Multi-Touch Attribution Explained in Plain English

A prospect sees your LinkedIn ad, ignores it, googles your category name three weeks later, reads a comparison blog post, signs up for a webinar a month after that, and finally converts after a sales rep follows up on a whitepaper download. Which of those five touches gets credit for the deal? Last-click attribution says the whitepaper download did all the work and everything before it was worthless. That’s obviously wrong, and multi-touch attribution exists specifically to fix it — but the term gets thrown around so loosely in vendor pitches that most marketers can describe what it’s supposed to do without actually understanding how any specific model works.

The problem multi-touch attribution is solving

Single-touch models — last-click and first-click — assign 100% of credit for a conversion to one interaction and ignore everything else in the journey. Last-click is the default in most analytics tools because it’s the easiest to measure: whatever channel the person was on right before they converted gets the credit. The problem is obvious once you think about a real B2B buying journey, which routinely spans 6-10+ touches across paid, organic, email, and direct channels over weeks or months. Crediting only the final touch systematically overvalues bottom-of-funnel channels (branded search, retargeting, direct) and undervalues the awareness and consideration channels that actually built the demand in the first place.

The practical consequence: teams that manage budget by last-click data alone tend to overinvest in retargeting and branded search — channels that look like they’re “working” because they capture credit for demand generated elsewhere — while cutting the top-of-funnel channels that were quietly doing the harder job of creating buyers in the first place. Multi-touch attribution tries to split credit across all the touches in a journey instead of dumping it all on one, so budget decisions reflect the whole path, not just its last step.

The five common models, explained without the jargon

Linear attribution splits credit evenly across every touchpoint in the journey. If there were five touches, each gets 20%. This is the simplest multi-touch model to implement and understand, and it’s a reasonable default when you don’t yet have enough data or confidence to weight touches differently. Its weakness is that it treats a passive ad impression the same as an active demo request, which rarely reflects reality — not all touches contribute equally.

Time-decay attribution gives more credit to touches closer in time to the conversion, on the theory that recent interactions were more influential in the final decision. This fixes linear’s biggest flaw for long sales cycles, where an ad seen five months before conversion probably mattered less to the final decision than a case study read last week. The tradeoff: it can still undervalue the very first touch that originally created awareness, especially in categories where the buyer researched slowly then converted quickly once they hit a trigger event.

U-shaped (position-based) attribution gives heavy weight — commonly 40% each — to the first and last touches, splitting the remaining 20% across everything in between. This directly addresses the two moments most marketers intuitively agree matter most: the touch that created initial awareness and the touch that closed the deal. It’s a popular middle-ground choice for teams who want more nuance than linear without building a full data-driven model.

W-shaped attribution extends the U-shape logic by adding a third heavily-weighted point: the moment a lead converts to a marketing-qualified or sales-qualified lead, in addition to first touch and closing touch. A common split is 30% first touch, 30% lead-creation touch, 30% closing touch, with the remaining 10% divided across everything else. This works well for B2B funnels with a clear, meaningful MQL-to-SQL handoff stage, because it explicitly credits whatever generated that qualification moment, not just the bookend touches.

Data-driven (algorithmic) attribution uses statistical modeling — typically some form of regression or a game-theory approach called Shapley value — to calculate each touchpoint’s actual incremental contribution to conversion, based on comparing journeys that converted against similar journeys that didn’t. This is the most accurate approach in principle because it derives weights from your actual data rather than assuming a fixed rule, but it requires substantial conversion volume (generally at least several hundred conversions per month) to produce statistically meaningful results, which rules it out for many mid-market and smaller B2B companies.

