Facebook ads for Amazon: how to run them and measure what they sell

Ads inside Amazon reach people who are already searching. That's valuable, but it mostly captures demand that already exists. Facebook and Instagram ads (both run through Meta Ads Manager) reach people who weren't looking for you yet. Done well, they create new demand.

The catch is that much of that demand ends up on Amazon in ways no report connects back to the ad. Someone sees your ad on Instagram, doesn't click, and searches your brand on Amazon two days later. The ad made the sale happen, but nothing records it. That's why Facebook ads for Amazon so often look worse in reports than they are, and why measuring them takes more than a dashboard.

There are three common setups. Each one changes what Meta can optimize for and what you can track.

• Straight to your Amazon listing or Brand Store. The simplest path for the shopper. Use an Amazon Attribution tag on the link so Amazon can report the clicks and sales. The downside: you can't put a Meta pixel on an Amazon page, so Meta never sees what happens after the click. It can optimize for clicks, not purchases.
• A landing page you own, then Amazon. The ad goes to your own page, which has the Meta pixel, and a "Buy on Amazon" button carries the Attribution tag. Meta can now optimize for an event you define, like clicks on that button. You also get a place to explain the product. The cost is one more step, and some people drop off at every step.
• Your own store. If you also sell on Shopify or your own site, Meta can optimize for real purchases there. But some shoppers who see the ad will still buy on Amazon, and those sales are invisible to Meta. The Facebook ads for Shopify guide covers that setup.

No setup is right for everyone. What matters is knowing which sales each one can see, and measuring the rest.

Amazon Attribution is Amazon's free tool for measuring ads outside Amazon. It's available to professional sellers enrolled in Brand Registry, vendors and agencies. You create a tag (a tracking link) for each campaign or ad, use it as the ad's link, and Amazon reports clicks, detail page views, add-to-carts, purchases and sales for the 14 days after each click.

Tag at the level you want to compare. If you want to know which ad set or creative sends the most buyers, each one needs its own tag.

The Brand Referral Bonus pays you for that traffic. Amazon says it averages 10% of qualifying sales from tagged links, applied as a credit against your referral fees, not cash. It's for brands in Amazon's US store, and the full credit can take about two months to show up. Amazon ads like Sponsored Products don't qualify; Facebook and Instagram ads do. The Brand Referral Bonus calculator shows what it does to your ad cost.

Any ad that links to Amazon without a tag earns no bonus and doesn't show up in Attribution at all.

Once tracking is set up, you'll have two reports, and both will be wrong in different directions:

• Meta Ads Manager counts purchases its pixel or Conversions API reports. It sees nothing on Amazon. If your ads link to Amazon, Meta's purchase numbers will be close to zero no matter how well the ads sell.
• Amazon Attribution counts Amazon purchases within 14 days of a click on a tag. It misses people who saw the ad and searched Amazon later, bought on another device, or bought after day 14.

Both also give credit for some sales that would have happened anyway, like loyal customers who clicked an ad on their way to reorder. Neither can tell you what the ads caused. That's incrementality, and it needs a test. The attribution gap calculator puts the reports and a test result side by side.

The direct way is a geo holdout test. Amazon's order reports include each order's ship-to state, so you can compare places where your Meta ads ran with places where they didn't:

1. Split the US states into two groups that have sold in step over the past few months.
2. Pause your Meta ads in one group for 4 to 12 weeks using location targeting. Change nothing else.
3. Compare Amazon orders in the two groups during the test with how they compared before it.

lift = (ads-on orders ÷ holdout orders, during) ÷ (ads-on orders ÷ holdout orders, before) − 1
caused Amazon sales ≈ what the holdout states would have sold with ads on − what they sold

Sales are noisy, so size the test first with the holdout test calculator. Brands with only a few orders a day usually can't see a realistic lift in a few weeks. The full method, including how the uncertainty range is worked out, is on how lift is calculated.

A made-up brand spends $4,000 a month on Meta ads, all linking to its Amazon listing with Attribution tags.

1. Meta Ads Manager reports almost no purchases, because it can't see Amazon.
2. Amazon Attribution credits $5,200 of sales: a ROAS of 1.3. On that number alone, the brand might cut the ads.
3. A geo holdout test finds the ads caused about $9,000 of Amazon sales: an incremental ROAS of 2.25. The extra $3,800 came from shoppers no tag tracked, called the halo effect.
4. On top of that, the Brand Referral Bonus at 10% of the tagged sales is a $520 credit on referral fees.

Whether 2.25 is good depends on your margin. Compare it with your break-even ROAS, not with the reported numbers. The Facebook ads ROI calculator shows how far your reported Attribution return is from break-even.

• Judging the ads on Meta's purchase numbers. If the ads send people to Amazon, Meta can't see the result.
• Judging them on Attribution alone. It only counts tagged clicks, so it often under-credits ads that build awareness.
• Untagged links. They earn no bonus and leave no trace in Attribution.
• Turning ads off and watching total sales. Sales move for many reasons: season, price, stock, competitors. Without a comparison group, you can't tell what the ads did.
• Testing too small. A test that can only detect a 40% lift will usually say "no clear effect", even when the ads work. Size it first.

Spillmark is built for exactly this question: how many Amazon sales do your Meta ads cause? It designs the geo holdout from your own Amazon order history, tells you up front if the test is too small, locks the design before it starts, and reports the result with its uncertainty range and the profit after Amazon fees and product costs. See how lift is calculated.