Campaign structure that makes bidding possible

Most advice about Amazon campaign structure is presented as taste. It is not. Structure has one job: make it possible for a bid to be correct.

Everything below follows from that.

Why averages hide winners

An ad group holds keywords and points them at products. Every keyword in it can be bid separately, but in practice most accounts set them broadly similar — and even where they do not, the ad group's data is what you have to reason from when any individual keyword is thin.

Put five keywords with different economics in one ad group and you get one blended picture. The strong one is subsidising the weak ones. Its true value is invisible, because the group's average conversion rate and average CPC describe none of its members.

This is the most common reason a hero term underperforms. It is not being bid badly on purpose; it is being bid as though it were its weakest neighbour.

The practical rule: if a term accounts for a meaningful share of a group's sales, it has earned its own ad group. Not because single-keyword ad groups are inherently virtuous, but because that is the only way its bid stops being an average.

Two layers, doing different jobs

A working account has a discovery layer and a conversion layer, and they should not be judged by the same standard.

Discovery — auto campaigns and broad match. Their job is to find queries you would never have thought of. Their ACOS will be worse, and that is correct: you are paying for information as well as sales. Judging a discovery campaign purely on efficiency and cutting it is how accounts slowly narrow until they are only bidding on what someone thought of two years ago.

Conversion — exact match and deliberate product targets. These own the terms that have already proven themselves. They should be efficient, because there is no uncertainty left to pay for.

The flow between them is the loop from the search terms guide: discovery finds, you harvest winners into conversion, you negate losers, discovery gets cleaner.

An account with no discovery layer stops finding new terms and gradually decays as the market shifts. An account that is all discovery never captures what it finds and pays average prices forever.

Match types as a funnel, not a preference

Match types are not three interchangeable options. They are stages.

Broad casts widest, matches loosely, converts worst, and finds the most. Phrase is narrower. Exact matches only that phrase and close variants, and converts best because there is no ambiguity about intent.

A term should move down this funnel over its life: discovered on broad or auto, promoted to exact once proven, blocked upstream so it does not compete with itself.

The mistake is running the same term on all three simultaneously with similar bids. You are then in three auctions for one query, and Amazon picks — usually not the one you would have chosen.

Grouping by economics, not by tidiness

Beyond ad groups, campaigns benefit from being grouped by something that shares a target.

The instinct is to group by product type because that is how your catalogue is organised. The better principle is to group by what they can afford — margin.

A 60%-margin product and a 25%-margin product should not share an ACOS target even if they sit in the same category, because their break-even points are completely different. One blended target means systematically over-investing in the thin product and under-investing in the fat one.

If your whole catalogue has similar margins, one account-wide target is genuinely fine. Do not manufacture complexity you do not need. The question is not "how many groups should I have" but "do my products have meaningfully different break-evens."

Signals that structure is the problem

Some symptoms point at structure rather than bids:

One term running in several campaigns at very different CPCs. You are bidding against yourself and paying a premium set by your own competition. The realized CPC spread across owners tells you how much.

A hero product with no dedicated campaign. Your biggest term is being priced by an average that includes four smaller ones.

A campaign capping its budget every day while under target. Structure is fine, budget is wrong — see the budgets guide.

One marketplace structured completely differently from another. Usually means one was built deliberately and the other grew by accident.

Ad groups with wildly mixed intent — a broad discovery term next to a specific converting one. Nothing can be bid correctly there.

Restructure carefully

Structural change resets your data. A new campaign has no history, so it has no norms, no conversion baseline, and nothing to bid from — you are back to step-up bidding until it accumulates evidence.

That cost is worth paying when structure is genuinely blocking correct bidding. It is not worth paying to satisfy a diagram.

Do it in pieces, largest problem first, and give each change several weeks before judging it. A wholesale rebuild leaves you unable to attribute any subsequent result to anything.

What AutoPPC does with this

Structural problems surface as findings with a quantified figure attached — how much spend is split across competing owners, what share of sales a term contributes to a shared ad group — rather than as generic advice.

Campaign groups are inferred from your campaign names and confirmed by name. Matching patterns exist underneath for edge cases but should never appear during ordinary setup.

Restructuring is never applied automatically. It arrives the way every other change does: as a proposal you approve row by row, with a reverse file that exists before anything ships.