A well-structured Sponsored Products account is one where every campaign controls a specific set of search terms, so you always know what you are paying for. Structure is the layer that decides whether your budget lands on the searches you want to win or leaks across thousands you never chose. Get it right and bid management, negative keywords, and reporting all become simple. Get it wrong and no amount of bid tweaking will save your ACoS.
Here is the situation that brings most sellers to us. You open the advertising console, you have somewhere between fourteen and ninety campaigns, the account ACoS is higher than you would like, and you cannot answer a simple question: which of these should I turn off tomorrow morning? The campaigns have names like SP – Auto – New and SP – Manual – Copy 2. Half of them contain the same product. Several of them are bidding against each other on the same search term. The spend is not obviously wasted, but it is not obviously working either, and the reports in front of you will not settle it.
That is a structure problem, not a bidding problem. Bidding decides what you pay. Structure decides whether you can ever know what you should have paid. In the accounts we take over, the bids have usually been tuned for months by somebody who could not read the structure underneath them, and the fix is nearly always to rebuild the shape before touching a single bid.
The goal of good structure is simple to state: maximum control over spend at the search-term level, with the least possible management overhead.
Three jobs, three kinds of campaign
A Sponsored Products account has to do three separate things, and the reason so many accounts are unreadable is that all three are happening inside the same campaigns.
Discovery is finding out what shoppers actually type before they buy your product. You do not know this at the start, and your keyword tool does not know it either — it knows what people search, not what converts on your listing. Automatic targeting campaigns and broad match campaigns exist to buy that information. They are a research budget. Judged as a sales channel they will usually look mediocre, and that is the correct result.
Harvest is taking the terms discovery proved out and giving them their own controlled bid. Exact match keyword campaigns and specific ASIN product-targeting campaigns do this. This is where the account should make its money, and it is the only place where a bid change has a clean, attributable effect.
Defence is your own brand terms and your own product pages. It is cheap traffic that would often have converted anyway, which means it flatters every efficiency metric you have. Keep it, by all means — but keep it in its own campaign so it can never be averaged into the numbers you use to judge anything else.
If a campaign is doing two of those three jobs, its ACoS is a blend of two different things and tells you nothing reliable about either.
What actually goes in each one
Sponsored Products gives you automatic targeting and manual targeting, and the choice is fixed when the campaign is created — you cannot convert one into the other later. Manual then splits into keyword targeting (broad, phrase and exact) and product targeting (individual ASINs, or categories with refinements such as brand, price band and review rating).
For discovery, run one automatic campaign per product or per tight product group. Auto campaigns separate their matching into close match, loose match, substitutes and complements, and you can bid on each group independently. That separation is more useful than it gets credit for: substitutes and complements are product-page traffic and behave nothing like the search traffic in close match, so bidding them the same is a decision you almost certainly did not mean to make. Alongside the auto campaign, run a broad match campaign holding your best guesses at the head terms. Amazon’s broad match can serve on synonyms and related phrases that do not contain your keyword at all, which is exactly why it is useful for discovery and exactly why you should not leave it running at a scale bid.
For harvest, one campaign per product family holding exact match keywords, and a second holding the ASINs you have proven you can win against. Whether each keyword gets its own ad group is a judgement call, not a rule. Single-keyword ad groups give you clean per-term budget and placement control, and they generate a great deal of admin. Below roughly a few thousand pounds of monthly ad spend, our own working view is that the admin usually costs more than the control is worth, and grouping five to fifteen closely related exact terms per ad group is the better trade. Above that, the control starts to pay.
Single-keyword vs multi-keyword campaigns
The first structural decision is how many keywords go into each campaign. Single-keyword campaigns (one keyword, tightly controlled) give you maximum precision. Multi-keyword campaigns (a themed group of related keywords) give you speed and simplicity. Most strong accounts use both, deliberately.
| Factor | Single-keyword campaign | Multi-keyword campaign |
|---|---|---|
| Bid control | Precise — one bid governs one term | Shared — the group moves together |
| Budget control | Isolated to that term | Pooled across the group |
| Best for | High-value head terms and proven converters | Long-tail, discovery, and lower-volume terms |
| Reporting clarity | Very high | Moderate |
| Management effort | High (more campaigns to maintain) | Low |
| Risk | Account bloat if overused | Winners and losers get averaged |
Bottom line: reserve single-keyword campaigns for the terms that deserve individual attention — your highest-volume, highest-intent, or best-converting keywords — and group the long tail into themed multi-keyword campaigns so you are not managing hundreds of near-identical campaigns for terms that each drive a handful of clicks. Our SKAG vs STAG comparison goes further into where that line sits.