A worked example: the same journey, five different answers

Go back to the five-touch journey from the opening — LinkedIn ad, branded search, comparison blog post, webinar signup, whitepaper follow-up — and assume the deal was worth $40,000 in first-year contract value. Here’s how each model would divide credit:

• Last-click: whitepaper follow-up gets $40,000. Everything else gets $0.
• Linear: each of the five touches gets $8,000 (20%).
• Time-decay (using a common 1-week half-life): the whitepaper follow-up might get roughly $16,000, the webinar $11,000, the comparison post $7,000, branded search $4,000, and the LinkedIn ad $2,000 — the exact split depends on the decay rate you choose, but the shape is always more credit to recent touches, less to distant ones.
• U-shaped: LinkedIn ad (first touch) gets $16,000, whitepaper follow-up (last touch) gets $16,000, and the remaining $8,000 splits across branded search, the blog post, and the webinar signup — roughly $2,700 each.
• W-shaped: if the webinar signup was the MQL-creation moment, it jumps to a heavily weighted $12,000 alongside $12,000 each for the LinkedIn ad and the whitepaper follow-up, with $4,000 left over for branded search and the blog post combined.

Run your own top five deals through this exercise by hand once. It’s the fastest way to build intuition for how differently each model treats the exact same set of facts, and it makes the abstract percentages from a vendor’s settings page feel concrete instead of arbitrary.

Picking a model based on your actual sales cycle, not what sounds sophisticated

There’s a strong pull toward picking the most sophisticated-sounding model available, but the right choice depends entirely on your funnel’s shape and your data volume, and picking beyond what your data can support produces a number that looks precise but is actually noise.

For a short sales cycle with high conversion volume (think self-serve SaaS with a days-long buying journey and thousands of monthly conversions), data-driven attribution is genuinely achievable and worth the setup effort, because you have enough volume for the statistics to mean something. For a longer B2B sales cycle with a clear MQL/SQL handoff and moderate volume (dozens to low hundreds of conversions per month), W-shaped or U-shaped attribution gives a solid balance of nuance and stability without requiring more data than you have. For very low-volume, high-ACV sales (enterprise deals closing in the single digits or low tens per month), even time-decay or linear may be more honest choices than a data-driven model, because with that little volume any algorithmic model is really just overfitting noise and dressing it up with a fancier name.

A useful gut check: if changing your attribution model from one option to another shifts your channel rankings dramatically and unpredictably month to month, that’s usually a sign you don’t have enough conversion volume to support the model’s complexity, not a sign the model is capturing something deep and true.

Sequencing the switch: how to change models without breaking trust in the data

Swapping attribution models mid-quarter, without warning, is one of the fastest ways to torch a marketing team’s credibility with finance and sales leadership — the channel rankings shift, someone asks “wait, why did paid social’s numbers just double,” and the honest answer (“we changed the model”) sounds like you’re moving the goalposts. Sequence a model change deliberately instead of flipping a setting.

Start by running the new model in parallel with the old one for at least one full reporting cycle (a full quarter for long B2B sales cycles, 4-6 weeks for faster-cycle businesses) without changing any budget decisions based on it yet. This gives you a side-by-side view of how much the channel rankings actually move and gives stakeholders a chance to see the new model’s output before it’s tied to any decision. Document the specific reasons for the change — usually “our sales cycle has gotten longer” or “we now have an MQL stage worth crediting separately” — so the change reads as a deliberate response to how the business evolved, not an arbitrary tweak.

Once you do cut over, keep a clearly labeled historical view of the old model’s numbers for at least a year of trailing comparisons. Nothing erodes confidence in attribution faster than a dashboard where last year’s numbers silently changed because the underlying model changed retroactively.

What multi-touch attribution can’t tell you (and why that matters)

Even a perfectly implemented multi-touch model only measures touches it can see, and it can’t measure the counterfactual — what would have happened without that touch. This is the crucial distinction between attribution and incrementality, and it’s the source of a huge amount of misplaced confidence in attribution data. A channel can show up as contributing meaningfully to conversions in your attribution model purely because it happens to reach people who were already going to convert anyway (branded search is the classic example — someone searching your company name by name was very likely already sold), without that channel having caused any of those conversions to happen.

