AMC Just Added Promotional Signals to Retail Purchases: How to Separate Discount-Driven Sales From Ad-Driven Sales
Until October 20, 2026, a purchase in Amazon Marketing Cloud looked the same whether the shopper paid full price or redeemed a 20 percent coupon on the way to checkout. That changed when Amazon added 13 promotional signal columns to the Amazon Retail Purchases dataset. Amazon's own release note says it plainly: until now, advertisers could not tell promotional purchases apart from full-price purchases in AMC.
If you run ads and promotions on the same ASINs, you have been reading two scorecards that cannot see each other. This post covers what launched, what it lets you measure for the first time, how coupons already distort the ROAS you report, and a routine for working out how much of a promoted sale belongs to the ad and how much to the discount.
Quick Answer: what changed and what to do with it
- What launched: Amazon's release note says the AMC Amazon Retail Purchases dataset now includes 13 promotional signal columns covering seller- and vendor-funded price discounts (Best Deals, Lightning Deals), point-of-sale discounts (Buy X Get Y), and coupons. The columns are available to Amazon Retail Purchases subscribers in the 24 countries where the dataset is available. The subscription is free on Amazon's side through December 31, 2026.
- What it provides: A purchase-level flag for whether an ad-attributed sale carried a promotion, and which type. You can now compare path to conversion, new-to-brand rate, and conversion rate for promoted and full-price purchases side by side.
- The caveat: The note does not name the columns, but the dataset schema does: three sets of four discount fields (price discount, point-of-sale discount, coupon) plus net_unit_price. What Amazon has not said is how far back the new fields are populated, so check the earliest dated row that carries values in your own instance.
- The risk: The ads console reports coupon and other in-cart discounts at the pre-discount price, so ROAS on coupon-heavy ASINs is overstated before you open AMC. Amazon has not documented this; the agency reporting is consistent on it.
- Our read: The single most useful number is the exposed conversion rate inside the promotion window compared with outside it. That delta tells you whether the ad or the discount is doing the work.
What Amazon shipped on October 20
What the release note confirms
The note describes 13 new columns in the Amazon Retail Purchases dataset. It names three promotion families: self-funded price discounts such as Best Deals and Lightning Deals, point-of-sale discounts such as Buy X Get Y, and coupons. It says advertisers can compare path to conversion and new-to-brand acquisition for promotional and full-price purchases, and see which promotion types and discount depths perform best. It also notes that the description of the existing unit_price column was updated to clarify how Lightning Deals are treated.
Access runs through AMC Paid Features. The dataset itself carries five years of Amazon store purchase history and, since June 2, 2026, Amazon's first-party paid features are free to subscribe through the end of 2026. You still have to subscribe on the Paid Features page; the columns do not appear in an instance that has not.
What the schema adds
The dataset schema fills in what the note leaves out. The 13 columns are three matching sets, price_discount, point_of_sale_discount, and coupon_discount, each with an amount, an ID, a name, and a type, plus one net_unit_price column. The definitions carry three details that matter for everything below.
- unit_price was already partly net. Its definition now reads that it includes price discounts except when the price discount type is a Lightning Deal. So a Best Deal has been inside unit_price all along; a Lightning Deal, a point-of-sale discount, and a coupon have not.
- net_unit_price closes that gap. It is the per-unit price after Lightning Deals, point-of-sale discounts, and coupons are applied. It is a price, not a margin: cost of goods, fees, and ad spend are not in it.
- price_discount_amt has its own baseline. It is measured against the ASIN's average selling price in the 90 days before the promotion started, not against list price. That is a better baseline than list, and a different one from what most teams use by hand.
Two things the schema still does not say: how far back the promotional fields are populated (the dataset holds multi-year history, but a new column can be backfilled for less), and the full list of values each type column can take. Pull both from your own instance before the steps below.
