Why Click-Based Metrics Overstate Amazon Display Ad Performance
Click-based metrics overstate Amazon Display ad performance because they measure the sales an ad was credited with, not the sales it caused. ROAS and ACoS rest on attributed sales, and click-through rate only counts clicks. A report attributes a sale to a display ad when the purchase follows the ad inside the attribution window, whether or not the ad changed anything. Many of those credited buyers were already on their way to checkout. Retargeted shoppers had looked at the product before. Repeat buyers come back on their own schedule. Shoppers who searched the brand name already knew what they wanted. When one of them clicks a display ad and then buys, the full sale is booked to the ad. The campaign looks more efficient than it is, and the gap grows as more budget goes to warm audiences. The true effect only shows up in a lift test, which holds a matched group of shoppers back from the ads and compares what they buy with what the exposed group buys. The difference between the two groups is what the ads added. The same short window can also miss in the other direction: sales from an upper-funnel ad that arrive weeks later may fall outside it, so attributed numbers can short-change prospecting even while they over-credit retargeting. Either way, attributed sales can’t settle whether display works. A lift test can, read alongside Total Advertising Cost of Sale (TACoS), which weighs total ad spend against total Amazon sales, paid and organic.
Last Updated: October 26, 2026
- • What Click-Based Metrics Actually Measure on Amazon Display
- • Why Attributed Sales Run Higher Than What the Ad Caused
- • Where the Extra Credit Comes From
- • How Repeat Buyers and Existing Customers Distort Display Reporting
- • Why Treating Clicks as the Scoreboard Fails Under Scale
- • How to Judge Display Without the Inflated Numbers
-
• Frequently Asked Questions
- • What is the difference between a click-through and a view-through conversion on Amazon?
- • Why do Amazon display ad reports show high ROAS while total sales stay flat?
- • What are New-to-Brand metrics and why do they matter for DSP campaigns?
- • If not clicks or ACoS, what is a better way to measure Amazon Display ad success?
- • How does last-touch attribution assign credit for a sale?
- • What to Do Next
What Click-Based Metrics Actually Measure on Amazon Display
Most Amazon Display reports lead with two numbers: Click-Through Rate and Advertising Cost of Sale. Operators default to them because the Amazon Ads console serves them up automatically, with no math required.
That convenience is the trap. Both numbers describe things the report can see: a click, and a sale the report decided to credit. Neither one shows where the shopper was headed before the ad appeared. For the wider question of when display spend earns its keep, read the complete guide to when display and DSP ads pay off.
Why Attributed Sales Run Higher Than What the Ad Caused
The money numbers on a display report lean on attributed sales. ROAS divides those sales by ad spend. ACoS flips the same math. Click-through rate counts the clicks that earn the credit. And attributed sales answer only one question: did an ad touch come before the purchase? They never answer the one that matters: would the purchase have happened without it?
So, display ads, retargeting most of all, often reach shoppers who are already close to buying. When those shoppers click and buy, the report counts the sale as the ad’s work. The space between what the report credits and what the ad caused is the overstatement.
| Metric | What It Counts | What It Misses |
|---|---|---|
| Click-Through Rate | How often shoppers clicked the ad after seeing it | Whether the shoppers who clicked were already planning to buy |
| Advertising Cost of Sale | Ad spend divided by the sales credited to the campaign | How many of those credited sales would have happened with no ad |
| Total Advertising Cost of Sale | Total ad spend measured against total Amazon sales on the account, paid and organic | Which single campaign did the work, so it needs a lift test beside it |
Where the Extra Credit Comes From
Think of attributed sales as a receipt with the ad’s name stamped at the bottom. The stamp proves the ad was there when the sale happened. It doesn’t prove the ad made the sale.
So, who fills those receipts? Four kinds of shoppers show up, and three of them were already headed to checkout.
| Shopper Who Clicked the Display Ad | What the Credited Sale Really Shows |
|---|---|
| Retargeted shopper who viewed the product page earlier in the week | The ad gets the full sale, though the shopper was already weighing the product before the ad appeared |
| Repeat buyer restocking a product bought every month | The ad gets the full sale, though the reorder was already due |
| Shopper who searched the brand name, then clicked a display ad on the way to the listing | The ad gets the full sale, though the search showed the shopper already wanted the brand |
| New shopper with no history with the brand who first met the product in the ad | Here the credit is closest to real, but only a lift test can confirm it |
How Last-Touch Logic Hands Over the Whole Sale
A 2025 arXiv paper on Amazon’s attribution methods defines last-touch attribution as giving 100% of the credit to the touchpoint that most closely precedes the outcome. Picture a shopper who has had the product in her cart for days. She sees a display ad, clicks it, and buys. The click came last, so the ad gets the whole sale.
This isn’t only a textbook model. Amazon Ads said it would introduce a shopping-signal enhanced last-touch attribution model on January 1, 2026.
Every brand pays for this kind of borrowed credit, but smaller brands feel it most, and how fixed display costs hurt low-revenue brands walks through why.
Why Credit Is Not the Same as Cause
Reading attribution as causation is the shortcut that inflates display results. Attribution tells you who stood closest to the register, not who brought the customer into the store.
Fancier models don’t fix this on their own. The same arXiv paper cites a 2023 study by Gordon et al. of almost 2,000 representative campaigns at Meta, where sophisticated models using observational data produced errors in estimated ad effects ranging from 488% to 948%, depending on the model and outcome type. Those were Meta campaigns, not Amazon ones. The lesson still carries: a model that only watches what happened can’t tell which sales the ad actually caused.
