Last Click Attribution: What It Really Credits, and When It Misleads You

Last click attribution hands the entire order to the final click before checkout. Everything earlier in the journey gets nothing. That is the whole rule, and it is why the model is both the easiest one to explain and the easiest one to misread.

The part that actually costs money is quieter. Most brands do not run one last click model, they run four. Shopify, Google Analytics 4, Google Ads and Meta each apply their own version, over their own window, to the events each one can see. They disagree in the same week on the same orders, and none of them is broken. This article covers what last click attribution credits, why your four dashboards return four different answers, when the model is genuinely the right one to use, and how to move past it without losing the reporting your team runs on every morning.

Illustrative comparison of how four common reporting surfaces apply a last click rule. Every row is a defensible definition. None of them is measuring the same thing as the row above it.

What is last click attribution?

Last click attribution is a single-touch model that assigns 100% of a conversion's value to the last click a shopper made before buying. If someone sees a paid social ad on Monday, reads a review on Wednesday, clicks a branded search ad on Friday and orders, last click attribution pays branded search the whole order and pays the other two nothing.

The model is a rule, not a measurement. It does not estimate how much each touchpoint contributed. It picks a position in the path and awards everything to whatever is standing there. That is worth holding onto, because most arguments about attribution accuracy are really arguments about whether a positional rule should be treated as a finding.

The one rule the model follows

Two settings decide what last click reports, and they do different jobs. The lookback window decides which touchpoints are eligible to be considered at all. The model decides how credit divides between the eligible ones. Last click makes the second decision trivial, since it always picks the final entry, which means the window is doing almost all of the work.

Shorten the window and earlier touchpoints fall out of the path entirely, so more orders resolve to direct or to whatever sits closest to checkout. Lengthen it and the path gets longer without changing who gets paid, because the last entry is still the last entry. Brands that shorten a window and watch a channel collapse often conclude the channel stopped working. The channel did not change. Its eligibility did.

Last click and last touch are not the same word

Last click counts clicks. Last touch counts clicks and, on some platforms, impressions. That distinction is the reason Meta's numbers run higher than everyone else's. Meta attributes conversions to ads a shopper saw as well as ads they clicked, within windows you set in Ads Manager. A view-through conversion is invisible to a click-based model, so the same order shows up as Meta's win and as direct traffic in Shopify at the same time.

There is a third variant that catches people out. Last non-direct click, which is what Shopify and GA4 use, throws away the direct session and credits the last identifiable source before it. It exists because "direct" is usually a measurement failure rather than a channel, and it means a returning customer who types your URL still gets counted against whatever brought them in last.

Why last click became the default, and mostly still is

Last click is not the default because anyone argued it was accurate. It is the default because it is cheap to compute, impossible to dispute, and it never double counts. One order, one owner, and the numbers add up to the revenue you actually banked. Every other model produces fractional credit that no finance team enjoys reconciling.

It also survived a decade of tracking loss better than the alternatives. Multi-touch models need a complete path to divide, and the path is exactly what cookie restrictions, consent banners and cross-device journeys broke. A model that only needs the final step degrades more gracefully than a model that needs all of them.

The market has not moved on either. Google Ads still ships last click as a selectable model alongside data-driven, and it retired first click, linear, time decay and position based rather than last click. Google Analytics 4 kept two of its three available attribution models as last click variants. Whatever the industry says about last click being dead, the platforms kept it and deleted the others.

Your four dashboards are running four different last clicks

This is the failure that wastes actual meetings. A brand pulls last week's numbers, finds Meta claiming 180 purchases, Google Ads claiming 90, Shopify showing 240 orders total, and GA4 showing something between the three. Somebody says the tracking is broken. The tracking is fine. Each tool is answering a different question and calling the answer the same thing.

The table above lays out why. Shopify sees storefront sessions and nothing else, so an ad a shopper saw but did not click never enters its path at all. GA4 sees browser events that survive consent and ad blockers, so it systematically undercounts, and it drops direct entirely by design. Google Ads only sees Google, and it counts a conversion back on the day of the click rather than the day of the order, so a Google Ads week and a Shopify week are not even the same seven days of revenue. Meta only sees Meta, and it counts views.

None of those systems is wrong on its own terms. They are wrong together. Adding platform-reported conversions across four tools produces a number larger than your order count, because each platform claims the orders it can see and no platform deduplicates against the others. That is not a bug anyone is going to fix for you, because no platform has an incentive to hand credit to a competitor.

