How the credit actually gets split
A multi-touch model needs three things before it can assign credit to anything: a way to tell that two touches came from the same person, a timestamp on each touch, and a rule for how much weight each one gets. Tools solve the identity problem with cookies, click IDs, or a login event (none of which catches every visit). Then they apply one of a handful of standard weighting rules to whatever touches they actually managed to capture, and that second part gets glossed over far more than it should.
The whole exercise exists to correct two blunt defaults. Crediting 100 percent of a sale to the very first ad someone ever saw flatters top-of-funnel spend and makes prospecting campaigns look better than they are. Crediting 100 percent to the last click flatters retargeting and branded search, which tend to catch people who were already going to buy. Multi-touch attribution spreads the credit across everything in between, using something more nuanced than a straight first-touch or last-touch model.
The main multi-touch attribution models
Every multi-touch model answers the same question differently: given a customer who saw an ad, clicked a retargeting banner, and then converted from an email, how much of the sale does each one earn? Here is what the common ones actually do.
| Model | How it weights each touch | Where it makes sense |
|---|---|---|
| Linear attribution | Every touch gets an equal share of the credit | Simple reporting, or when you genuinely do not believe any one touch matters more |
| Time decay attribution | Touches closer to the conversion get more credit, usually on an exponential curve | Sales cycles with a clear urgency spike near the close, like ecommerce carts or limited offers |
| U-shaped (position-based) | First and last touch each get roughly 40 percent, the rest split the remainder | Teams who want to protect credit for both discovery and closing channels |
| W-shaped | First touch, lead-creation touch, and closing touch each get a fixed share | B2B pipelines with a clear lead-capture moment in the middle of the funnel |
| Data-driven (algorithmic) | Weights are learned from your own historical conversion paths | Accounts with enough conversion volume to train on, usually a few hundred a month minimum |
Linear attribution: the honest baseline
Linear attribution is the model most people mean when they say "just split it evenly." It is simple to explain to a client or a boss, and it does not require enough data volume to train anything. Its weakness is also its selling point: it treats a passive display impression the same as the touch that closed the deal, which is rarely true in practice. And for a lot of small accounts, that is honestly close enough. Most teams use linear as a sanity check against a fancier model, not as their reporting standard.
Time decay attribution in practice
Time decay attribution assumes the touches nearest to conversion had more to do with the decision than the ones weeks earlier. Reasonable enough, as assumptions go. A common implementation halves the credit for every seven days back from the conversion, so a click three days out might carry twice the weight of one that happened three weeks earlier. It suits products with a short, urgent consideration window better than it suits long B2B sales cycles, where the early research touches often matter as much as the final demo booking.
The instrumentation problem most explanations skip
Here is the part that gets left out of most write-ups: a multi-touch model can only weight the touches it actually recorded. A touch it never captured is not down-weighted, it is invisible, and the model runs its math on a partial graph while presenting the output as though it were complete. A friend forwarding a screenshot in a DM, a customer who saw an ad on a device that never got cookied, a Safari user whose cross-site cookie got wiped by Intelligent Tracking Prevention: none of these show up as a zero-weight touch, they show up as nothing at all. The "multi" in multi-touch attribution is bounded by what your tracking stack can see, and the sophistication of the weighting formula does not change that. Most vendors do not say where that boundary sits.
This is also why two tools running the same model on the same conversion can disagree by a wide margin. If one tool's attribution window is seven days and the other's is twenty-eight, they are not weighting the same set of touches to begin with. The second one simply has more raw material to distribute credit across.
What multi-touch attribution needs in order to actually run
Three things have to be in place before a multi-touch model produces anything trustworthy: consistent UTM parameters so touches can be grouped by campaign, a defined attribution window that matches your actual sales cycle, and a stitching method that survives a device switch or a cleared cookie. Miss any of the three and the model still runs, it just runs on a thinner dataset than its output implies.
For a lot of smaller operations, the practical version of multi-touch attribution is a two-node graph rather than a five-touch weighted model: first touch and last touch, full stop. Raydar's link tracking captures UTM data along with first-touch and last-touch cookies on every click that hits a bio link, which covers the two points most small teams actually act on without paying for a dedicated attribution platform built for a much bigger touch count.
Multi-touch attribution and the ad-platform view-through problem
There's one kind of touch worth pulling out on its own: a person who saw an ad, never clicked it, and converted anyway. Platforms log that as a view-through conversion, and fold it straight into their own multi-touch reporting using a matching method your other tools cannot see or verify. It is one of several structural reasons your ad platform's numbers and your analytics tool's numbers refuse to line up, even when both are configured correctly.
Multi-touch vs single-touch: which should you actually use
If you are running one or two channels with a short sales cycle, a heavier model buys you very little. Single-touch attribution is the right amount of machinery for that job. The complexity of a weighted multi-touch model only pays for itself once you are running enough simultaneous channels that a single-touch view would obviously mislead you, say crediting a branded search click for a sale that a two-week Instagram campaign actually built. Below that threshold, spend the effort on cleaner UTM discipline instead of a heavier model, though even good tagging will not rescue a channel nobody bothered to tag in the first place. Clean inputs beat a sophisticated formula running on messy data, every time.