The right Amazon PPC campaign structure is a funnel: auto and broad match campaigns discover search terms, exact match campaigns run the proven ones at precise bids, branded and non-branded targeting stay separated, and every search term that matters has exactly one owner. Structure is the foundation under every other optimization. Bids, budgets, negatives, harvests: all of them assume the data underneath is clean, and structure is what keeps it clean. In the accounts I audit, most "performance problems" trace back to structure: harvests that never happened, negatives that never got added, conversion data fragmented across campaigns until no single row justified a decision. This guide covers the discovery-to-performance funnel I build, the campaign stack per product, when to group products and when to isolate them, brand versus non-brand separation, and how the whole system prevents search term bleed.
Why campaign structure decides everything else
Campaign structure decides what you can see and what you can control. Campaigns hold the settings: budget, placement multipliers, bidding strategy. Ad groups hold the targeting: which keywords promote which product, at what bid. Products carry the performance: conversion rate, revenue per click, relevancy. Mix things that need different settings, and no bid change can separate them again.
That three-level model settles most structural arguments before they start. Anything that needs its own budget or placement strategy needs its own campaign, because those controls exist nowhere else. Any product that needs its own bids needs its own ad group. So before grouping anything, I run a four-part check: similar budgets, similar placement performance, compatible bidding settings, aligned target ACOS. One significant divergence, and the product earns its own campaign.
The proof that structure is usually the real diagnosis sits on my results page. A health and personal care account was doing everything right on the surface (campaigns built, bids adjusted, budgets allocated), and the problems were structural underneath: search term bleed splitting its conversion data, wasted spend on keywords that looked active, a bid architecture that rewarded impressions over conversions. The rebuild took ACOS from 39.66% to 27.02% in 90 days with no budget increase. The levers themselves are in my guide on lowering ACOS without killing sales; the point here is that every one of those levers needed a structure to pull against.
The discovery-to-performance funnel: discovery finds, exact match runs
Every campaign in a well-built account has one of two jobs. Discovery campaigns (auto, plus broad and phrase match) exist to find search terms and competitor ASINs you did not know to target. Performance campaigns (exact match, plus specific product targeting) exist to run proven terms at precisely controlled bids. The search term report is the conveyor belt between them.
Each match type sits at a different point on that dial. Auto targeting casts the widest net Amazon offers. Broad match shows on synonyms, misspellings, and reordered words. Phrase match keeps your phrase intact and collects the long-tail variations around it. Exact match runs only your keyword and its close variants, which is where conversion concentrates and where bids deserve precision. I keep match types in separate campaigns, because that is what lets research money and performance money live in different budgets: you can throttle discovery in a rough month without touching a single proven keyword.
Terms graduate on evidence, not enthusiasm. My harvest bar is two or more orders (one order can be coincidence), confirmation that the advertised product itself got the order rather than a halo sale to a sibling, and a relevancy check against the live search results. A harvested term enters the exact campaign at a bid of revenue per click times target ACOS, and the source campaign gets a negative exact so discovery stops re-buying traffic performance now owns. The report mechanics are in my search term report guide, and my n-gram analysis guide covers the word-level version of the same hunt.
Discovery also has a budget ceiling. On my benchmark table, auto campaigns should hold at or under 40% of spend at launch, tightening to 15% or less in the profitability phase, because a mature account should already know its converting terms and own them at exact match. An auto share stuck high is the signature of a broken harvest loop: you are paying Amazon, every month, to rediscover terms you already bought.
The campaign stack I build per product
Most products need four or five campaigns: an auto campaign for discovery, a manual research campaign holding broad and phrase match, an exact match performance campaign, a product targeting campaign for competitor and complementary ASINs, and a brand defense campaign once branded search exists. Each carries its own budget, its own job, and its own definition of success.
| Campaign | Job | Targeting | Funnel role |
|---|---|---|---|
| 1. Auto | Discover search terms and ASINs | Amazon's four auto segments | Discovery |
| 2. Manual research | Extend discovery into the long tail | Broad + phrase | Discovery |
| 3. Manual exact | Run proven terms at precise bids | Exact only | Performance |
| 4. Product targeting | Conquest competitors, cross-sell | Specific ASINs | Performance |
| 5. Brand defense | Protect your branded searches | Exact branded keywords | Defense, kept separate |
The build order matters less than the separation, but this is the sequence I follow:
- Launch the auto campaign. One per product, always on, for the life of the product.
- Add the manual research campaign. Broad and phrase versions of your seed keywords, on a research budget.
- Build the exact match performance campaign. Proven terms only, each priced at revenue per click times target ACOS.
- Add the product targeting campaign. Competitor ASINs where you compare well on price and reviews, plus complements.
- Separate brand defense. Branded keywords in their own exact campaign, and your brand terms negated in every non-brand campaign.
