The Amazon Prime Day Advertising Strategy That Actually Works

The three phases, in order

Most guidance treats Prime Day as one event with one set of ads. It's actually three separate jobs, and they need different budgets, different ad types, and different success metrics.

• Pre-event (3-5 weeks out): Sponsored Brands and DSP display build awareness and retarget people who visited your listings or Brand Store but didn't buy. This is the phase most sellers skip because it doesn't show a same-day return, but the shoppers you reach here are the ones who come back during the event already familiar with your brand.
• Event days: Sponsored Products on your top sellers and deal-enrolled SKUs carries most of the volume. This is also where budget management matters more than bid management — a campaign that hits its daily cap at 11am on the biggest traffic day of the year stops delivering for the rest of that day, no matter how competitive the bid is.
• Post-event (2-8 weeks after): Retarget people who engaged but didn't convert, and look at what the event actually did to your customer base, not just what it did to that week's revenue.

The mistake we see most, including in our own early campaigns years ago, is spending the whole planning cycle on phase two and treating phases one and three as afterthoughts. Phase two is the easiest to plan for because it's the most visible. It's also the phase where differentiation is hardest, because every competitor is doing the same thing at the same time.

A worked example, on real numbers

Here's what a Prime Day-style push actually looks like at scale, using our own book rather than a hypothetical. Across 30 advertisers we manage, July 2026 delivered 6.04x return on ad spend measured across the whole portfolio — not the best line item pulled out to make a slide. That same book ran 78.4 million impressions at a $4.00 CPM and a blended $1.42 cost-per-click. If you've seen a $0.41 CPC quoted for this category elsewhere, that figure is online-video only; it is not what a full display and sponsored mix actually costs blended.

Blended cost per acquisition was $5.49 across 57,137 attributed purchases, and 20.1% of those purchases came from a shopper who had never bought the brand before. That last number is the one worth sitting with: one in five sales wasn't a repeat customer converting on a deal, it was a new customer the brand didn't have last month. If your Prime Day reporting only shows same-day ROAS, you won't see this number at all, because standard reporting doesn't split new-to-brand out by default.

The practical takeaway: don't judge a Prime Day campaign, or a vendor's pitch, on a single ROAS number without asking what it's measured across and whether new-to-brand is broken out separately.

Common mistakes, including ones we've made

• Raising bids the morning of, not the budget days before. A higher bid does nothing if the campaign is already flagged budget-limited. Check that flag on every campaign three to five days out, and raise the daily budget in steps, not the bid.
• Treating last-click as the whole truth. Last-click attribution will tell you a display campaign did almost nothing, because it credits the final click, usually a branded search, not the display impression that put the product in the shopper's head three days earlier. Last-click cannot prove incrementality on its own — it was never built to.
• Pulling DSP display the moment the event ends. The traffic surge doesn't stop the second Prime Day does. Cutting display on day five throws away cheap retargeting reach on shoppers who are still in-market.
• No negative targeting. Expensive, irrelevant keywords or competitor ASINs that never convert eat budget fastest exactly when budget is tightest — during the event itself.
• Judging success on day one. Attribution has a lag. A campaign that looks flat 24 hours in may look very different once the full purchase window closes.

When the numbers look bad mid-event

This is the part most guides skip. If your Prime Day numbers look wrong while the event is live, work through this in order, not all at once:

• Check the budget flag first, not the bid. A campaign marked budget-limited is invisible for the rest of the day regardless of how good the bid is. Fix the budget, then look at the bid.
• Check the time window before you panic. Purchases attributed to an impression or click can land 24-48 hours later. A campaign that looks like it's underperforming at hour six may be perfectly on pace once the window closes.
• Check whether display and sponsored ads are double-counting the same sale. If a shopper saw a display ad and later clicked a Sponsored Products ad for the same purchase, some reporting will credit both channels for the full sale, which inflates the apparent return of whichever number you look at last. Reconciling this requires looking at DSP and sponsored ads together, not as two separate reports.
• If the fix doesn't work, say so. Not every campaign should be rescued mid-event. If a targeting group is genuinely not converting after budget and timing are ruled out, the right move is to reallocate that spend to what's working, not to keep raising the bid on something broken.

What to measure once it's over

The event-day number is the easiest one to get and the least useful one on its own. Three things matter more:

• New-to-brand rate. What share of Prime Day purchases came from customers who'd never bought the brand before. This is the number that tells you whether the event grew your customer base or just discounted sales to people who were going to buy anyway.
• Longer-window value, not just same-day. Amazon's own long-term sales insights capture purchases and projected revenue from new customers over the following months, not just the event itself. A cookware brand or a beauty brand that picks up a new customer during Prime Day at a loss can still come out ahead if that customer buys again in month three.
• Whether display actually added anything. This is the honest, harder question, and it's why holdout tests and matched controls exist — comparing exposed and unexposed audiences directly, rather than trusting a last-click report that was never designed to prove incrementality.