The practical fix isn’t to abandon multi-touch attribution, it’s to pair it with periodic incrementality testing — geo holdouts, matched-market tests, or simple channel pause-and-measure experiments — especially for channels where attribution and instinct seem to disagree. Attribution tells you where credit lands across a modeled journey; incrementality testing tells you what actually moved the needle. Teams that only ever look at attribution data, without occasionally validating it against a real experiment, tend to drift toward optimizing for whatever the model rewards rather than what’s actually growing the business.

Dark funnel and untracked touches: the model’s blind spot

Every attribution model, however sophisticated, only accounts for touches your tracking infrastructure actually captured. A prospect who saw a podcast ad, heard your company mentioned by a colleague, read a review on a third-party site, and only then searched your brand name directly has had several influential touches that no pixel or UTM parameter ever recorded. This “dark funnel” effect is a genuine, structural limitation, not a fixable bug — some portion of real influence on a buying decision will never show up cleanly in any attribution model, no matter how well-instrumented your tracking is.

The practical response is to treat direct and branded-search traffic with some humility rather than either ignoring it or crediting it fully to itself. A spike in direct traffic and branded search following a podcast sponsorship or a PR push is a reasonable (if imprecise) signal that the untracked activity is working, even though no attribution model will draw that line explicitly. Surveying new customers with a simple “how did you first hear about us” question, cross-referenced against what your attribution model shows for the same customers, is a low-tech but genuinely useful way to estimate how much dark-funnel influence is happening and adjust your confidence in the attribution numbers accordingly.

The identity-stitching failure mode nobody budgets for

A separate, more mechanical problem sits underneath all of this: attribution models can only stitch together touches that get matched to the same person. Cross-device journeys (research on a phone, convert on a work laptop), cookie loss from browser privacy changes, and B2B buying committees where three different people from the same account interact with your marketing independently all break the chain of touches the model is trying to assemble. When identity stitching fails, the system doesn’t fail loudly — it just quietly reassigns those broken-off touches to whichever channel happened to complete the visible part of the journey, usually last-click-adjacent channels again, for the same structural reason last-click overweights the bottom of the funnel in the first place.

The fix isn’t glamorous: keep your CRM’s account-level activity view as a sanity check against your attribution tool’s user-level view, especially for enterprise deals where a buying committee is the norm rather than the exception. If your attribution tool shows a deal converting from a single three-touch journey but your CRM shows five different stakeholders from that account engaging with content over four months, trust the CRM’s fuller picture over the attribution tool’s narrower one, and treat the attribution number as a floor on the number of real touches involved, not the whole truth.

Making attribution actually change budget decisions

None of this matters if the output sits in a dashboard nobody acts on. The useful discipline is a recurring (monthly or quarterly) review where you compare your multi-touch attribution’s channel-level credit against your current budget allocation, specifically looking for channels that are getting meaningfully more or less credit than budget, or vice versa. A channel receiving 25% of attributed credit but only 8% of budget is a strong candidate for a test increase; a channel receiving 30% of budget but 10% of credit deserves scrutiny, ideally backed by an incrementality test before you cut it, since attribution alone (for the dark-funnel and counterfactual reasons above) shouldn’t be the sole trigger for a budget cut on a channel you have other reasons to believe is working.

How to tell if the switch to multi-touch actually worked

Adopting a multi-touch model is only worth the setup effort if it changes what the team does, so measure the change itself, not just the elegance of the new dashboard. Three signals are worth tracking for the two quarters after a model change: whether budget actually moved toward previously under-credited top-of-funnel channels (if the numbers shifted but the budget didn’t, the exercise was decorative); whether the incrementality tests you ran to validate specific channel calls confirmed or contradicted the attribution model’s ranking (persistent contradiction is a sign the model’s weighting assumptions don’t fit your funnel); and whether blended CAC and pipeline velocity improved over the following two quarters, which is the actual business outcome all of this modeling exists to influence. A more sophisticated attribution model that never changes a budget line or a channel decision hasn’t earned its complexity, no matter how defensible its math is.