Picture the report that lands on a marketing director's desk every Monday. Branded search is the hero again. Retargeting is close behind. The prospecting campaigns, the YouTube spend, the early-funnel social, all of it sits near the bottom looking like a rounding error. The obvious move suggests itself: shift money toward the channels that win, away from the ones that lose. Do that for four quarters and you have quietly defunded the part of marketing that creates demand while pouring budget into the part that only collects it.
That report is not lying about what it measured. It is lying about what it means. Last-click attribution, the model that hands 100 percent of the credit for a conversion to the final touchpoint a customer interacted with, is the most widely understood measurement method in marketing and one of the most misleading. It is not misleading because it is complicated. It is misleading because it is simple, and the simplicity is the trap.
Origin: the model that won by asking nothing of you
Last-click did not win an argument. It won by default, because it was the only thing the early web could measure without effort. A cookie recorded the last click before a conversion, the ad platform read that cookie, and credit was assigned. No statistical model, no assumptions, no analyst. The answer was deterministic: the same data produced the same result every time, and you could trace any conversion back to a specific click and feel you had proof.
Three properties made it the industry standard. It is simple, so anyone can explain it in a sentence. It is deterministic, so it never needs defending as an estimate. And it is native to every ad platform, baked into Google Ads and the social networks as the original way conversions were counted. Ruler Analytics describes it as the most basic form of attribution, which is exactly why it spread. A method that needs no modeling and no setup will always out-compete one that does.
The platforms that promoted it have since walked away from it. Google made data-driven attribution the default in Google Ads in September 2021, and in 2023 it removed First Click, Linear, Time Decay and Position-Based models from Ads and GA4 entirely, leaving advertisers with data-driven attribution or last-click and nothing in between. The vendor that built its business on the last click no longer recommends it. The model survived the endorsement that created it.
Present: what the final touch quietly erases
A real purchase has a history. Someone sees a brand on TikTok, reads a review three weeks later, gets a retargeting ad, ignores it, then one evening types the brand name into Google and buys. Last-click looks at that chain and records one thing: branded search. Every touch before the final one is erased, not discounted, erased. The model does not say the TikTok video mattered less. It says it did not happen.
That erasure is not random. It runs in a consistent direction, and the direction is what makes it dangerous. Last-click systematically overcredits whatever sits closest to the purchase and undercredits everything that sits further back. The reliable winners are branded search, retargeting, email, and direct traffic. The reliable losers are prospecting, video, display, influencer content, and any awareness channel whose job is to plant a memory that pays off weeks later. Rockerbox puts it plainly: brands that rely on last-click alone systematically underinvest in the upper-funnel work that makes later conversions possible.
Branded search is the clearest case, and there is hard evidence rather than opinion behind it. Researchers ran a field experiment with eBay, published in the journal Econometrica, in which eBay switched off its paid search ads. When the brand-keyword ads stopped, 99.5 percent of the traffic that those ads would have delivered showed up anyway through organic results. The customers were searching for eBay. They were going to arrive whether or not an ad sat above the organic link. For non-brand keywords the experiment found a statistically insignificant lift and near-zero effect on frequent buyers. Last-click had been crediting that branded spend with conversions it did not cause. Procter and Gamble reportedly found the same when it cut brand search spending and watched business outcomes hold steady.
Retargeting follows the identical pattern. A retargeting ad shows almost exclusively to people who already visited your site, which means it shows to people already partway to buying. When they convert, last-click hands the retargeting campaign the credit. Withhold those ads from a control group and a large share of that audience converts regardless, returning directly or searching the brand. The reported return looks excellent because the model cannot see that the demand already existed. The ad did not create the customer. It stood near the finish line and took the photo.
So the model is wrong in a specific, structural way. It mistakes the channels that harvest demand for the channels that create it. And because it cannot tell the difference, neither can anyone reading its reports.
The decisions it actually breaks
A flawed measurement is harmless until someone spends money on it. Last-click is not harmless, because budget allocation is exactly what marketers reach for it to decide.
Start with the budget itself. Read a last-click report as a scoreboard and the rational move is to cut what loses and feed what wins. The losers are the awareness channels. The winners are the demand-harvesting channels. Run that logic across a few planning cycles and you have starved the top of the funnel that produces the searches and the site visits, then spent the freed-up money on the channels whose entire job is to convert demand that already exists. Fewer people enter the funnel. The harvesting channels keep posting strong numbers because they are still harvesting, just from a shrinking pool. The decline is slow, it looks like a market problem, and the measurement that caused it stays in place because it never reported anything wrong.
The clearest documented version of this is a geo experiment Uber ran, described by the agency 9AM, where a controlled test showed certain paid social spend was close to non-incremental and the company cut roughly 35 million dollars that last-click style measurement had been quietly defending. That is the gap between credited and incremental, converted into a number.