Both scorecards can be right, and that is the problem
Most Amazon teams run price and ads as two programs with two reviews. A coupon that ran during a campaign gets credit in the pricing review for the lift it drove, and the same purchases get credit in the ads review as attributed sales. Neither review is lying. A shopper can see a Sponsored Products placement, land on a detail page with a coupon badge, and buy. The ad brought the shopper, the coupon closed the sale, and both teams can point at the order.
The question was never which one gets the purchase. It is how much of the purchase each one earned. Before October 20, AMC could not help with that, because the purchase row carried no record of the discount. Now it can show you, per purchase, whether a promotion was present and what kind. That does not settle attribution on its own, but it lets you split your ad-exposed purchases into promoted and full-price groups and compare how each behaves. The behavior of the full-price group is the cleanest read you have on what the ad does alone.
This matters most on promoted ASINs, which are exactly the ones agencies and brands scale during deal events. Spend climbs because ROAS looks strong, and ROAS looks strong partly because price was cut. When the promotion ends, the bid targets that were set against the promoted result stay where they were.
Does your ROAS already account for the coupon? Mostly not
Before the new columns, the natural question is whether the ads console already handles this. The answer depends on the type of promotion, and Amazon has not published it. The clearest account comes from Pacvue's write-up on how retail promotions affect ad metrics: discounts that change the price shown on the page (Lightning Deals, Best Deals, Deals of the Day, price discounts) are reported at the discounted price, so attributed sales roughly match what the shopper paid. Discounts applied in the cart (coupons, Prime-exclusive and Prime-member promotions, Buy X Get Y) are reported at the site price at the time of conversion, so attributed sales carry the pre-coupon number.
Pacvue's example is a $10 ASIN with a $3 in-cart discount: the ads report shows $10 of sales against $7 of revenue. A coupon-heavy deal event therefore reports a better ROAS than the same event run as a site-price deal, with no difference in what customers paid.
Two limits on that claim. Pacvue's account is written from the vendor reporting side, and at least one seller-tool glossary says Seller Central ordered product sales are recorded net of coupons, so check your own payment reports before assuming which side you are on. And none of this changes the purchase count, which is why the conversion rate comparison in the next section tells you more than ROAS does.
Separating discount-driven from ad-driven purchases takes six steps
- Pull the schema and confirm the columns. Check that the three discount sets and net_unit_price are present in your instance, and find the earliest dated row where they carry values. That date is the start of your usable history.
- Flag every purchase as promoted or full price. A row with any non-null discount ID is promoted. Keep the three types as a second label, and split Lightning Deals out of the price-discount group, because unit_price already nets the other price discounts and the console reports in-cart discounts differently.
- Join ad exposure to the flagged purchases. Compare the exposed and unexposed groups inside promoted purchases and inside full-price purchases separately. Do not mix them. Mixed groups are how a coupon gets credited to an ad.
- Compare exposed conversion rates across promoted and full-price windows. This is the cut that tells you most. If exposed shoppers convert at a similar rate with and without a promotion, the ad is carrying the ASIN. If the rate collapses without the promotion, the discount is carrying it, and the ad is mostly delivering shoppers to a price.
- Restate ROAS on net_unit_price. Replace unit_price with net_unit_price on ad-exposed purchases before multiplying by units. The gap between the two figures is the part of your result that Lightning Deals, point-of-sale discounts, and coupons supplied.
- Re-run after the promotion ends and reset the targets. Break-even bids that assumed the promoted price should be recalculated against the price the ASIN now sells at.
The table below shows what each cut tells you and what it cannot.
Step six is the one that breaks in practice
Steps one through five are an analysis, and a good analyst finishes them in an afternoon. Step six is where it falls apart. The promotions calendar lives in one place, bids and budgets live in another, and nobody re-runs the AMC query after the coupon ends. The bid targets stay set to a break-even that assumed the discounted price.
