Prime Day 2026 ran four days, June 23 through June 26, and it worked exactly as designed. US online spend across all retailers reached $26.4 billion during the window, up 9.3% year over year, according to Adobe data reported by Retail Dive. That number narrowly beat Adobe’s own pre-event forecast of $26.3 billion, and it is now inching toward the roughly $32.45 billion consumers spent across Thanksgiving, Black Friday, and Cyber Monday combined last year. The headline is a win. Amazon moved the event up two weeks into the second quarter, Walmart and Target ran competing events, and the whole market lit up.
The headline is also the only number the earnings call will reward. And that is the problem I want to put in front of every Chief Digital Officer and CMO reading this: the retail system that produced $26.4 billion tells you the top line grew.. It cannot separate the demand it earned through relevance from the demand it rented through discount depth and financing. When you run a promotional engine that reports volume but not unit economics, you are running a subsidy program with a leaderboard, and the leaderboard is the only part anybody talks about.
The number everyone reads and the number nobody asks for
Look at what sat underneath the 9.3%. Buy-now-pay-later orders jumped 9.5% to $2.1 billion, accounting for 6.6% of all online orders during the event, per the Adobe data in Retail Dive. Discount depth held or crept up across categories: electronics peaked at 24% off list versus 23% last year, apparel held at 24% off, appliances at 16%, and toys rose from 19% to 20%. Telsey Advisory Group’s research team tracked 68 retailers and brands and found roughly 40% of them were more promotional around this event than they were in July of last year. So the market grew, and it grew while pushing more discount and more financing into the funnel at the same time.
Now hold that against the customer signal. Shopper satisfaction with deals fell to 59% reporting high satisfaction, down from 68% in 2025, according to the Prime Day statistics compiled by SmartScout. Nearly 45% of shoppers reported buying an item they had specifically been waiting to purchase until it went on sale, which means a large slice of that $26.4 billion was demand the retailer already had, timed to a discount the retailer chose to give away. Sixty-nine percent of items purchased were priced under $40. Day one alone did $8.3 billion, up 5.3% year over year. The top line is real. The question the top line does not answer is whether the incremental dollar carried any margin.
Growth that was bought, not growth that was earned
The most honest data point in the entire event came from Omnia Retail’s proprietary pricing analysis of Prime Day 2025, and it should be printed on the wall of every merchandising war room. Omnia found that 45.5% of products were priced higher during Prime Day than they had been the week before the event, and only 0.6% of products saw price reductions of more than 20% from that pre-event baseline. For almost half the catalog, the celebrated Prime Day discount simply reverted a price to roughly where it had been a month earlier. Two things can be true at once: Adobe measures peak discount against list price and reports double-digit markdowns, while Omnia measures against the actual street price the week before and finds the discount often evaporates. Both are accurate. The shopper feels the difference, which is why usage of price-tracking tools like CamelCamelCamel and Keepa spiked during the event. Consumer skepticism about promotional pricing is climbing, and the 59% satisfaction figure is what that skepticism looks like on a dashboard.
This is the core of the pattern. A promotional calendar that leans on discount depth and financing manufactures volume that is very hard to distinguish from earned demand. You see the orders. You do not see, in the same view, how many of those orders would have happened at full price, how many were pulled forward from a future full-margin purchase, and how many carried a BNPL fee, a deeper markdown, and a higher return probability that will land in a later quarter. The leaderboard says you won. The ledger has not been asked yet.
There is also a genuine relevance signal buried in this event, and it is worth naming because it points to the fix. Traffic to US retail sites from generative AI sources almost doubled year over year during Prime Day 2026, building on a 3,300% surge in 2025. More important than the volume is the quality: AI-referred shoppers spent 49.9% longer on sites, browsed 20.5% more pages, and had a 33% higher add-to-cart rate than non-AI traffic, and in 2026 they converted 50.7% better than non-AI sources, a full reversal from 2025 when AI traffic actually converted 23% worse. That is what earned demand looks like. Shoppers arrive with intent, find the product answers they need inside the listing, and convert without a subsidy attached. The channel is still small next to paid search, which drove 28.5% of online retail revenue in 2025, but it is the one channel where relevance, not discount, is doing the conversion work.
The marginal cost you cannot see
Here is where the operating gap becomes concrete. Take two brands that both treated Prime Day 2025 as a growth lever and pulled it in opposite directions, as documented in SmartScout’s marketplace data.
