No dashboard is the true count, and you should never sum conversions across platforms. If Google and Meta both claim one sale, you have one conversion with two claimants. Pick one source of truth for the count, usually your CRM or backend where the money is real. Every other source explains where those conversions came from, not how many there were. Each source is biased in a known direction. Ad platforms over-count themselves through view-through and their own windows, GA4 pushes the unknown into direct, CRMs depend on what reps and forms captured, and surveys favor the memorable touch. Trust directional agreement over exact matches, use self-reported attribution and incrementality tests as tiebreakers, and report the gap honestly instead of forcing the numbers to tie.
How the Marketing Attribution skill works
- 1
Pick the pillar and boundaries
The agent decides whether you need interpretation (choosing a model, reconciling numbers) or first-party build work (instrumenting attribution on a site you control). It reads your product context for business type and sales cycle, and hands off what it doesn't own: event tracking and UTMs to analytics, pixels and CAPI to conversion-tracking, pipeline stages to revops.
- 2
Set the source of truth
It names one system that defines the conversion count, the CRM for B2B or the store backend for e-commerce, and de-duplicates every platform's claims against that total. It also checks that each source counts the same event, since a form fill, a key event and a closed deal are different things.
- 3
Compare models side by side
For long cycles, it shows first-touch and last-touch together, because the gap between them is the insight. It explains how each model distorts the picture (first-touch, last-touch, last non-direct, linear, time-decay, position-based, data-driven) and notes that Google Ads and GA4 now offer only data-driven and last-click. Data-driven needs about 200 conversions and 2,000 ad interactions in 30 days to be reliable.
- 4
Expose the blind spots
It digs into direct (the junk drawer), branded search (where demand created elsewhere cashes out), dark social and AI-assistant influence. It adds a "How did you hear about us?" field at the moment of conversion, with a pick-list plus free text, and discounts survey share instead of treating recall as credit.
- 5
Match rigor to the decision
Small budgets get clean UTMs, last non-direct and a survey. Mid budgets add MTA plus periodic incrementality tests on the biggest line items. Large or offline-heavy spend adds MMM, which needs 2–3 years of weekly data with real spend variation, validated by geo holdouts. Incremental CPA (spend divided by incremental conversions) is what drives budget.
- 6
Build first-party attribution
When you control the surface, it closes the identify() gap at each real conversion and stitches third-party bookings or checkouts through metadata passthrough and a webhook merge. It applies fail-closed anonymity guards, excludes OAuth and checkout referrers from first-touch, and syncs source, confidence and basis onto the CRM record, rolled up to the account.
When to use it
Your agent reaches for this skill when you ask for things like:
What's inside
The playbook covers
Reference library — 4 deep-dives
See it in action
Google Ads says we got 50 conversions last month, Meta says 40, GA4 says 60, and our CRM shows 35 closed deals. Which one is right? I need to know our real numbers before I set next quarter's budget.
Google Ads 50 and Meta 40 are platform conversions, probably leads or form fills counted under each platform's own window and view-through rules. GA4's 60 is likely key events. Your CRM's 35 is closed deals. Those are three different things, so none of them is "wrong" yet. Source of truth for budget: CRM. 35 closed deals is the count. The others explain where those deals came from. They don't redefine how many there were. Never add 50 + 40. That's 90 claims, not 90 conversions.
Question: where should next quarter's budget go? | Channel | Platform-reported | GA4 | CRM (closed) | Self-reported | Our read | | Paid search | 50 | ? | ? | ? | pull CRM first-touch | | Paid social | 40 | ? | ? | ? | expect over-claim | | Direct / branded search | n/a | ? | ? | ? | demand created elsewhere | Fill each column from its own source, de-duplicated against the 35, not summed. Then show first-touch and position-based side by side on CRM deals.
