Amazon Campaign Structure — The Complete Playbook

Part 1

Why structure is everything

Campaign structure is the foundation everything else sits on. Get it right and every optimization that follows — bidding, budgets, placement changes — becomes easier and more precise. Get it wrong and you spend months chasing performance problems that were structural problems all along.

This is a contested corner of Amazon advertising. People argue hard about single keyword campaigns versus ad groups, how many keywords belong in a campaign, whether to segment by match type, and whether to negate harvested terms from their source. Most of those opinions are incomplete — they describe one structure that worked for one account at one moment.

Part 2

Foundational principles

The granularity spectrum

Every structure decision is a trade between control and simplicity. Total aggregation — everything in one campaign — is easy to manage and impossible to optimize precisely. Total segmentation — every keyword in its own campaign — gives surgical control and drowns most teams in operations. The sweet spot for the large majority of accounts sits in between: single product ad groups, with granularity added only where data, budget, and your own bandwidth justify it.

The two jobs of Amazon's algorithm

Before any structure decision, remember what the algorithm optimizes for. It has exactly two objectives: give shoppers the most relevant product at the best price, and make Amazon as much money as possible — advertising is one of its largest revenue lines alongside AWS and retail. That second objective matters for structure debates later: Amazon has no incentive to limit your impressions because you run "too many keywords." Limiting your impressions would mean leaving money on the table.

Campaigns have settings. Ad groups have targeting. Products have performance.

This is the mental model that makes grouping decisions mechanical. The campaign level carries budget, placement modifiers, and dynamic bidding. The ad group level carries product-to-keyword relationships, individual bids, and ad group negatives. The product level carries conversion rate and revenue per click — the numbers that should be setting your bids in the first place.

So the grouping test is simple: if products inside a campaign share similar budgets, placement behavior, bid settings, and ACOS goals, grouping them costs you nothing. Where any of those diverge, they need their own campaign.

Part 3

Naming conventions that scale

A naming convention sounds like bureaucracy until the first time you need an answer fast. With consistent names, you can tell what a campaign promotes, how it targets, and what its goal is without opening it — and you can filter an entire account by product, tactic, or match type with one search instead of five merged spreadsheets.

Naming earns its keep in three places: reporting (slice performance by product or tactic instantly), organization (understand a campaign's purpose from the name alone), and management (troubleshoot without guessing which campaign does what).

Brand vs. non-brand: the sacred separation

If branded and non-branded targeting live in the same campaigns, your data is compromised. Branded keywords convert at rates non-brand traffic never touches and drag blended ACOS down, making non-brand performance look healthier than it is — so you keep funding campaigns that only look profitable because they harvest your own brand demand.

Part 4

The four campaign structures

There are exactly four architectures on Amazon. Each has a place; the skill is knowing when, not marrying one across a whole account.

1. Multiple product ad groups (MPAGs)

One ad group, many products, one set of keywords and bids. It is the default structure most accounts drift into — and the structure behind the classic catch-all failure: a single ad group stuffed with dozens of products, where one product runs at 15% ACOS next to another past 100%, and no bid change can fix the expensive one without starving the profitable one.

Why MPAGs fail: you cannot see which product converted on which keyword; a $15 product and a $150 product need fundamentally different bids but share one; and only one ad per ad group can display on any search — five products in one ad group can win at most one placement, while five products in five ad groups can win up to five.

The one real upside: aggregated campaigns can produce cheap clicks, because Amazon's pairing of products to queries is genuinely good. For a brand-new account, that simplicity has value — but treat it as a starting point, not a home.

2. Single product ad groups (SPAGs) — the baseline

One campaign per parent product, each child variation in its own ad group with its own keywords and bids. This is the structure to default to on virtually every account, because it is the single biggest step-change in control available:

3. Single product campaigns

Everything SPAGs offer, plus budget and placement control at the product level. Use when the catalogue is small enough to manage (roughly under 50 products), when specific products need their own spend caps, when placement performance differs sharply between products, or when high-spend hero products justify every lever having its own setting.

4. Single keyword campaigns

One keyword per campaign — maximum granularity, covered in depth in Part 9. The short version: a surgical ranking tool for a few VIP terms, not a default architecture for an account.

Part 5

Match types — the targeting dial

Auto campaigns

Auto targeting has four segments: close match, loose match, substitutes, and complements. Loose match and substitutes tend to be the most productive discovery engines; complements are hit-or-miss.

Manual match types

Should you segment by match type?

Yes — and it is one of the highest-leverage structural decisions available. Each match type behaves differently, discovers different terms, and wants a different bid strategy. Mixed into one campaign, they compromise each other: broad needs room and flexible bids to discover; exact wants precision bids on proven terms. Separated, you can budget research and performance independently and set placement modifiers per match type.

