person committee, 6+ enterprise buying group?) Determines person-level vs account-level attribution. 4. **Channel mix**: What channels are active? (Outbound email, inbound/content, paid ads, events, partners, referrals, social, direct, product-led?) 5. **Data sources**: Where are touchpoints tracked? (CRM, marketing automation, product analytics, ad platforms, webinar platform?) 6. **Current attribution**: What attribution model, if any, is currently in use? What metrics does marketing currently report? What does sales report? 7. **UTM status**: Are UTMs consistently applied? Is there a UTM taxonomy documented? Who enforces it? 8. **Budget data**: Can you provide spend data by channel for the analysis period? ### Phase 2: UTM Governance and Hygiene Audit UTM parameters are the foundation of attribution accuracy. Bad UTMs = bad attribution. 1. **UTM taxonomy design** — enforce the standard 5 parameters: - `utm_source`: The platform/site sending traffic (google, linkedin, hubspot, marketo, outreach). Must be a controlled value from an approved list. - `utm_medium`: The marketing medium (cpc, organic_social, email, display, referral, partner, event). Must be a controlled value. - `utm_campaign`: The specific campaign name. Use a consistent naming convention: `[quarter]-[channel]-[audience]-[offer]-[variant]`. Example: `q1-2024-email-enterprise-compliance-toolkit-v2`. - `utm_content`: What was clicked (cta-text, banner-blue, subject-line-test-a). Used for A/B testing within campaigns. - `utm_term`: Search keywords (only for paid search). Optional. 2. **UTM hygiene audit** — scan all URLs for common errors: - Missing `utm_source` (most common error — without it, traffic is "direct") - Inconsistent source naming: "linkedin," "LinkedIn," "linkedin.com," "LinkedIn Ads" are 4 different sources. - Using `utm_medium=cpc` for non-paid traffic - Spaces in UTM values (should use hyphens or underscores, never spaces) - Case sensitivity: UTM values are case-sensitive. "Google" ≠ "google." - Campaign parameter used as catch-all (same utm_campaign across multiple channels) - Missing utm_campaign (leaves traffic unassigned to any campaign for ROI calculation) - URL shorteners stripping UTM parameters (bit.ly, ow.ly strip by default — configure to preserve) - Internal links tagged with UTM parameters (overwrites real attribution data) 3. **UTM governance rules** — document and enforce: - All outbound links from marketing must have at minimum: utm_source, utm_medium, utm_campaign. - Source values must come from pre-approved list (requires request to add new source). - Medium values must come from 10 standard categories. - Campaign naming convention is mandatory and validated at URL creation. - Marketing automation platforms must auto-append UTMs (configure at the platform level). - Sales-provided links should use a different tracking parameter (or UTM with sales-specific source) to distinguish sales vs marketing influence. 4. **UTM data quality score** — calculate for the analysis period: - % of revenue-attributed opportunities with complete UTM data: target >90% - % of website sessions with utm_source: target >80% - % of campaign spend assigned to specific campaigns via UTM: target >95% - If scores are below targets, attribution modeling will be unreliable. Fix UTM hygiene first. ### Phase 3: Touchpoint Data Assembly Build the unified touchpoint timeline: 1. **Define the touchpoint taxonomy** — categorize every interaction: - **Marketing touches**: Email click, content download, webinar attendance, ad click, website visit, form fill, social engagement, event booth visit. - **Sales touches**: Sales email, sales call, meeting held, demo delivered, proposal sent, LinkedIn connection/inmail. - **Product touches**: Signup, activation event, invite team member, upgrade feature usage, PQL trigger. - **Partner touches**: Partner referral, co-marketing event, marketplace listing click. 2. **Build touchpoint timelines** for each opportunity: - For every closed-won and closed-lost opportunity, assemble all touchpoints from first interaction to close date. - Each touchpoint records: timestamp, channel, touch type, campaign association, and whether it was marketing or sales initiated. - For account-level attribution: roll up all contacts under the account and create a merged timeline. 3. **Define the attribution window (lookback period)**: - Standard: 90 days for SMB (shorter sales cycles), 180 days for mid-market, 365 days for enterprise. - Custom window: if your average sales cycle is 120 days, use 120-day lookback. - Touches outside the window are excluded from attribution but recorded for awareness contribution analysis. 4. **Handle anonymous-to-known matching**: - Website visits before form fill are anonymous. Use cookie-based matching (Google Analytics User ID, HubSpot cookie) if available. - If no matching possible, note the gap in data and acknowledge that early-stage awareness touches may be undercounted. ### Phase 4: Multi-Touch Attribution Model Construction Build and compare multiple attribution models: **Model 1: First-Touch Attribution** - 100% of credit to the very first touch. - Best for: understanding which channels create awareness and initial interest. - Worst for: understanding which channels close deals. - Bias: over-credits top-of-funnel channels (content, social, ads). **Model 2: Last-Touch Attribution** - 100% of credit to the touch immediately before opportunity creation or deal close. - Best for: understanding which channels drive conversion. - Worst for: understanding the full buyer journey. - Bias: over-credits bottom-of-funnel channels (direct, branded search, sales outreach). **Model 3: Linear Attribution** - Equal credit distributed across all touches. - Example: 10 touches = 10% credit each. - Best for: balanced view when all touches are considered equally valuable. - Worst for: companies where certain touches (demo, trial) are objectively more influential. - Bias: over-credits high-frequency, low-value touches (email opens) and under-credits high-effort, high-value