1. Add "How did you first hear about us?" (pick-list plus free text) to the demo form now. It's the strongest B2B signal you're missing. 2. Break MQL → SQL → closed-won down by first-touch channel. A channel that fills forms but never closes is a trap. 3. Before cutting or doubling either paid channel, run an on/off or geo holdout on the bigger one and budget on incremental CPA, not reported CPA. Confidence: directional until the holdout reads out.
More things to ask
Real requests the skill is tested against:
Should we use first-touch or last-touch attribution? We're a B2B SaaS with a sales cycle that runs about two months and multiple people involved in each deal.
Our biggest conversion is a sales call people book through SavvyCal, but that happens on savvycal.com so PostHog loses the whole journey. How do we connect a booking back to where the visitor originally came from? We control the marketing site.
Meta says our retargeting campaign has a 6x ROAS so we keep scaling it, but revenue isn't really going up. What's going on?
Half of our conversions show up as 'direct' in analytics and a big chunk of the rest is branded search. Does that mean direct traffic is our best channel?
Can you set up GA4 and our event tracking plan so we can start measuring conversions? We don't have any analytics installed yet.
FAQ
Why do Google Ads, Meta, GA4 and my CRM all show different conversion numbers?
Each one counts differently. Ad platforms credit themselves using view-through and their own attribution windows, so several can claim the same sale. GA4 loses cross-device and cookie-blocked users and pushes the unknown into direct. A CRM reflects what forms and reps captured, often a later-stage event like a closed deal. Pick one source of truth, usually the CRM, de-duplicate against it, and read whether the sources agree on direction.
Should I use first-touch or last-touch attribution?
For long sales cycles, use both side by side, because the gap between them is the insight. First-touch over-credits awareness and ignores what closed the deal. Last-touch over-credits branded search and direct, which defunds the top of the funnel. For B2B with clear created and closed moments, position-based (40% first, 40% last, 20% middle) on your own CRM data is defensible. Short, impulse e-commerce purchases can live with last-touch.
Why is so much of my traffic showing up as direct?
Direct is a junk drawer, not a channel. It holds typed URLs and bookmarks, but also stripped referrers, app-to-web handoffs, dark social like DMs, Slack and podcasts, and any touch your tracking dropped. A large direct share often means a working top of funnel that attribution hides. Rule out existing customers and lost referrers, segment branded from non-branded search, then check self-reported answers or run an incrementality test.
What is incrementality testing and when should I run one?
Incrementality tests withhold a channel from a control group, such as some regions, a random holdout audience, PSA or ghost-ad controls, and measure the difference. Lift is the exposed rate minus the control rate, and incremental CPA is spend divided by incremental conversions. If the confidence interval includes zero, you can't claim a win. Spend these tests on high-stakes questions like retargeting, branded search, or which of two big channels really drives growth.
When does a company need marketing mix modeling (MMM)?
When offline and brand spend is material and user-level tracking clearly can't see it. MMM regresses outcomes against spend per channel over time, so it captures TV, podcasts, PR and saturation curves without cookies. It needs 2–3 years of weekly data and real swings in spend, and it reacts in quarters, not days. For e-commerce, the skill suggests it becomes worthwhile around six figures a month of blended spend. Validate its coefficients with holdouts.
How do I attribute Calendly or SavvyCal bookings back to the marketing source?
Pass the visitor's anonymous analytics ID to the booking tool at click time. A capture-phase link decorator adds it to the outbound URL as metadata. When the tool's webhook returns it, merge identities (booking email as the known ID, the anonymous ID as the alias) and fire a conversion event. Only ever pass the anonymous ID. Reject anything that looks like an email and send nothing when identity is ambiguous, or profiles get corrupted.
Related skills
Use it in your agent
Marketing Attribution is a plain SKILL.md folder, so the same skill loads in any agent that reads the skills format. Setup guides:
Build a tool for this workflow?
Marketers run this skill mid-task — the exact moment your tool matters. Partner tiers put your tool inside the skill their agent runs.
See partner tiers