Part 6

The standard campaign stack

For most products, the complete setup is four to six campaigns — a funnel from discovery to defense:

Source campaigns vs. destination campaigns

Keyword harvesting — moving converting search terms from broad campaigns into precise ones — only works if the funnel is explicit. Auto and broad/phrase campaigns are sources: their job is discovery, and their budgets are research budgets. Exact and product-targeting campaigns are destinations: their job is performance on proven terms. Terms flow one way, from source to destination, when they meet criteria (Part 7).

All variations bid on the same keywords

If a parent has five variations, all five should be bidding on the main keywords — with bids that reflect each variation's own performance. Amazon personalizes results by shopper: the premium buyer sees your Pro version, the bargain hunter sees the Basic, the shopper who always buys black sees the black variant. Bid only your "best" variation and you are invisible to every other customer segment.

Part 7

Keyword harvesting done right

The number one mistake: over-harvesting

Across hundreds of accounts, the most common harvesting failure is aggression: grab every term with a single order, stuff it into manual campaigns. Four things go wrong:

Where research terms come from

Your search term reports are the most direct signal, but they only show terms your bids could actually win impressions on. Search Query Performance shows the terms that matter in your category whether you were competitive on them or not — which makes it the better pure research source. Vendor Central's Search Frequency Rank and reverse-ASIN tools round out the picture.

The debate: negate harvested terms from source campaigns?

The purist argument for negating from source is clean data and zero overlap. In practice, our default is don't negate from sources, for four reasons: auto and broad often earn cheaper CPCs on the same terms than exact does; overlapping placements multiply your presence on one search page; negating everything that converts from an auto campaign breaks its bid data — you end up optimizing it on non-converters only; and over time a starved source campaign stops discovering anything new. The exception is genuinely tight-budget scenarios where total spend control matters more than visibility.

Part 8

Negative keywords — cutting the fat

Over-negating damages more accounts than over-harvesting. Three criteria cause most of the damage:

Negate when a term is clearly wrong for the product — kids' shoes showing on "shoes for teenagers," silicone products showing on "steel" searches, the wrong gender, wrong material, wrong category. And aim negation at terms with volume: a one-click long-tail that will never be searched again doesn't need a negative, it needs to be ignored.

Negative exact vs. negative phrase

Negative exact blocks one specific term — use it when an irrelevant term shares words with relevant ones. Negative phrase blocks every search containing the phrase — use it when a whole theme is wrong ("steel" when you sell silicone). Apply at campaign level when the theme is wrong for everything in the campaign; at ad group level when it is wrong for one product but fine for its siblings — which is exactly the resolution single product ad groups give you.

Part 9

Single keyword campaigns and the impressions myth

The myth: fewer keywords = more impressions

The claim never dies: Amazon throttles impressions when a campaign holds too many keywords, so splitting campaigns "unlocks" reach. Tested repeatedly, the result is consistent: zero growth in impressions or clicks from splitting. Think about Amazon's incentive: a $100 budget across 25 keywords is $100 Amazon can charge for; the same $100 across 100 keywords is more chargeable opportunities, not fewer. Amazon is not leaving that money unspent.

The illusion has a mundane explanation: splitting a campaign usually also resets budgets — five new campaigns each get their own daily budget, so total potential spend rises. The budget changed, not the keyword count. Correlation, not causation.

Why keywords actually get zero impressions

When single keyword campaigns do make sense

SKCs are surgical ranking tools, not architecture. They earn their keep on a high-volume, high-importance term that needs its own budget cap, its own placement modifier, or precise daily spend control during an aggressive ranking push — and only when you have the tooling to manage hundreds of campaigns without losing discipline.

Part 10

The complete playbook

Play in the gray

If this guide has one lesson, it is that there is no single right answer. Single keyword campaigns are perfect for some accounts and a disaster for others. Over-harvesting destroys some accounts; under-harvesting starves the rest. The managers who win are the ones who can diagnose an account, read its constraints, and deploy the structure it needs — not the ones married to one system.

The principles that don't change

The weekly rhythm

Structure without cadence decays. A sustainable weekly loop: bids one day (RPC-based, 7–30 day windows), search term review the next (harvest and negation candidates), bulk-sheet changes the third (harvest with performance-based starting bids, negate where relevant), budgets and placements the fourth (feed capped winners, adjust top-of-search modifiers on data), and TACOS plus tactic-level review to close the week. Compress or spread it as capacity allows — what matters is that each activity happens on a cadence, not only when something looks broken.

Bulk and cut

Scaling follows seasons, like training. In the bulking phase you add keywords, launch campaigns, raise bids, expand targeting — higher ACOS is expected, because you are buying data and reach. In the cutting phase you optimize bids, negate waste, pause underperformers, tighten budgets — converting the investment into efficiency. Neither phase is a home. Accounts scale by oscillating between them: bulking forever bleeds money, cutting forever stagnates.

This is the exact structure discipline we apply to every account we take on — the same stack, the same harvesting rules, the same weekly loop. If you want it built on yours, the first diagnosis is free.