B2B marketing attribution assigns credit for closed deals to the touchpoints that influenced them. The honest answer to “is mine lying?”: every model is wrong, because committees, dark social, and word of mouth stay invisible to tracking. The question is whether yours is still useful.
Every few months, somewhere in your company, a version of this meeting happens: marketing presents attribution numbers showing which channels drove pipeline, sales mentions that the biggest deal of the quarter “actually came from a referral”, finance asks why the attributed revenue doesn’t sum to actual revenue, and everyone leaves slightly less convinced than they arrived.
Here’s the uncomfortable truth that makes the meeting easier: all attribution models are wrong. B2B buying involves committees, dark-social conversations, podcast listens, and word of mouth that no tracking script will ever see. The question is never “is our attribution accurate?” — it can’t be. The question is “is it useful, and do we know where it’s lying?”
The four lies to check your model for
Lie one: the visible-touch lie. Attribution can only credit what it can see — clicks, forms, tracked visits. The influences it can’t see (a colleague’s recommendation, a conference chat, three months of podcast listening) get credited to whatever visible touch happened to come last, usually a branded Google search. This is why paid search perpetually looks like a hero: it harvests demand that something else created. Check: how much of your attributed pipeline lands on “direct” and branded search? That share is roughly the size of what your model can’t see.
Lie two: the single-person lie. Most models attribute at the contact level, but B2B deals are bought by committees. The person who filled the form is often not the person who drove the decision — and the champion who never touched your website is invisible. Check: on your last ten closed-won deals, how many contacts were involved, and how many had any tracked touches? The gap is the lie.
Lie three: the hygiene lie. Attribution inherits every upstream data problem. Duplicate contacts split journeys in half; broken sync mappings orphan touchpoints; missing source data (check six of the hygiene health check) means a chunk of pipeline is attributed to a shrug. If your source coverage is 80%, your attribution model is a survey with a 20% non-response rate that nobody’s correcting for.
Lie four: the model-choice lie. First-touch, last-touch, U-shaped, W-shaped — each answers a different question, and switching between them reshuffles credit dramatically while changing nothing real. Teams sometimes discover this the hard way when a model change makes a channel look suddenly brilliant or suddenly dead. If a decision flips depending on which model you view it through, the honest conclusion is that attribution can’t make that decision.
The audit: three tests in an afternoon
1. The reconciliation test. Take last quarter’s closed-won revenue and compare it to attributed revenue by any model. Document the gap and the “unattributable” share. This is your model’s confidence interval — and it should be printed on every attribution report, the way polls print margins of error.
1. The deal-review test. Pick five recent wins. Interview the reps: how did this deal really start? Compare against what attribution says. The pattern of disagreement tells you which lie dominates in your business.
1. The decision test. List the decisions actually made on attribution data in the last two quarters. For each: would a different (equally defensible) model have flipped it? Decisions that survive all models are safe; the rest were coin flips wearing a dashboard.
The reckoning usually comes from putting the attribution dashboard next to a handful of actual deal reviews. The model confidently credits the last touch — a branded search, a demo request — while the deal review shows the real origin was a webinar six months earlier, or a champion who’d been quietly reading for a year. Neither the model nor the reps are lying; the model only sees what’s instrumented, and most of the influence wasn’t. The decision that changes is usually a budget one: a channel that looked like a top performer turns out to be harvesting demand it didn’t create, while the spend that actually generated pipeline was being starved.
Using attribution like an adult
The mature posture isn’t abandoning attribution — it’s demoting it from accounting to evidence:
- Use it directionally, at channel level, over quarters. “Webinars consistently show up early in journeys that close” is a supportable claim. “Webinar X generated £312,400 in pipeline” is theatre with decimal places.
- Triangulate. Attribution + self-reported “how did you hear about us?” (add it to your forms — the answers will humble your model) + rep deal reviews. Where all three agree, act with confidence. Where they diverge, investigate before spending.
- Fix hygiene before models. Upgrading to a fancier attribution tool on top of dirty source data buys you more precise lies. Data first, model second — the same bottom-up rule as rebuilding reporting trust generally.
Attribution’s real job in B2B isn’t to grade channels to two decimal places. It’s to make resource allocation slightly less blind, quarter after quarter. Held to that standard, even a flawed model earns its keep — as long as everyone in the room knows where it lies.
—
Attribution integrity is lens four of the Pipeline Leak Audit — including the reconciliation test against your actual revenue. Book a free health check if you want a read on how big your model’s blind spot is.