The practical consequence: last click is only meaningful inside one system that sees every channel and every order. Applied to four partial views, it produces four partial answers that cannot be added, averaged or reconciled by argument.

What last click attribution systematically overpays

Last click does not fail randomly. It fails directionally, and the direction is always the same: credit flows to whatever sits closest to the checkout button.

Branded search takes the biggest unearned share. Somebody who already wants your product searches your name and clicks the ad above the organic result. Last click pays that ad the full order. The ad did not create the demand, it collected it, and it collected it from a shopper who would very likely have scrolled two centimetres further and arrived anyway. Search Engine Journal has been making this argument about last click and PPC for years, and it is the single most reliable way to manufacture an impressive ROAS.

Retargeting is the same trick with a different ad unit. Retargeting only reaches people who already visited, which means it is structurally guaranteed to be the last touch for a large share of buyers. A retargeting campaign that shows up last on 40% of orders will report a superb return no matter how little incremental purchasing it caused.

Email and SMS land last more often than they earn. Lifecycle messaging usually reaches people who are already customers or already deep in consideration. The abandoned cart flow that "recovers" a cart is frequently the receipt for a decision made an hour earlier.

The pattern is worth stating once, plainly: last click rewards the channel that was present at the close, and being present at the close is something you can buy cheaply without changing anyone's mind.

What it underpays, and what it cannot see at all

The mirror image is prospecting. Paid social, video, creators, podcasts and affiliate placements do their work early. They introduce the brand to somebody who had never heard of it, and then they hand off to search or email or direct, which take the credit. Under last click, an introduction is worth exactly zero.

Underpaid is recoverable. You can see a channel appearing early in paths and argue for it. The harder problem is the demand that never appears in any path at all.

• Word of mouth arrives as direct traffic or branded search, wearing whatever costume the last click gave it.
• Podcast and creator reads generate a search rather than a click, so the referral is invisible unless someone uses the code.
• Connected TV, direct mail and out of home produce no click by definition.
• Practitioner and professional recommendation, which drives a large share of demand in health, supplements and specialist categories, leaves no digital trace whatsoever.
• In-store and marketplace discovery that ends in a direct online order looks like free revenue.

For brands whose growth genuinely runs on those channels, last click is not merely imprecise. It reports that the thing driving the business does not exist, and then it recommends spending the budget on the channels that were standing near the finish line.

Five ways last click quietly moves your budget

None of these announce themselves. They show up as a reallocation that felt data-led at the time.

1. The prospecting cut that shrinks the whole funnel

Prospecting reports a weak return, so it gets cut. Four to eight weeks later, branded search volume falls, retargeting audiences thin out, and the channels that looked efficient stop performing. They were efficient because prospecting was feeding them.

2. The scaling ceiling nobody can explain

Budget concentrates into the channels with the best last click numbers. Those are the ones closest to purchase, and demand capture cannot exceed the demand that exists. Spend goes up, orders do not, and the reported return degrades without anyone having changed the creative.

3. The window change that looks like a channel collapse

Somebody shortens a lookback window to be more conservative. Longer consideration journeys stop resolving, more orders land in direct, and a channel appears to have died overnight. Nothing happened to the channel.

4. The seasonal misread

During peak, discovery compresses and more shoppers arrive already decided. Last click credits closing channels even more heavily than usual, which makes peak look like proof that prospecting is unnecessary. It is proof that prospecting already happened.

5. The agency scorecard that pays for the wrong thing

When a partner is measured on last click return, the rational move is to bid harder on branded terms and retarget more aggressively. Both improve the reported number. Neither necessarily sells anything extra.

When last click attribution is the right model

Most articles on this topic are written by companies selling multi-touch attribution, so they end here with a verdict. The honest answer is that last click is correct for a specific set of questions, and reaching for something more sophisticated on those questions makes the answer worse rather than better.

Last click is the right model when:

• The buying cycle is genuinely short. Impulse categories under about $40, one session, one decision. When the path really is one touch, a multi-touch model is fitting complexity to noise.
• You are asking a closing question. Which creative converts a warm shopper, which landing page finishes the job, which flow recovers a cart. Last click answers exactly that and answers it cleanly.
• You are diagnosing tracking rather than performance. A sudden spike in direct or unattributed orders under last click is one of the fastest ways to catch a broken tag, a missing app embed or a UTM convention somebody changed.
• You need a number that ties to revenue. Last click never double counts, so it reconciles against orders. Fractional models do not, which is why finance tends to prefer the blunt one.
• You have one meaningful acquisition channel. With a single channel there is nothing to split, and the model question is academic.

