Ask five people on a marketing team which channel deserves credit for last quarter's biggest deal and you'll get five different answers, all defensible, all wrong in some way. The ad they clicked once. The webinar they half-watched. The cold email a rep sent two months later.
The case study a colleague forwarded internally that nobody on the marketing team even knows exists. Attribution promises to settle this argument with data. In a B2B sales cycle with six stakeholders and a nine-month timeline, it usually can't, not cleanly, and pretending otherwise is where most attribution programs quietly start lying to the people relying on them.
What marketing attribution actually is
Marketing attribution is the practice of assigning credit for a conversion, a signup, a pipeline deal, a closed contract, to the specific marketing touches that influenced it. That's the textbook definition, and it's accurate as far as it goes.
What it leaves out is the assumption baked into every attribution model: that a buying decision is a chain of discrete, trackable events with a clean causal line running through them. For a single-touch consumer purchase, that assumption mostly holds. Someone sees an ad, clicks it, buys a $40 product nine minutes later. There's your attribution, uncontroversial.
B2B doesn't work that way, and treating it as though it does is the single biggest reason attribution reporting gets quietly ignored by the people who most need it to be right. A real B2B buying journey involves multiple people at the target account doing independent research on different channels, at different times, some of it not trackable at all: a Slack message from a peer at another company, a conference conversation, a Google search done from a personal phone that never touches a tracked device.
Attribution software can only assign credit to what it can see. It has no mechanism for admitting what it can't, so it quietly divides 100% of the credit among the touches it did capture, understating how much of the real influence happened somewhere the tracking pixel never reached.
The attribution models, explained
Every attribution tool ships with a menu of models, and the names get thrown around in meetings as if everyone agrees on what they mean. They mostly don't overlap as neatly as the dropdown menu implies. Here's what each one actually does with the credit for a single converted deal.
First-touch attribution
100% of the credit goes to the very first tracked interaction, the ad click or organic search that started the trackable journey. It answers "what got this account into our world," which is a genuinely useful question for evaluating top-of-funnel channels, but it gives zero credit to everything that actually closed the deal months later.
Last-touch attribution
100% of the credit goes to the final interaction before conversion, usually a demo request or a form fill. It's still the default in most CRMs because it's the easiest to implement, and it systematically overvalues bottom-funnel activity like retargeting and branded search while erasing every touch that built the awareness those final clicks depended on.
Linear attribution
Credit is split evenly across every tracked touchpoint in the journey. It's the fairest-looking model on paper and the least useful in practice, because it treats a passive newsletter open and a 30-minute sales call the same way, which flattens exactly the distinction a marketing budget decision needs to make.
Time-decay attribution
Credit increases the closer a touch sits to the conversion date, on the theory that recent interactions mattered more. It's a reasonable default for sales cycles under a few weeks. For a nine-month enterprise cycle, it quietly reproduces last-touch bias with extra math, because everything from the first six months gets a rounding error's worth of credit.
U-shaped (position-based) attribution
40% of the credit goes to the first touch, 40% to the touch that converted the lead (usually a form fill), and the remaining 20% is split across everything in between. It's a genuine improvement for B2B because it explicitly credits both discovery and conversion, the two moments that are hardest to replace, while still acknowledging the middle happened.
W-shaped attribution
The same idea as U-shaped, extended to give a third meaningful credit chunk to the touch that created a sales-qualified opportunity, not just a marketing lead. This is the model that maps most honestly onto a real B2B funnel with a distinct lead-to-opportunity handoff, at the cost of being harder to configure correctly.
Algorithmic (data-driven) attribution
A model, usually built on Markov chains or Shapley value calculations, that assigns credit based on which touches statistically correlate with conversion across the full dataset, rather than a fixed rule. It's the most defensible answer to "what is attribution marketing supposed to solve," but it needs a large volume of conversions to train on.
Most B2B companies close too few deals per month for the model to find a statistically stable pattern, which means the "smart" model quietly reverts to something close to a guess dressed up in more confident language.
Choosing a model for a long B2B sales cycle
The honest answer to "which attribution model should we use" is that the model matters less than what decision it's being used to make, and most teams pick one model and try to make it answer every question, which is where the trouble starts.
Separate the demand question from the conversion question
"What's creating awareness in accounts that eventually buy" and "what's converting an already-aware account into a lead" are different questions and deserve different models. First-touch or a U-shaped model answers the first; last-touch or the conversion-side weighting of a W-shaped model answers the second. Trying to answer both with one model is how a single dashboard number ends up satisfying nobody.
Default to U-shaped or W-shaped unless there's a specific reason not to
For a sales cycle longer than a month with more than one stakeholder, first-touch and last-touch both distort the picture in opposite, predictable directions. Position-based models split the difference in a way that's defensible in a budget conversation, and they're supported natively in most marketing automation platforms without custom engineering.
Track influenced pipeline alongside attributed pipeline
Attribution assigns exclusive credit; influence tracking counts every touch that appeared anywhere in a deal's history without forcing it to add up to 100%. A channel can show weak attributed credit and strong influence presence across nearly every deal that closes, which is a completely different, and often more useful, signal about whether to keep funding it.
Build a dark-funnel line item into the report, don't pretend it's zero
Ask closed-won buyers directly, in a one-question post-sale survey, how they first heard of the company. Compare that self-reported answer against what the attribution software claims. The gap between the two numbers is the size of the dark funnel your tracking can't see, and reporting that gap honestly builds more trust with a skeptical CFO than a suspiciously clean attribution chart does.