This is the point where Qore fits. Qore, from Trellis, is where you turn a review you run by hand into a workflow that runs itself on a schedule. Here that means a recurring check that compares promoted ASINs against net_unit_price, flags the ones whose break-even has moved, and proposes the target change. Actions default to manual approval, so the proposal waits for you. Once you trust a workflow, you can let it execute. Q, the assistant inside Qore, helps you shape the rule, such as which ASINs, how large a gap counts, and what the proposed change should look like.
Qore is in open beta, and you can sign up here. It does not remove the need to know your own break-even. You set the standard, and it holds the schedule.
Promotional signals still describe correlation, not cause
Seeing a coupon on a purchase that also had ad exposure does not tell you the ad caused the purchase, and it does not tell you the coupon did. Nothing in this dataset substitutes for a holdout. If a promoted ASIN is a candidate for a large spend increase, run an exposure holdout before committing, and treat the reconciliation above as the way to pick which ASINs deserve the test.
The split also has a boundary. It explains why a promoted ASIN's ROAS moved during a coupon window. It does not explain two full-price windows that differ, or two promoted windows at the same discount depth that differ. Those gaps are real, and the promotion columns are not the alibi for them.
Three limits sit with Amazon. Attribution windows are Amazon's, and they differ for sellers and vendors. AMC access and minimum-audience thresholds are Amazon's. Query behavior on small ASIN sets can return too little data to read. Qore does not change any of them.
What to do this week
Subscribe to Amazon Retail Purchases if you have not, while it is free. Confirm the 13 columns are in your instance and find the earliest row they populate. Restate your last deal event on net_unit_price for the five ASINs that took the most spend. Compare exposed conversion rates inside and outside the promotion window. Then list the promoted ASINs whose bid targets still assume the discounted price, and reset them. If that list keeps coming back every deal event, it belongs on a schedule. For related reading, see what Amazon's free first-party AMC signals changed, how Trellis uses Claude and AMC to analyze and act, and the Amazon coupon performance report.
Frequently Asked Questions
Amazon's October 20, 2026 release note says the dataset now includes 13 promotional signal columns covering self-funded price discounts (Best Deals, Lightning Deals), point-of-sale discounts (Buy X Get Y), and coupons. The schema shows three sets of four fields (amount, ID, name, type) for each discount family, plus net_unit_price.
The dataset holds multi-year purchase history, but Amazon has not said how far back the promotional columns are populated. Find the earliest dated row where the discount fields carry values in your instance before planning around past events.
Mostly not. In AMC, unit_price includes price discounts except Lightning Deals, so revenue computed from it has never netted out Lightning Deals, point-of-sale discounts, or coupons; net_unit_price does. Amazon has not documented how the ads console treats a clipped coupon, so compare a coupon-heavy day's attributed sales against your payments report.
No. It is the per-unit price after Lightning Deals, point-of-sale discounts, and coupons are applied. It does not include cost of goods, fees, or ad spend. Other price discounts, such as Best Deals, are already reflected in unit_price.
No. The columns show which purchases carried a discount and which had ad exposure. That is correlation. An exposure holdout is the way to test causation.
No. They are available to Amazon Retail Purchases subscribers through AMC Paid Features, in the 24 countries where the dataset is available. The subscription is free on Amazon's side through December 31, 2026, but you still have to subscribe.
If sales are reported at a price the customer did not pay, discounted units are counted at the pre-discount price. In AMC, restating revenue on net_unit_price instead of unit_price shows how much of the result came from Lightning Deals, point-of-sale discounts, and coupons.
Run it after each promotion ends, and before each deal event for the ASINs you plan to scale. If the same ASINs keep needing a reset, schedule the check rather than relying on someone to remember it.
Yes. The columns are in the audience version of the Amazon Retail Purchases dataset, so you can build rule-based audiences such as coupon-only buyers or full-price buyers of promoted ASINs, and use them in Sponsored Display, as bid boosts in Sponsored Products and Sponsored Brands, or in Amazon DSP. Audiences from this dataset exclude UK and EU data.