The failure is not using last click. The failure is using last click to decide what to spend on next. It is a good report on what closed and a poor guide to what to fund, and those two jobs get confused because the same dashboard serves both.

Last click compared with the other models

The models differ in one respect only: where they place the credit. Here is the same single order under each, and what each one is actually good for.

Illustrative splits on a three touchpoint path. Note what the table does not contain: a column for whether the channel caused the sale. No positional model has one, which is the limit of the entire category, and it is why a model comparison eventually stops being the interesting question.

If you want the longer version of this, we walk each model and its failure modes in our guide to attribution models compared side by side.

How to move past last click without breaking your reporting

The mistake here is switching models on a Monday and asking the team to trust numbers that moved 40% overnight for reasons nobody can explain. Sequence it.

Step 1: fix the inputs before you touch the model

Changing models on incomplete data changes which incomplete answer you get. Before anything else, find out how many orders currently resolve to nothing. If a meaningful share of revenue is unattributed or sitting in direct, that is a tracking problem, and no model fixes it.

The usual causes are unglamorous: a missing app embed, UTM conventions that differ between the agency and the in-house team, sales channels that carry no referrer at all, and journeys that break when a shopper moves from phone to laptop. Fix those first. The gap between models is almost always smaller than the gap between tracked and untracked.

Step 2: pick one system as the ledger

One system needs to see every channel and every order, and every conversation about performance needs to happen inside it. Platform dashboards remain useful for optimising inside their own walls, which is what they are good at, but they stop being the score. This single decision removes most of the four-dashboards argument, because the argument was never really about models.

Step 3: use last click as one lens, not the verdict

Once the numbers live in one place, run the same period under last click and under a model that pays introductions. The channels whose share moves most between the two are the ones you are most likely mispricing. That comparison is more useful than either model on its own, and it costs nothing but a toggle.

Step 4: prove the ones that matter

Attribution says who was present. It cannot say what would have happened otherwise, and no amount of model sophistication changes that, because every positional model is dividing up correlation. For the decisions that genuinely matter, whether to keep funding a channel at all, the answer comes from a holdout test: turn the channel off in some geographies, leave it on in matched ones, and measure the difference in total revenue. That is incrementality testing, and it is the only method that answers a causal question with evidence rather than a rule.

Run it on the two channels where last click and a first-touch view disagree most violently. Those are where the money is being decided badly, and they are usually branded search and prospecting social.

How Polar handles last click

Polar's position is that the model is a setting, not a truth, so you should be able to change it and see what moves.

The Polar Pixel is first-party and server-side, so it is not dependent on third-party cookies surviving in the browser. It stitches sessions into one journey across devices and over time, including late-arriving signal, so a shopper who researches on a phone in March and buys on a laptop in May resolves to one path rather than two strangers. Attributed revenue reconciles back to actual Shopify orders, which is what makes the number arguable in a finance meeting rather than just in a marketing one.

On top of that data, the model and the window are both yours to set. Last click is available, as are first click, linear, time decay, U-shaped and a data-driven model that assigns fractional credit from observed paths. Lookback windows are configurable rather than fixed at 7 or 30 days, and there is a paid-only mode when you want to see how credit distributes among paid channels alone. Because every report reads the same underlying journeys, switching models changes the lens rather than the dataset, so you can tell how much of a result is the channel and how much is the setting.

Two capabilities exist specifically for the demand last click cannot see. View-through and offline touchpoints, including connected TV and direct mail impressions, can enter the journey rather than being lost. And post-purchase survey answers can be treated as real touchpoints, which is how brands whose growth runs on word of mouth, podcasts or professional recommendation get those channels into the attribution report instead of watching them disappear into direct.

For the causal question, Causal Lift runs geographic holdout tests, so the channels that look expensive under last click can be judged on whether turning them off actually costs revenue. The broader picture, including how these pieces fit together, is in our overview of multi-touch attribution.

Last click attribution is a lens with a known bias, and a known bias is workable. Use it to understand what closes, do not use it to decide what to fund, and prove the expensive decisions with a test rather than a rule.

FAQ