- Run the harvest-then-negate loop. Review the search term report on a cadence, promote what qualifies, negate at the source.
What the funnel costs on real numbers
Discovery has a price you can compute, and structure is what keeps it payable. On the dataset I grade across these guides ($36,303 in spend, $111,058 in ad sales, 4,140 orders from 45,672 clicks), the account converts at 9.1% with a $0.79 average CPC, which prices every stage of the funnel.
At 9.1%, an average order takes roughly eleven clicks, so eleven clicks is a fair trial for one search term: about $8.77, dividing spend by orders. My harvest bar of two orders costs roughly double that, around twenty-two clicks. And when a term graduates, its exact match bid is not a guess: revenue per click times target ACOS, which on this account's $2.43 revenue per click and a 30% target prices the click at $0.73. Every number in that chain assumes one thing: that the term's whole record sits in one place. Scatter those eleven clicks across three campaigns and no row ever finishes its trial, which is why fragmented structure quietly turns discovery spend into wasted spend.
Single product or grouped: how many products per campaign
One product per ad group is the baseline I recommend for nearly every account. Group products into a shared campaign only when they match on budget, placement behavior, bidding settings, and target ACOS. The structure I unwind most often is the opposite: multiple products sharing one ad group, one keyword list, and one set of bids.
Shared ad groups fail for mechanical reasons. You cannot tell which product earned which keyword's orders, so harvesting and negation go blind. Products with different prices need different bids on the same keyword, because a click is worth what that product's revenue per click says it is worth, and in a shared ad group they split one bid. And only one ad per ad group can show on a given search, so five products in one ad group compete with each other for a single placement; five products in five ad groups can win up to five.
Moving to single product ad groups is the biggest single jump in control an account can make. From there, granularity is earned, not defaulted: a product gets its own campaign when it needs its own budget cap or its own placement settings, and small catalogs can afford that for everything. Very large catalogs go the other way, aggregating the long tail into catch-all campaigns grouped by similar target ACOS so the structure stays manageable. Mixing is not a compromise, it is the point: isolate the top sellers with granular campaigns, aggregate the tail, and match the complexity to what you can actually manage.
| Structure | Best for | Control | Complexity |
|---|---|---|---|
| Multiple products per ad group | Very large catalogs, grouped by shared target ACOS | Low | Minimal |
| Single product ad groups | The baseline for most accounts | High | Moderate |
| Single product campaigns | Smaller catalogs, product-level budget caps | Very high | High |
| Single keyword campaigns | VIP ranking keywords only | Maximum | Very high |
Separate brand from non-brand, always
Branded and non-branded targeting never belong in the same campaign. Branded searches convert at a different level than non-branded, so any brand traffic leaking into a non-brand campaign inflates its numbers and hides real performance. Separate the campaigns, then add your brand terms as negative exact in every non-brand campaign you run.
The separation protects two things. The first is your data: a non-brand campaign quietly collecting branded searches through broad or phrase match will report a conversion rate and ACOS it did not earn, and you will scale a "winner" that is really your own customers finding you the long way around. The second is your budget: defense and recruitment are different jobs with different economics, and you cannot price either one if they share a wallet. My benchmarks guide holds branded sales to 5% or less of ad sales at launch, allows up to 25% during a market share defense, then brings it back to 15% or less; without separated campaigns you cannot even measure that number honestly. The enforcement mechanism is the ownership negative covered in my negative keywords guide, and the leak check is a regular pass through the search term report. The free Negative Keyword Finder builds the upload file from your own data.
A clean structure is what prevents search term bleed
Search term bleed is what structure failing looks like: multiple campaigns buying the same customer search term, splitting its conversion data so no campaign ever accumulates the statistics to price, harvest, or negate it. Prevention is structural. Every proven term gets one owner, the exact match campaign, and a negative exact in every other campaign that bought it.
The most reliable producer of bleed is the half-finished harvest: a term gets promoted into exact match, nobody adds the negative at the source, and from that day on the discovery campaigns pay to re-find traffic the account already owns. A funnel with the loop closed defends itself. Discovery finds, performance owns, and negatives keep the lanes separate, which is exactly the ownership fix my search term bleed guide walks in full. The free Audit Dashboard runs a dedicated bleed panel as one of its nine checks; the demo shows the read on a real anonymized account.
The nuance: your campaigns do not bid each other up
Amazon's auction is account-based: when several of your campaigns qualify for one search, only your highest bid enters against other advertisers. Overlap does not inflate your CPC, and deliberate overlap (one term live in exact and auto for two placements on the same page) is a legitimate play. The damage from unmanaged overlap is split data, not a bidding war.
