Attribution

When a company ran ads in one place, nobody needed the word. Put an ad in the newspaper, watch revenue rise that month, thank the newspaper. Then channels multiplied. One person reads a blog post from a search result, sees an Instagram ad four days later, searches the company name directly three days after that, and pays. One payment, three touchpoints in front of it. A touchpoint is any single place where a customer comes into contact with you, and a conversion is whatever action the team has decided counts as success — a free signup for some teams, a first payment for others. Somebody has to divide one 900-dollar conversion across three touchpoints, and no document said how. The blog post, the Instagram ad and the branded search all happened, all of them plausibly mattered, and the payment arrives as a single indivisible event.

Since nothing said how, everyone counted their own way. The same scene repeated at Pagedeck every quarter. The person running the blog and the newsletter said the reader found them through search and eventually paid, so the credit was theirs. The person running paid ads said the last click was on their banner. Add the two spreadsheets together and the total came to 1.8 times the revenue the company actually booked. Nobody had entered a wrong number. Both sides were counting the same conversion once each. A table like that can't split next month's budget: if both are right the total is 180 percent, and if only one is right there's nothing behind that call. Meetings usually ended with whoever argued loudest, or whoever had won the previous quarter, and the other side returned the favor three months later.

Attribution moves the argument into a rule. Rather than uncovering which touchpoint truly caused the sale, the team writes down how credit will be divided and then compares every channel through that one lens. Once the rule exists, the shares always sum to 100 percent and channels get measured with the same ruler. The rule does not have to be right to be useful; it has to be fixed, because a fixed rule makes one month comparable to the next. It's an agreement about the ruler, not a search for the truth, and that single fact explains most of how the concept behaves. Teams that pick an imperfect rule and hold it for six months end up with better judgment than teams that spend those months debating which rule is correct.

In the wider picture it sits in the middle of marketing measurement. Tracking records who arrived from where and what they did, attribution divides credit over that record, and analysis reads the divided numbers by time and cohort. The order never reverses. A newsletter sent without UTM parameters — the tags appended to a link so analytics can tell one source from another — collects no credit under any rule, because those visits land in an undifferentiated direct-traffic bucket. The ground is also shifting underneath. With third-party cookies blocked and mobile advertising identifiers hidden behind a permission prompt, stitching one person's three touchpoints into a single chain no longer works the way it once did. A third-party cookie is one set by a domain other than the site being visited, which is what let an ad network recognize the same browser across unrelated sites. With those gone, a growing share of conversions arrive with an unknown path, and the rule gets applied to a record that has holes in it.

The terms under attribution fall into two groups: models, which decide the shape of the split, and settings, which decide how far back the split reaches. A model picks the method and a setting picks the pool, so a standard that names only one of the two is half a standard.

Models occupy four slots. The first three differ in how they distribute credit; the fourth changes what counts as a touchpoint at all.

1. First-touch attribution — 100 percent to the earliest touchpoint. In the example above, the blog post takes the whole 900 dollars. It answers the question of what brings strangers in. Everything that removed hesitation right before payment scores zero, so a pricing comparison post that settled the decision never appears in the table.
2. Last-touch attribution — 100 percent to the final touchpoint. It's the most common rule and the default in most ad platforms because the math is trivial. The trouble is that the final touchpoint is usually a branded search. It's the doorstep of someone who already decided to buy, so this model inflates branded search ads. More than half of the people who searched for Pagedeck by name had already heard about the product from a colleague.
3. Multi-touch attribution — credit gets split across several touchpoints, often 30, 30 and 40 across three, sometimes weighted toward whatever sits closest to the conversion. Every step scores something, which is closer to how buying actually works, but no evidence sets the weights, so the team sets them. Change 30, 30, 40 to 20, 30, 50 and the channel ranking shifts a slot on identical data. Some teams fit the weights statistically instead of choosing them, which removes the arbitrariness of the numbers but needs far more conversions per month than a young product produces.
4. View-through conversion — counts people who never clicked, saw the ad, and converted later. Someone who scrolled past a banner and typed the address in two days afterward shows up here. It exists to make brand advertising visible, but a generous definition of "saw" pulls in people who merely scrolled past. Leave it undecided whether a banner that appeared for one second at the bottom of the screen counts as an impression, and the platform's number and yours end up more than twice apart.

Settings occupy one slot. The attribution window is the deadline: how many days after a touchpoint a conversion still counts. At seven days, a person who saw the ad and paid ten days later is treated as unrelated to it; at thirty days the same person becomes credit. Leave the model alone, move this one value from seven to thirty, and the channel ranking changes. Products with long deliberation need long windows — enterprise buyers can spend two months on security review and procurement, which a seven-day window catches almost nothing of. Stretch it to ninety days, though, and a touchpoint from a quarter ago keeps collecting credit, so a campaign that ended long ago keeps taking this month's results. Most teams end up somewhere between fourteen and thirty days for click-driven channels and shorter than that for view-through, on the grounds that a glance decays faster than a click.

Once both choices are made, the remaining work is writing them down. Changing the rule flips the entire channel ranking of the same month's report, so without a record of what changed and when, this quarter can't be compared to the last one. In practice teams name one standard for budget decisions and keep the other models as second angles on the same month. Put first-touch and last-touch side by side and you can see which channels bring people in and which ones close them, separated on a single screen. A channel that ranks high on both is usually doing the whole job; one that ranks high only on last touch is standing near the door.

"This table is last-touch. On first-touch the newsletter jumps to number one."

Check which rule produced a number before reading it, even within the same month.

"Conversions from search ads went up 40 percent when we moved the window from 7 days to 30. Nothing actually went up."

The increase came from a settings change, not from performance.

"Our rule gives the ad credit here, but this person looks like someone who would have paid anyway."

The touchpoint that collects credit and the touchpoint that produced the outcome can be different.

The loudest this got at Pagedeck was the quarter the team considered doubling paid ad spend. In the table Ravi brought, ads accounted for 62 percent of new paying teams. Mia pulled the same period on first-touch and the share of ads dropped to 24 percent, while the path through search into the help docs rose to the top. Both numbers were right. Ads were often sitting in the last clicked position, and documentation was what brought people in first. The two tables looked like they were fighting because they were answering different questions, not because either was wrong.

Grace put both on one screen and split the decision. Budget allocation would key off multi-touch and nothing else, and quarterly targets would be set on that number. Separately, the ads running on branded search would go dark for a month to see how many signups actually disappeared. The rule didn't settle anything by itself, but it made visible what the team was arguing about. Nobody left the room believing the ranking was the truth, which was the part that had been missing.

The most common error is reading the channel with the largest share as the cause. Tables label the column "contribution," so the number sounds causal. It is a distribution rule applied to a record, which means the same data for the same month produces a different ranking once the rule changes. Move budget without noticing that difference and money piles onto activities that merely stand in the last position, like branded search ads, while the activities that brought people in shrink for scoring low. Two quarters later there's more money aimed at the final touchpoint and fewer people reaching it — the ad share on the table climbs from 62 to 71 percent while total conversions hold flat or fall. The table looks healthier precisely as the business gets thinner, because the share is a slice of a shrinking pie.

The other frequent mistake is believing a better model makes the number accurate. Multi-touch looks more sophisticated than first or last touch, so it seems like the more correct answer. But nothing justifies giving three touchpoints 30, 30 and 40. The rule got more elaborate; it's still a split somebody chose. Accuracy comes from turning ads off and measuring the difference, not from swapping models. Miss that and a team rotates through models each quarter, collects a different ranking every time, and stops trusting any of them — at which point the loudest voice wins again, which is exactly where the rule was supposed to help.