How Repeat Buyers and Existing Customers Distort Display Reporting
Repeat buyers deserve a closer look, because they’re the easiest sale for a display campaign to claim. Giving them that closer look is part of how Amazon Ads Management reads display results. A shopper who reorders the same product every month will keep doing it. If a display ad sits in her path when the reorder comes due, the report books that purchase to the ad, month after month.
Retargeting stacks on top of this. A retargeting audience is made of people who already viewed or bought from the brand, so its credited sales lean on shoppers who needed little push. A campaign like that can post a strong ROAS while adding very few buyers. Separating repeat buyer bias from real campaign gains shows how to pull those buyers out of the numbers.
Why Treating Clicks as the Scoreboard Fails Under Scale
The overstatement gets more expensive as budget grows. A small test aimed at warm shoppers can look excellent. Scale it, and much of the new money keeps buying credit for the same kind of sale, so each extra dollar does less than the report suggests.
Why a Better ACoS on Flat Sales Isn’t Progress
Optimizing Advertising Cost of Sale on a stagnant revenue base does not build a stronger business. It tightens a number that was never the right target. A campaign can report an improving ratio while total sales stay flat, because the ratio only measures spend against attributed credit, not against the business’s actual trajectory.
How to Judge Display Without the Inflated Numbers
So, start with the split the report already offers. New-to-brand data flags whether a buyer had purchased from the brand within a set lookback period. A campaign whose credited sales are mostly repeat orders is mostly being paid for loyalty. A high new-to-brand share is a better sign, but it still doesn’t prove the ad caused the sale.
| Evaluation Step | Input Reviewed | Decision It Supports |
|---|---|---|
| Review new-to-brand share | New-to-brand share of credited sales, campaign by campaign | Whether the campaign mostly reaches people who already buy the brand |
| Run a lift test | Sales from the test’s two groups | How many sales the display ads actually added |
| Track TACoS over time | TACoS read week over week while display spend changes | Whether the whole account grows as display spend rises |
| Check account and catalog health | Listing integrity, account standing and unauthorized seller activity | Whether weak results come from the ads or from problems in the account itself |
Running a Lift Test Before Trusting the Report
A lift test, also called a holdout test, is the read that measures cause. Split a matched audience in two. One group sees the display ads. The other is held back. Run the test, then compare purchases. The gap between the groups is what the ads added. If the attributed report claims far more sales than that gap, the report is running high. Even Amazon’s own multi-touch attribution, as the arXiv paper describes it, uses randomized controlled trials alongside machine learning models to allocate credit across Amazon Ads touchpoints.
Where Attributed Numbers Can Run Low
Attributed numbers don’t only miss high. An Amazon Science study of 15 US brands across three verticals found that within the two-week attribution window, upper- and middle-funnel ad products show only 30 to 50 percent of their total effects, compared with 60 to 90 percent for lower-funnel products. So a prospecting campaign aimed at new shoppers can look weaker in the report than it is, because part of its effect lands after the window closes. That doesn’t rescue click-based metrics. It means the same report can flatter a campaign aimed at warm shoppers and shortchange one aimed at cold shoppers. A lift test catches both mistakes, as long as it runs long enough for the late buyers to count.
Frequently Asked Questions
Once the attribution problem clicks, a few mechanical questions keep coming up. These are the ones worth settling plainly.
What is the difference between a click-through and a view-through conversion on Amazon?
A click-through conversion happens when a shopper clicks an ad and buys afterward. A view-through conversion happens when a shopper only sees the ad, never clicks, and still buys within the attribution window.
Why do Amazon display ad reports show high ROAS while total sales stay flat?
Because ROAS is built on attributed sales, not added sales. If most of a campaign’s credited buyers were repeat customers, retargeted shoppers or people who searched the brand, the report books their purchases to the ad. ROAS climbs while total sales stay where they were.
What are New-to-Brand metrics and why do they matter for DSP campaigns?
New-to-brand metrics flag orders from shoppers who haven’t bought from the brand within a lookback period. For a DSP campaign, that shows how much credited revenue came from outside the existing customer base. It’s a sign of reach, not proof the ad caused the purchase, and a lift test supplies that proof.
If not clicks or ACoS, what is a better way to measure Amazon Display ad success?
Use two reads together. A lift test shows how many sales the display ads actually added. TACoS shows whether the whole account grew as display spend went up. Clicks and ACoS can stay on the report as activity checks, not as proof the ad worked.
How does last-touch attribution assign credit for a sale?
It hands the whole sale to the touchpoint that came last before the purchase. Order is the only test. The model never asks whether that last touch changed the shopper’s mind or simply happened to come last.
What to Do Next
Start with the campaigns you already run. Pull new-to-brand share for each display campaign, and mark the ones aimed mostly at past visitors or past buyers. Those are the campaigns most likely to be over-credited.
Next, pick one of those campaigns and plan a lift test before its next budget increase, not after. Keep the read open past the two-week mark, so slow buyers get counted.
Then chart TACoS week over week while the test runs, so the test result and the account trend get read side by side.
Display is one line in a channel P&L, so we read it next to listing health and account standing as part of Amazon Account Management, not as a stand-alone scoreboard.
To see how much of your own display credit is real, get a free Amazon audit.
Results vary by brand, category, and investment level.