So the case for ownership is statistical, not auction mechanics. Auto campaigns often earn cheaper clicks on the same search, and a second placement is real visibility; I have run intentional overlap and will again. But that is a per-term decision, made with eyes open and monitored. What I find in audits is not a decision. It is duplication nobody chose, with all of the data cost and none of the intent, and the default that prevents it is one owner per term.
The control structure actually buys: budgets, placements, bids
Campaign structure is budget and bid control wearing a filing system. Budgets, placement multipliers, and dynamic bidding strategy all live at the campaign level, so where you draw campaign lines determines exactly which products, tactics, and terms you can cap, boost, or reprice independently. A structure you cannot steer is just tidy.
With discovery and performance in separate campaigns, budget becomes a dial instead of a hope: open the budget on a performance campaign that hits target ACOS but runs dry by afternoon, and throttle research without touching a proven keyword. With products in their own campaigns, budget follows evidence. The skincare account on my results page is what that looks like: six ASIN variants with only three carrying their weight, spend concentrated onto the three converters, and wasted spend fell from 41% to 16%. Placement multipliers are campaign-level settings too, which is why you never need separate campaigns per placement; the multiplier gives you the control without the bloat. That lever carried a $310K per month home goods brand that was leaking roughly $11K a month to top-of-search placements on terms converting better lower on the page: repricing the placements alongside a full bid rebuild took ACOS from 32.1% to 21.6% in three months and added $28,400 a month to profit.
When a single keyword campaign earns its place
A single keyword campaign is a surgical ranking tool, not a default structure. It earns its place when one high-volume term needs its own budget cap, its own placement modifier, or precisely controlled daily spend during an aggressive ranking push. What it does not do is increase impressions. Splitting campaigns has never bought me impressions.
The myth that Amazon limits impressions when a campaign holds "too many" keywords refuses to die, and it fails both logic and testing. Amazon makes money on your clicks; more keywords in a campaign are more opportunities to charge you, not fewer. When splitting appears to lift impressions, the usual explanation is that every new campaign shipped with its own daily budget, so total potential spend rose. A keyword with no impressions almost always has a boring cause: the bid is uncompetitive, the budget exhausts early, the search volume does not exist, or Amazon does not consider the product relevant. Where the single keyword campaign genuinely shines is the budget cap play: bid high enough to win the premium placement, cap the daily budget so the total stays controlled, and hold order velocity on a ranking target without letting it eat the account.
Frequently asked questions
How should I structure my Amazon PPC campaigns?
As a discovery-to-performance funnel. Auto campaigns and broad or phrase match campaigns discover search terms; exact match and product targeting campaigns run the proven ones at precise bids. Keep one product per ad group, separate branded from non-branded targeting, and give every harvested term one exact match owner with negative exact everywhere else. Every other optimization assumes this foundation.
How many campaigns should each Amazon product have?
Four or five for most products: an auto campaign for discovery, a manual research campaign holding broad and phrase match, an exact match performance campaign, a product targeting campaign for competitor ASINs, and a brand defense campaign once shoppers search your brand. Each gets its own budget and its own job. Large catalogs can aggregate the long tail; the funnel logic stays the same.
Should I put multiple products in one Amazon campaign or ad group?
Only when they match on four things: similar budgets, similar placement performance, compatible bidding settings, and the same target ACOS. Even then, keep one product per ad group so each product's performance on every keyword stays visible and you can bid each relationship separately. Multiple products in one ad group share bids, and only one ad per ad group can show on a given search.
Should auto and manual campaigns run at the same time?
Yes, permanently. The auto campaign is not training wheels; it is the discovery engine that keeps feeding new search terms into your manual structure. What changes over time is its share of spend: at or under 40% of spend at launch, tightening to 15% or less in a mature account, because proven terms should keep graduating into exact match.
Do single keyword campaigns get more impressions?
No. The claim that Amazon limits impressions when a campaign holds too many keywords has never survived testing in my experience; splitting campaigns produced no impression growth. The effect people report usually comes from budget resets, because every new campaign gets its own daily budget. Use single keyword campaigns for control (budget caps and placement modifiers on VIP terms), not for reach.
Should branded and non-branded keywords be in separate campaigns?
Always. Branded searches convert at a different level, so brand traffic leaking into non-brand campaigns inflates the numbers and hides how well the account actually recruits new customers. Separate the campaigns, add your brand terms as negative exact in every non-brand campaign, and check the search term report regularly for leakage. Your data integrity depends on this split.
Check the structure you already have
The free Account Health Snapshot grades nine metrics from your bulk file, and several of them are pure structure: auto campaign share, branded versus non-branded split, budget concentration, spend and revenue alignment. About 60 seconds, parsed in your browser, no email, no account. If the grades say the structure needs work but you are not sure which of the four architectures fits your catalog and your capacity, that is a judgment call, and the free 30-minute diagnosis call is where I make it with you, campaign map on the table.