Re-evaluate the model when the sales motion changes
A model built around a self-serve, single-buyer motion stops fitting the moment the company moves upmarket into multi-stakeholder enterprise deals. The model isn't broken; it was built for a funnel that no longer exists. Revisit it whenever average deal size or buying-committee size shifts meaningfully, not on a fixed annual schedule.
Where attribution breaks down in practice
Three specific failure modes account for most of the distrust marketing attribution earns inside a company, and none of them are fixed by switching models.
The dark funnel. Peer recommendations, private Slack and Discord communities, word of mouth at industry events, dark social shares of a link inside a group chat, none of it carries a UTM parameter, and all of it demonstrably influences B2B purchase decisions. Attribution software can only report on what it can instrument, so it silently treats untracked influence as though it didn't happen, which systematically undervalues brand, community and word-of-mouth work relative to anything with a trackable click.
The buying committee. A deal with six stakeholders doesn't have one journey, it has six overlapping ones, often on different devices, sometimes without ever logging into the same account. Most attribution tools stitch activity to a single contact or a single account record, which either double-counts the same touch across stakeholders or silently drops touches that never got matched to a known contact at all.
Offline and human touches. A conference conversation, a customer reference call a rep arranged personally, an existing customer's casual comment to a peer at a different company, none of these show up in a marketing automation platform's touch log unless someone manually logs them, and almost no one manually logs them consistently. The result is a report that looks complete and precise while systematically missing the touches that are hardest to scale, which are often the ones that actually close enterprise deals.
What to track instead of chasing a perfect number
The mistake isn't using attribution, it's treating an inherently incomplete measurement as though it were a precise one and making budget decisions on the third decimal point. A more honest reporting setup pairs an attribution model with three supplementary signals: a self-reported attribution question on every closed deal ("how did you first hear about us"), influenced-pipeline tracking that doesn't force exclusivity, and a simple before/after test where a channel gets paused for a defined period to see whether pipeline volume actually moves.
That third one, the pause-and-watch test, catches what attribution models structurally can't: a channel with weak attributed credit that's quietly load-bearing for demand no dashboard ever traced back to it, because it was doing brand-building work that doesn't convert to trackable performance metrics on any single-touch view. The goal isn't a perfect number. It's a set of imperfect signals that agree often enough to make a budget decision defensible.
Marketing attribution isn't broken, it's just answering a narrower question than most reports imply. No model can see a private Slack recommendation, a hallway conversation at a conference, or six stakeholders researching independently across personal devices, so a clean single-number attribution report is describing the trackable minority of a much larger decision.
Pick a model that fits the actual sales motion (usually U-shaped or W-shaped for a multi-stakeholder B2B cycle), pair it with self-reported and influence-based signals, and report the size of what the model can't see instead of pretending it's zero.
FAQ
What is attribution in marketing?
Attribution in marketing is the practice of assigning credit for a conversion or closed deal to the specific marketing touchpoints that influenced it, so a team can judge which channels and campaigns are actually driving revenue rather than just activity. It's a measurement framework, not a single fixed number, and different models will assign very different credit to the same customer journey.
What is attribution marketing, exactly?
Attribution marketing usually refers to the broader discipline of building a measurement system, choosing a model, instrumenting the right touchpoints, reporting the results, around answering which marketing investments deserve credit for revenue. It's often used interchangeably with marketing attribution, though "attribution marketing" more often shows up when the emphasis is on the reporting and decision-making process rather than the underlying model math.
What is a marketing attribution model?
A marketing attribution model is the specific rule used to divide credit for a conversion across multiple touchpoints. Common models include first-touch, last-touch, linear, time-decay, U-shaped, W-shaped and algorithmic (data-driven) attribution, each of which assigns very different credit to the same journey. There's no universally correct model, only one that fits a given sales motion and the specific question being asked.
What's the difference between first-touch and last-touch attribution?
First-touch attribution gives all the credit to the very first tracked interaction, which is useful for evaluating what drives initial awareness. Last-touch gives all the credit to the final interaction before conversion, which is useful for evaluating what closes an already-warm lead. Neither reflects the full journey in a multi-touch B2B cycle, which is why position-based models like U-shaped or W-shaped attribution exist as a middle ground.
Do I need attribution software, or can this be done manually?
Attribution software is close to mandatory once a company runs more than a couple of marketing channels, purely because manually stitching together touchpoint data across email, ads, organic and events isn't realistic at any volume. The software isn't the hard part, though. Choosing a model that fits the sales motion and being honest about what the software can't see (the dark funnel, offline touches, multi-stakeholder journeys) matters far more than which vendor's dashboard displays the number.
How accurate can B2B attribution realistically be?
Meaningfully incomplete, by design, for any sales cycle involving more than one stakeholder or more than a few weeks of consideration time. Dark social shares, peer recommendations, and offline conversations aren't trackable by any current attribution tool, so even a well-configured model is reporting on the visible fraction of a larger, partly invisible decision process. Treating attribution as directionally useful rather than precisely accurate avoids most of the bad decisions that come from over-trusting a single dashboard number.
What should a marketing team actually do with attribution data?
Use it to spot clearly underperforming or clearly overperforming channels at the extremes, not to make fine-grained budget calls between two channels with similar attributed credit. Pair the model's output with self-reported attribution from closed-won buyers and a periodic pause test on individual channels; where all three signals agree, the budget decision is on solid ground, and where they disagree, that disagreement is more informative than the attribution number alone.
Read next
Trying to make a real budget call from attribution data?
I help teams build measurement systems that admit what they can't see instead of hiding it, so channel decisions hold up under a skeptical CFO's questions instead of falling apart under them. See how I work.
Book a Call