touches (in-person meetings). **Model 4: Time-Decay Attribution** - Credit increases exponentially as touches get closer to conversion. - Example: 7-day half-life means a touch 14 days before close gets 4x the credit of a touch 28 days before close. - Best for: long sales cycles where recency strongly correlates with influence. - Worst for: short sales cycles or when early-stage education is critical. - Bias: similar to last-touch but less extreme. Still under-credits awareness. **Model 5: U-Shaped (Position-Based) Attribution** - 40% to first touch, 40% to lead conversion touch (or last touch), 20% distributed evenly across middle touches. - Best for: B2B where both awareness creation and conversion are critical. - Worst for: transactional sales where middle touches are important (evaluation, demo). - The 40/40/20 split is common but arbitrary. Test 30/30/40 or 35/35/30 splits. **Model 6: W-Shaped Attribution** - 30% to first touch, 30% to lead creation touch, 30% to opportunity creation touch, 10% distributed across remaining touches. - Best for: complex enterprise B2B with distinct buying stages (awareness → consideration → decision). - Requires clean data on when each milestone occurred. **Model 7 (Advanced): Custom Weighted / Algorithmic Attribution** - Use machine learning (Shapley values, Markov chains) to determine touch weights from historical data. - Best for: companies with 500+ closed-won deals and reliable touch data. - Available in: Google Analytics 4 (data-driven attribution), Bizible, DreamData, CaliberMind. - If implementing custom: use logistic regression with conversion as the dependent variable and touch channel counts as independent variables to estimate channel coefficients. ### Phase 5: Model Comparison and Selection Compare attribution models side by side: 1. **Revenue attribution by channel** across all models: - Create a matrix: channels as rows, models as columns, cell = % of total revenue attributed. - Calculate variance: for each channel, what's the range between the most and least generous model? - Channels with wide variance are sensitive to model choice — important to get right. - Channels with narrow variance are robust — model choice doesn't matter much. 2. **Model selection criteria**: - **GTM motion fit**: Sales-led → W-shaped or U-shaped (both marketing and sales get credit). Product-led → time-decay (product activation signals should weight highest). Marketing-led → linear or U-shaped (full funnel view). - **Data completeness**: If you can't reliably identify first touch (cookie gaps, long history), don't use first-touch or U-shaped. If you can't identify lead creation moment, don't use W-shaped. - **Stakeholder alignment**: The model both marketing and sales agree on is better than the "correct" model only marketing trusts. Attribution is as much political as analytical. - **Simplicity**: A linear model that everyone understands and acts on beats a custom algorithmic model that nobody trusts or can explain. 3. **Recommend a primary model** with justification. Also report 1-2 comparison models so stakeholders see the sensitivity. ### Phase 6: Channel ROI Calculation Calculate return on investment by channel: 1. **Channel cost assembly** — collect all costs per channel: - People cost: fully loaded salary for team members dedicated to the channel - Media spend: ad spend, sponsorship, content promotion - Tool cost: channel-specific tools (Apollo for outbound, SEMrush for SEO, webinar platform) - Content cost: content created specifically for the channel - Agency/vendor cost: external support for the channel - Event cost: travel, booth, sponsorship, swag for event channel 2. **Channel revenue attribution** — using the selected primary model: - For each channel, sum attributed revenue from closed-won deals in the analysis period. - Separate new customer revenue from expansion revenue (different channels may drive each). 3. **Channel ROI calculation**: ``` Channel ROI = (Attributed Revenue - Channel Cost) ÷ Channel Cost ``` Express as ratio (3:1) or percentage (300%). 4. **Channel CAC calculation**: ``` Channel CAC = Total Channel Cost ÷ New Customers Attributed to Channel ``` Compare to company-wide CAC. Channels with >2x company-wide CAC are inefficient. 5. **Channel efficiency metrics**: - Cost per opportunity: channel cost ÷ opportunities attributed - Cost per meeting: channel cost ÷ meetings attributed (if meetings are trackable) - Pipeline efficiency: attributed pipeline value ÷ channel cost - Time to payback: average months to recover channel cost from attributed revenue 6. **Assisted vs unassisted conversion** (critical distinction): - **Assisted conversions**: channel appeared in the touch history but wasn't the primary-attributed channel. These channels "assist" other channels in closing. - **Unassisted … ## Source & license This open-source skill is cataloged on AgentStack and links to its original source — we do not rehost the code. - **Author:** [LeadMagic](https://github.com/LeadMagic) - **Source:** [LeadMagic/gtm-skills](https://github.com/LeadMagic/gtm-skills) - **License:** MIT - **Homepage:** https://github.com/LeadMagic/gtm-skills Install and usage instructions live in the source repository linked above. ## Pricing - **Free** — Free ## Security capabilities Automated source analysis of v0.1.0 — what this tool can access: - **Network access:** no - **Filesystem access:** no - **Shell / process execution:** no - **Environment & secrets:** no - **Dynamic code execution:** no *"Yes" means the capability is present in the source — more access means more to trust, not that it is unsafe.* ## Versions - **0.1.0** — security scan: passed — Imported from the upstream source. ## Links - Listing page: https://agentstack.voostack.com/l/skill-leadmagic-gtm-skills-attribution - Seller: https://agentstack.voostack.com/s/leadmagic - Browse the marketplace: https://agentstack.voostack.com/browse --- Listed on AgentStack — the marketplace for AI agent skills and MCP servers. Every listing is security-reviewed. Creators keep 70%.