Most marketing attribution lies. Not deliberately — through omission. The standard last-touch model says "the email closed the sale." The first-touch model says "the ad opened it." The truth lives in the 17 touches in between.
Over 286 client deployments, we've built and refined an attribution model that tracks every dollar to every touchpoint that influenced it. Here's the actual data structure, the dashboards we ship, and the lessons learned.
What standard attribution gets wrong
The three dominant attribution models — first-touch, last-touch, linear — all share the same fatal flaw: they assume the customer journey is a sequence of independent touches, each with measurable contribution.
Real customer journeys aren't sequential. They're networks. A prospect sees an ad, ignores it, then sees a friend mention you on LinkedIn, then Googles your name, then reads three blog posts over a week, then signs up for a newsletter, then opens 4 of 8 emails, then visits the pricing page twice, then books a call.
Last-touch attribution credits the call-booking page. That page didn't do the work. The blog posts and the email sequence did. The ad opened the loop. The friend's LinkedIn mention closed the credibility gap. The pricing page visits resolved the cost concern.
Crediting just the last touch loses 95% of the actual mechanics.
The model we use
Three components:
- Event tracking on every touchpoint — ad impressions, page views, email opens/clicks, video watches, support interactions, social mentions (where measurable), referrals.
- Behavioral signal weighting — not all touches are equal. A 30-second page view weights different than a 5-minute one. A click-through from a search query weights different than a click-through from a remarketing ad.
- Conversion-path reconstruction — when a deal closes, we reconstruct the full journey: every touchpoint, every signal, every time-gap. Then we apply a weighted attribution across all touches.
The weighting isn't arbitrary. It's based on historical data — across thousands of closed deals, which touches actually preceded conversion? Which gap-patterns predicted decay? The model is empirically tuned.
The data structure
For one client, the attribution database typically has these tables:
- Touchpoints — every recorded interaction with timestamp, channel, content, behavioral metadata
- Conversions — every revenue event with deal-size, timestamp, customer-id
- Journeys — the reconstructed sequence of touchpoints leading to each conversion, with weights applied per touch
- Channel-performance rollups — periodic snapshots of channel-level attributed revenue, weighted by the journey model
The schema isn't proprietary. The weighting model is. The data engineering to maintain it across a live funnel — that's where most agencies fall down.
The dashboards we ship
For Growth-tier clients, the attribution dashboard typically includes:
- Per-channel attributed revenue (weighted, not last-touch) — what did each channel actually earn?
- Touch-count distribution — how many touches did the average closed customer have before conversion? The shape of this distribution is more informative than the average.
- Channel-pair interaction — which channel combinations close deals together? (Ad + Email is one of the strongest combos for B2B; Ad alone almost never closes.)
- Time-to-close per source — which channels produce fast-converting customers vs. slow-converting? (Important for ROI calculations.)
- Funnel-stage attribution — at which stage of the funnel does each channel contribute most? (Some channels open journeys; others close them; few do both.)
Lessons from running this across 286 clients
Three patterns that show up reliably:
- Email is undervalued by last-touch attribution by 3-5×. Email rarely gets the last click but it's almost always in the middle of the journey, doing the work of resolving objections and maintaining engagement.
- Organic search is undervalued by 2-3×. When prospects Google you, last-touch attribution doesn't capture that they Googled you because they saw an ad or heard about you from a friend a month ago.
- Remarketing is overvalued by last-touch attribution by 2-4×. A remarketing ad that delivers the click that becomes a conversion didn't actually do the work — it just caught the prospect in the moment of decision that was set up by every prior touch.
The biggest mistake clients make
Optimizing channels based on last-touch attribution. We see this constantly: a CFO cuts the email budget because "email isn't producing revenue" (according to last-touch), and three months later the ad performance collapses because the email sequence was what was keeping prospects warm long enough to convert.
The right move is to optimize the journey, not the channel. Some channels are openers. Some are middle-movers. Some are closers. Cutting an opener kills the closers down the funnel. The attribution model needs to make that visible before the CFO makes the wrong call.
What to ask any vendor selling attribution
Two questions:
- "Show me a real customer's full journey — every touch from first impression to closed deal."
- "How does your model weight a 30-second page view vs. a 5-minute one? What's the actual coefficient?"
If the answer is "we use Google Analytics' multi-touch model" — they're not doing real attribution. They're presenting Google's defaults as their work.
Real attribution is data engineering plus empirical tuning plus willingness to publish dashboards that make uncomfortable channels look bad. Most agencies don't do it because the dashboards would reveal which of their efforts aren't working.
The clients who insist on it tend to be the ones running the biggest funnels — they have enough revenue to optimize against, and enough sophistication to know that last-touch is a lie.