How to Tell If Display Ads Are Winning New Customers or Not

Last Updated: October 26, 2026

By Justin W. Boggs, Chief Revenue Officer

What Counts as a Real New Customer on Amazon Display

Before any measurement question gets answered, the vocabulary has to be nailed down. New-to-brand, incrementality, and attribution are not interchangeable words. They describe three different layers of the same transaction, and conflating them is where display reporting goes wrong.

Defining Incrementality Versus Attribution

Attribution answers a narrow question: did a shopper touch this ad before buying? Incrementality answers a harder one: would that shopper have bought anyway? Incrementality is the sales lift the ad itself caused, and nothing else.

So knowing when programmatic display advertising earns its keep on Amazon starts with keeping those two words apart.

What ‘New-to-Brand’ Actually Measures

New-to-brand status flags a shopper who hasn’t bought the brand within a set lookback window. That makes it a useful filter for reach: a display campaign whose buyers are almost all repeat customers isn’t finding new people.

What the label can’t show is cause. A first-time buyer may have found the brand through search and only passed the ad on the way.

Why Standard Amazon Display Reporting Overstates Its Own Impact

Standard display reporting tends to overstate display campaigns, especially when view-through conversions are counted. So it’s worth pulling the mechanics apart one at a time.

Why View-Through Credit Inflates Display Performance

A view-through conversion credits an impression a shopper may never have consciously registered. The shopper scrolls past a banner, forgets it within seconds, then buys later through a branded search. The display report still counts the display ad as the cause.

That crediting logic does not test whether the impression changed behavior. It only tests whether exposure and purchase happened in that order. The problem gets worse when repeat buyers skew display campaign results, because the exposed audience already leaned loyal.

Why ACoS-Only Reporting Hides the Underlying Problem

ACoS can’t catch this, because a view-through sale counts toward a campaign’s ACoS the same way a sale the ad truly caused does. Spotting this is part of what Amazon Ads Management does when display is in the mix. A campaign can look efficient on ACoS while mostly collecting credit for purchases that were coming anyway.

What the Attribution Numbers Actually Measure

The attribution report is not lying. It is answering a narrower question than most dashboards let on, and treating it as a verdict on causation is where the distortion starts.

How Amazon’s Multi-Touch Attribution Model Allocates Credit

A 2025 arXiv preprint describes Amazon’s multi-touch attribution as combining randomized controlled trials with machine learning models to allocate credit for Amazon conversions across Amazon Ads touchpoints, in proportion to their likely contribution to shopping decisions.

It’s a way to share credit across ads, not a test run on one brand’s own campaign. What counts as genuine incremental ad revenue goes deeper on where that line falls.

How Holdout Testing Isolates True Incremental Lift

A holdout test settles the question attribution cannot. It withholds display exposure from a comparable audience slice and compares what that slice buys against what the exposed group buys. The gap between the two groups is the lift the ad actually caused.

Building a Holdout Test Inside Amazon DSP

Inside Amazon DSP, a holdout group gets excluded from the campaign’s targeting while everything else about the audience stays matched. The excluded segment still gets tracked for purchases over the same window.

That tracked behavior becomes the baseline the exposed group is measured against. Setting it up is ongoing campaign work, not a one-time report pull.

Reading Results Without Fooling Yourself

Read the gap over a full purchase cycle, not a few days, and set it next to TACoS. A gap that never shows up in total revenue isn’t lift the business can bank. Keep both groups the same kind of shopper, and don’t change bids or creative partway through.

A change in either group’s conditions shows up as a gap the ad didn’t cause. Give the test enough time, too. An Amazon Science study found that within the two-week attribution window, upper- and middle-funnel ad products show only 30–50% of their total effects, so a display test read inside two weeks misses half or more of what the ads did.

Where Display Spend Quietly Cannibalizes Sales Already in Motion

Cannibalization is the quiet failure mode behind many inflated attribution numbers. A display ad doesn’t have to find a new customer to log a win. It just has to step in front of a shopper who was already walking toward the same purchase through organic search or a branded query.

The clearest warning sign is a display campaign whose credited buyers had mostly bought from the brand before. Those shoppers knew the brand before the ad reached them, and counting them as wins is how a campaign earns more budget for work it didn’t do. Branded searches are harder to read.

A study in MIS Quarterly found that mere exposure to display advertising increases people’s propensity to search for the brand and the product. So a buyer who arrived through a branded search may have been sent there by the ad, or may have been coming anyway. The attribution report can’t tell those two apart.

Using Amazon Marketing Cloud to Build a Custom Attribution View

A standard reporting dashboard can only take attribution logic so far. Amazon Marketing Cloud lets a brand build its own view, joining display exposure, search behavior and purchase data at the signal level. That means Amazon’s default crediting rules stop being the final word.

What Amazon Marketing Cloud Can and Can’t Prove

An AMC instance can compare exposed and unexposed shoppers, but that split isn’t randomized, so treat it as a lead to test with a holdout, not as proof. The hard part is reconciling that signal-level data against unit economics, not pulling another campaign report.

Frequently Asked Questions

Once incrementality enters the conversation, the same few questions keep coming up. Here are straight answers to the ones that matter most.

How do holdout tests prove if Amazon DSP ads are actually working?

By comparing two matched groups over a full purchase cycle: one sees the ads, one doesn’t. If the exposed group buys meaningfully more, the ads caused that difference. If it doesn’t, the ads added little on their own.

What’s the real difference between ACoS and TACoS for measuring display ad success?

ACoS grades one campaign against its own spend, in isolation. TACoS weighs total ad spend against total revenue across the business. So a campaign can post a strong ACoS while total revenue stays flat, and only TACoS shows that.

Can Amazon’s new-to-brand metrics be trusted for display ads?

Trust them for what they show: who bought, not why. A new-to-brand buyer can still have been ready to buy with or without the ad.

Why do view-through conversions make display ad ROI so hard to calculate on Amazon?

Because the display report credits the ad whenever a view came before the purchase, whatever actually drove the sale. So the ROI math runs on credited sales, not on sales the ad caused.

What is Amazon Marketing Cloud and how does it help with display ad attribution?

It’s an Amazon tool where a brand writes its own queries on ad and purchase data instead of reading fixed reports. That lets it trace the ads a shopper passed before buying and check that path against what the standard report credited.

How does Amazon’s multi-touch attribution model allocate credit across ad types?

Amazon’s multi-touch attribution splits credit for a sale across the Amazon Ads touchpoints a shopper encountered beforehand, weighting each by its likely contribution to the decision. That’s why a display ad can share credit with a search ad for the same purchase.

Where This Leaves You

Start with the two reads you already have: new-to-brand share for each display campaign, and TACoS charted week over week. Then plan a holdout test before the next display budget increase, not after it.

Running that test inside a live account is what Amazon DSP management is built around.

Results vary by brand, category, and investment level.