Match types, and the job each one does
Match types are not three flavours of the same thing. Each has a distinct role, and structure works best when you let them play those roles instead of stacking all three on every keyword at full bid.
- Exact match is your control layer. It only serves for the search you specify, so it is where you place proven terms and your most confident bids.
- Phrase match is your middle ground — it captures close variations around a term while still keeping reasonable control.
- Broad match is your reach layer. It surfaces variations, synonyms, and adjacent searches you may not have thought of. It is the widest net and needs the most policing.
A common, sensible pattern is to bid most confidently on exact, slightly lower on phrase, and lowest on broad — because the wider the match type, the less certain you are that any given search is worth full price. Broad match earns its place as a discovery tool, not as a place to spend aggressively.
Where negatives belong, and why the level matters
This is the part that most often undoes an otherwise sound structure. Negative keywords in Sponsored Products can be applied at campaign level or at ad group level, in negative exact and negative phrase. Negative product targeting is available as well, for excluding specific ASINs and brands; the levels at which you can apply it have changed over time, so check what your console offers rather than assuming.
Two rules do most of the work.
- Terms you never want, anywhere, go on at campaign level. The wrong material, the wrong size, a competitor model number you do not compete with, the word “free”. Campaign level, so you add it once per campaign instead of once per ad group.
- Terms you are moving out of discovery and into harvest go on in the discovery campaign only. This is the harvest negative, and it is the one people get wrong. You have promoted a search term into an exact campaign; you now negate it in the auto or broad campaign that found it, so the two are not bidding on the same query. Negate it in the wrong campaign and you block the term from the very campaign you just built for it.
Know the difference between the two negative types before you use them at volume. Negative exact suppresses only that precise search term; negative phrase suppresses every search term containing that phrase. Negative phrase is the sharper instrument and the easier one to cut yourself with. “Case” as a negative phrase will also remove “hard case”, “case for” and, in some categories, a large part of your best traffic.
There is no account-level or portfolio-level negative list. Every negative lives inside a campaign, which means a term you have decided you never want has to be added everywhere it could appear. That is one more argument for having fewer, better-defined campaigns.
Bids, placements and budgets
Structure and bidding are linked. Once campaigns are cleanly separated, bids become meaningful decisions rather than guesses averaged across a mixed bag of keywords.
Start conservative and let data raise your bids. It is far cheaper to increase a bid on a term that is proving itself than to claw back overspend on one that isn’t. New campaigns have no history, so they earn budget rather than being handed it.
Sponsored Products offers fixed bids, dynamic bids down only, and dynamic bids up and down. The up-and-down setting can raise your bid above the figure you typed, with different ceilings depending on placement, and Amazon has changed those ceilings before — read the current values in the console rather than a number you saw in an article. Down only is the sane default for a discovery campaign, where you want the information without the volatility. Up and down belongs on proven exact terms, once you know what a conversion from that term is worth to you.
Placement adjustments — top of search, rest of search, product pages — multiply on top of all of that, and the multiplier available on top of search is very large. Treat placement as a lever you pull after a keyword is proven, never as a way to rescue a campaign that is not converting. If a term does not convert in rest of search, buying it at the top of search buys you the same non-conversion at a higher price.
On budgets: set them per campaign, keep discovery budgets deliberately small, and set harvest budgets high enough that the budget is never the constraint. Amazon can spend above a stated daily budget on a busy day and balance it out across the month, and the specifics of that behaviour have changed over time, so read a daily budget as an average rather than a hard ceiling.
The specific numbers depend on your margins and your break-even ACoS. The discipline — conservative start, data-led increases, placements set on evidence — is universal.
Launch vs scale: structure follows the goal
Structure is not static. The right shape for a brand-new product with no sales history is different from the right shape for an established one.
At launch, with no search-term data of your own, the job is discovery and momentum: lean on autos and broader match types to find what converts, with tight control on the few terms you are confident about. In the scale phase, you already know your winners, so the structure tilts toward controlled, single-keyword and exact campaigns on proven terms, with discovery running quietly in the background to keep the pipeline of new keywords full.
A structure that never changes is a sign the account has stopped being managed.
Naming, so the account survives you
Every campaign name should answer four questions without being opened: which product, which job, which match type, which marketplace. SP | ProductA | Harvest | Exact | UK does that. SP - Auto - New does not. This sounds like housekeeping and it is not. An account becomes unreadable because nobody can filter it, and nobody can filter it because the names carry no data. When you hand the account to someone else — a hire, an agency, yourself in eight months — the naming convention is the only documentation that goes with it.
