Amazon Advertising Strategy: A Framework Built From Live Campaign Data

Strategy is a budget allocation problem, not a bidding trick

Most pages that rank for this query list tactics: negative keywords, match types, dayparting. Those are real but they're maintenance, not strategy. A strategy is the answer to one question: which product, on which ad type, gets which share of budget, and why.

Start by splitting your catalog into three groups based on margin and organic rank, not gut feel. Core SKUs — high margin, already converting — can absorb aggressive bids because they still turn a profit at higher ACoS. Developing SKUs — decent product, low velocity — get selective support with a time limit; if organic rank hasn't moved in a defined window, pull back. Defensive SKUs — thin margin or declining — get minimal spend, just enough to hold position, or nothing at all.

Once SKUs are tiered, map them to funnel stage. Sponsored Products does the conversion work near purchase intent. Sponsored Brands builds consideration by putting your name and range in front of someone who's already searching the category. Sponsored Display and DSP handle reach and retargeting — people who saw your product but didn't buy, or haven't seen it yet. Budget follows the tier and the stage together, not one or the other.

Bidding: what automatic, manual, and dynamic actually buy you

Automatic bidding is a data-gathering tool, not a set-and-forget setting. Run new keywords in auto or broad match first to see which search terms actually convert before you commit a fixed bid to them. Manual bidding is where you spend real money on proven terms — exact match, dedicated campaign, bid set by what the click is worth to that specific SKU. Dynamic bidding (up and down, or down only) adjusts in real time based on conversion likelihood; down-only is the safer default for stable, already-profitable campaigns, up-and-down suits high-intent placements where you're willing to pay more for a near-certain sale.

Here's what that discipline looks like at scale, not in theory. Across 30 advertisers we manage in July 2026, the portfolio delivered a 6.04x return on ad spend — measured across the whole book, not cherry-picked from the best line item. 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's not what a blended account, spanning search and display and video, actually pays. Quoting a single-channel number as if it's your whole account is one of the fastest ways to set a budget that's wrong from day one.

Blended cost per acquisition on that book was $5.49 across 57,137 attributed purchases, and 20.1% of those purchases came from a shopper new to the brand. That new-to-brand share matters for strategy because it tells you whether your ad spend is buying incremental customers or just harvesting demand that would have converted anyway.

The campaign stack, stage by stage

A full-funnel structure isn't four random ad types running at once — it's a sequence, and each stage has a different job and a different metric to judge it by. The table below is the structure; the comparison_table field lays it out in full.

• Awareness uses Sponsored Brands headline placements and DSP display or video to put the brand in front of category shoppers who haven't searched your name yet. Judge it on new-to-brand rate and reach, not ACoS — ACoS on a pure awareness placement will almost always look bad, because that's not its job.
• Consideration pairs Sponsored Brands with Sponsored Display retargeting to bring back people who viewed a detail page. Judge it on click-through and repeat detail page views.
• Conversion is Sponsored Products, manually bid, on exact-match branded and high-intent category terms. This is where ACoS is the right metric, because the shopper is close to buying.
• Retention and incrementality uses DSP audiences built from purchase and browse signals, tested against holdout groups. This is the only stage where you can honestly say whether the spend added a sale that wouldn't have happened anyway.

Brands selling considered-purchase products — premium cookware, DTC accessories, hobby kits — tend to need all four stages because the buying decision doesn't happen in one search. A brand selling a commodity with near-zero consideration time can often run the first three stages and skip DSP until volume justifies it.

When the numbers say the strategy isn't working

ACoS improving while total sales stay flat usually means one of two things: your organic rank is falling as ads take over the same real estate you used to win for free, or you're advertising SKUs that were already going to sell. Check TACoS — Total Advertising Cost of Sales, ad spend against total revenue including organic. If ACoS drops but TACoS doesn't, your ads are substituting for organic sales, not adding to them.

If a bid change doesn't move the needle after two full weeks, the problem usually isn't the bid — it's the keyword or the placement. Pull the search term report before you touch the bid again. And if a DSP campaign's reported ROAS looks great but overall brand revenue didn't move, check for double-counting: sponsored ads and DSP can both claim credit for the same purchase under last-click attribution. That's a measurement bug, not a performance win, and it's why we reconcile DSP and sponsored ads in Amazon Marketing Cloud rather than trusting either channel's own dashboard in isolation.

Mistakes that quietly wreck strategies — including ones we've made

Treating every SKU with the same target ACoS is the most common one. A hero product and a long-tail product don't have the same margin, so they shouldn't have the same bid ceiling. Set targets by SKU tier, not by category average.

Reporting a single-channel CPC or CTR as if it represents the whole account is another. It flatters the number and misleads the next budget conversation.

We've made the double-counting mistake ourselves, early on — reading DSP-reported conversions and sponsored-ads-reported conversions as additive without checking for overlap. Last-click attribution can't tell you if a sale was incremental; it can only tell you which ad happened to fire last. Holdout tests and matched controls can. That's the difference between a number that looks good and a number you can defend.

Where a managed DSP partner fits, and where it doesn't

Sponsored ads are self-serve by design and most brands should run them in-house or with a PPC-focused partner until spend and complexity outgrow that setup. DSP is a different animal — it requires a seat, audience-building infrastructure, and a way to measure incrementality that last-click attribution structurally cannot provide. reMKTR runs Amazon DSP as a managed service across 109 live advertiser seats, inside the Full Circle group, which has managed more than $500M in Amazon spend across 100+ brands including HexClad, Ridge, BK Beauty, Beardbrand, Epic Gardening, The Woobles, and Walkize. If your sponsored ads program has plateaued and you need to know whether display actually added a sale, that's the specific problem this kind of setup is built to answer — not a replacement for sponsored ads, a stage that comes after them.