Then there is the overspending the model invites directly. Branded search and retargeting look like the most efficient line items on the plan, so they get protected and expanded. Both are largely harvesting channels. Both can show a strong return while adding little the business would not have captured anyway. Last-click makes non-incremental spend look like the safest spend you have, which is the exact inversion of the truth.
The third cost is organizational and the least discussed. Attribution decides which team looks good. The team running branded search and retargeting collects conversions it did not create. The team running awareness gets a report that says its work does not convert. Budgets, headcount, and credibility follow those numbers. People are smart, so they optimize for the model: everyone crowds toward the bottom of the funnel where last-click hands out credit, and the unglamorous, hard-to-measure demand creation gets quietly abandoned because the scoreboard never rewards it. This is part of why multi-touch attribution emerged in the first place, an attempt to spread credit across the journey rather than dump it all on the final step.
Why it refuses to die
If last-click is this flawed, and the platforms have moved on, it should be gone. It is not. Plenty of teams still run on it, and the reasons are worth being honest about, because they are not stupidity.
It is organizationally simple, and organizational simplicity beats statistical accuracy in almost every meeting. Last-click gives one number, with one owner, traceable to one click. A defensible answer in five minutes outcompetes a better answer that needs a modeling caveat and a confidence interval.
Finance prefers it, and finance often holds the budget. A deterministic count looks like accounting. A modeled estimate looks like marketing asking to be trusted. Faced with a probabilistic credit split nobody can reduce to a clean rule, the finance instinct is to fall back on the number that behaves like a ledger entry. Across the industry, organizational understanding is repeatedly cited as a bigger blocker to better attribution than the technology itself.
It is the path of least resistance. It is already on in the ad platforms and the analytics tool. Better measurement is a project: incrementality tests to design, a marketing mix model to build or buy, an analyst to run it. Last-click is the default that is already running and costs nothing to keep.
And it is consistent. This is the genuinely fair point. Last-click is wrong, but it is wrong the same way every week. A team that has watched its last-click numbers long enough develops a rough feel for the bias and can read direction through it. Even so, the marketers who lean on it are not happy: in one survey only about one in five US marketers said they were confident last-click reflected a platform's real long-term impact. They keep it not because they trust it but because the alternatives ask for more than they have.
Future and impact: a directional tool, not a budget tool
The honest position is not that last-click is worthless. It is that last-click is fine for one job and dangerous for another, and the whole problem is that people use it for the dangerous one.
The job it can do is directional and operational, the fast tactical signal. Which landing page closed more sessions this week, which keyword converted, which checkout variant performed better. For short paths and single-session decisions, last-click is quick, clear, and good enough. Rockerbox notes it is genuinely useful for lower-funnel tactics where the final touch and the conversion really are tightly linked. As a daily operating dashboard it earns its place.
The job it cannot do is decide where the budget goes. Allocation is a causal question. It asks what would happen if you moved a million dollars from one channel to another, and last-click cannot answer causal questions because it only ever describes the final click, never tests a counterfactual. That job belongs to methods built for it. Incrementality testing answers it directly: hold a channel out from a control group and measure what changes, which is exactly how the eBay and Uber findings surfaced. Marketing mix modeling answers it at the portfolio level, using aggregate spend and outcome data with no user-level tracking at all, which is also why it works in a world of decaying cookies and signal loss. The market has already turned that way. In a July 2025 eMarketer and TransUnion survey, 27.6 percent of US marketers named MMM their most reliable methodology, the single top answer, and 46.9 percent planned to invest more in it.
The mature setup is not one model. It is last-click or platform attribution for tactical signal, incrementality tests for causal proof, and marketing mix modeling for strategy, reconciled into one estimate of incremental impact. AI agents are starting to make that triangulation faster, drafting tests and flagging when a channel's credited numbers and its incremental numbers drift apart. But the agent does not rescue last-click. It just makes it cheaper to see what last-click was always hiding.
Keep last-click on the dashboard. Read it on Monday morning. Just never let it sign the budget, because the report that makes branded search the hero every week is not describing your best channel. It is describing the last thing your already-decided customer happened to touch.
Council summary
This post argues that last-click attribution is not wrong by accident but wrong in a fixed direction: it overcredits the channels that harvest existing demand and erases the channels that create it, and that bias quietly corrupts budget allocation, invites overspending on branded search and retargeting, and warps which teams get credit. The critique is anchored in causal evidence rather than opinion, the eBay paid search field experiment in Econometrica and Uber's geo test that exposed roughly 35 million dollars of non-incremental social spend, and it is honest enough to grant last-click its one real use as a fast tactical signal. The reader's takeaway is a clean division of labour: keep last-click on the daily dashboard, but never let it set the budget, because allocation is a causal question that only incrementality testing and marketing mix modeling can answer. A smart marketing decision-maker should leave knowing the report that crowns branded search every Monday is describing the last thing an already-decided customer touched, not the channel that earned the sale.
Comments