A worked example
Take a seller doing £18,000 a month on one product family, spending £4,500 on Sponsored Products at a 34% ACoS, spread across nine campaigns with no clear division of labour. Restructured, that becomes four: one automatic campaign at £15 a day, one broad campaign at £15 a day, one exact harvest campaign holding the eleven terms the search-term report shows have actually converted, and one brand-defence campaign. Roughly £900 of the monthly spend now sits in discovery, where it is bought deliberately; about £600 sits in brand defence, quarantined out of every efficiency judgement; and the remaining £3,000 sits in harvest, where a bid change produces a readable result within a fortnight.
Those figures are an illustration, not a forecast. The point of the exercise is not that ACoS falls in month one — it often does not, and on a well-run account it sometimes rises while total sales rise faster. The point is that after the rebuild you can answer the question you could not answer before.
Five structural mistakes that quietly waste budget
These are the recurring issues that cost accounts the most, in rough order of how often they show up:
- No negative-keyword strategy. Broad and auto campaigns pull in irrelevant searches, and nothing blocks them. This is usually the largest single source of wasted spend.
- Everything in one or two campaigns. Winners and losers get averaged, visibility disappears, and bid decisions become impossible to make well.
- Underbidding the head term. The highest-volume, most important keyword ends up with a timid bid and never gets the impressions it deserves. It is worth checking the bid on your single most important term first, every time.
- Auto campaigns with no harvesting. Autos are left running as a permanent spend line instead of being mined for converting terms that should graduate into controlled campaigns.
- Duplicate or colliding campaigns. The same product and keywords get built twice, so two campaigns compete for the same search and split the data — inflating cost and muddying every report.
What to do on Monday morning
- Pull the last 60 days of the search-term report for one product family. Not the whole account — one family, so that you finish.
- Sort by spend. Mark every term with meaningful spend and zero orders. Those are your first campaign-level negatives.
- Mark every term with orders at an ACoS you would accept. That is your harvest list.
- Build one exact campaign for the harvest list, then negate each of those terms in the campaign that found them.
- Move brand terms into their own campaign, whatever they are currently doing to your averages.
- Rename what survives so the names answer the four questions. Then leave the bids alone for a fortnight and let the structure show you something.
Two cautions before you compare any numbers that come out of this. First, check which attribution window each report is using: Sponsored Products reporting offers more than one, the console default is not the same in every view, and a 7-day figure set beside a 14-day figure will make a campaign look better or worse than it is. Second, if the same product sits in two campaigns targeting the same query, only one of your ads can take that slot, and Amazon’s guidance has generally been that the higher bid is the eligible one. The usual cost of that overlap is not doubled spend — it is that your reports stop meaning anything, which is the more expensive problem and much the harder one to notice.
Frequently asked questions
What is the best campaign structure for Amazon Sponsored Products?
There is no single template, but the strongest accounts share a principle: each campaign controls a defined, related set of search terms so spend is visible and controllable. In practice that usually means single-keyword or exact campaigns for proven, high-value terms, themed multi-keyword campaigns for the long tail, and auto campaigns used purely for discovery — all held together by disciplined negative keywords.
How many keywords should I put in one campaign?
It depends on the keyword’s value, not a fixed rule. Put high-value, high-intent, or proven-converting terms into their own tightly controlled campaigns so you can bid on them individually. Group lower-volume, long-tail terms into themed campaigns so you are not drowning in near-identical campaigns that each drive only a few clicks.
Why is my Amazon ACoS so high even after lowering bids?
Lowering bids treats the symptom. A high ACoS is usually a structure and search-term problem: your campaigns are spending on searches you never intended to target, and those searches are averaged in with your good ones. The fix is to separate campaigns so you can see the waste, then use negative keywords to cut it and redirect traffic to the terms that convert.
Do I still need auto campaigns if I have good keyword research?
Yes — but as a discovery tool, not a workhorse. Auto campaigns surface real customer search terms that no keyword tool will predict. Run them to find new converters, promote those into your controlled campaigns, and negate the rest. Treat them as a feeder into your structure rather than a place to spend aggressively.
How often should I review campaign structure?
Search-term reports and negatives deserve a regular cadence — for active accounts, at least weekly in the early stages of a campaign. The overall structure should be revisited whenever a product moves between phases (for example, from launch to scale) or whenever reporting stops giving you clear, actionable numbers.