Apple listed 379 Prime Exclusive deals at an average discount of just 8%, ran only three Lightning Deals at 11% off, and held an average Buy Box price of $709.74 against a deal price averaging $656.42. Year over year, Apple expanded its deal count from 247 to 379 while holding discount depth flat or lower. It created more entry points for buyers and protected its pricing architecture. Apple treated Prime Day as a conversion event for demand it had already earned, and it knew, precisely, what each incremental order cost it.
Kitsch ran the opposite playbook: 662 deals at an average 20% discount, more than double Apple’s deal count at two-and-a-half times the depth. The result was a 90% month-over-month revenue jump from June to July 2025 and 80.1% annual growth, all while spending roughly 12% less on Amazon advertising than category competitors. Kitsch treated Prime Day as a discovery engine to pull shoppers into 70 categories, 29 of which now generate more than $1 million each. Its growth was deal-fueled.
Neither approach is wrong. What separates them is that each brand knew which growth constraint it was solving and instrumented the discount accordingly. Apple was converting existing demand and priced to protect margin. Kitsch was buying new demand and priced to buy reach. Both could answer the question that matters: what did the next order cost, and what is it worth over the customer’s life?
That is the question most enterprise retail promotional engines cannot answer in real time. When your merchandising, finance, and marketing systems each hold one piece of the marginal-cost picture, and none of them reconcile during the event, a Kitsch strategy looks like an accident. Whether you bought reach at a rational price or just eroded your own baseline remains a mystery.. The $26.4 billion market is full of both, and the reporting stack treats them identically because it counts orders, not economics.
The Cost of Doing Nothing
The Cost of Doing Nothing here is not the discount itself. Discounting is a legitimate instrument. The cost is the gap between the top-line lift you report and the unit economics nobody makes you disclose, compounded over every promotional event on the calendar.
Walk the mechanism. Every incremental order won on a reverted price, meaning a discount that only undoes a pre-event price hike, trains the customer to wait, and 45% of shoppers already told us they waited. Every order pushed through BNPL, 6.6% of orders and $2.1 billion this event, carries a financing cost and a higher return and default profile that the Prime Day headline never nets out. Every point of discount depth you add on top of the 40% of the market that is already more promotional is a point you are unlikely to claw back next event, because your competitor and the price-tracking apps both remember it. And the satisfaction slide from 68% to 59% is the leading indicator: you are spending margin to buy volume from a customer who trusts the deal less each cycle.
None of that shows up on the earnings call, because the earnings call asks about the leaderboard. The compounding damage is that a company optimizing a number it can see (gross volume) while ignoring a number it cannot (contribution margin per incremental order) will keep pouring discount and financing into the funnel until the quarter finally forces the reconciliation. By then the customer has been retrained and the baseline has moved down. That is the real bill for doing nothing about instrumentation.
The operating model that fixes it
The fix is an operating model where every promotion is decided against marginal contribution, not gross volume, and where the answer is available while the event is live. Four moves get you there.
First, instrument the marginal cost of the next order before the next event, not after it. That means one reconciled view that joins the discount depth, the BNPL fee, the ad cost of acquisition, and the expected return rate against the price actually charged, at the SKU and offer level. Apple can price 379 offers at 8% and protect $709.74 Buy Boxes because it can see this. Build the view that lets your merchants make the same call.
Second, price against your own durable baseline, and let the customer verify it. With CamelCamelCamel and Keepa usage spiking and Amazon itself surfacing 365 days of price history on product pages, the pre-event price hike followed by a headline discount is a visible tactic that costs you trust. A genuine discount off a durable baseline converts and retains. The 59% satisfaction number is recoverable, and recovering it is a margin decision.
Third, treat the AI channel as the relevance engine that reduces your dependence on subsidy. AI-referred traffic converts 50.7% better than non-AI and adds to cart 33% more often, because those shoppers arrive with intent and find structured answers in the listing. Make your product content answer the questions a shopping assistant asks, and you convert intent without paying for it in markdown. That is the highest-leverage substitution available to a retailer right now.
Fourth, put a unit-economics number on the earnings-call agenda alongside the top line. If leadership only asks about gross volume, the organization will only optimize gross volume. Ask for contribution margin per incremental promotional order, event over event, and the whole system reorients toward earned demand.
Prime Big Deal Days returns October 6 and 7, 2026, and the fourth-quarter promotional calendar behind it is the largest of the year. That is a hard deadline to close the gap between the number you report and the number that determines whether the growth was worth having. The retailers that walk into October able to price the next order will compound margin through the holidays. The ones still reading the leaderboard will grow their top line again, and quietly subsidize it, one more time.